2/14/2011 --------- [bash-4.2 released] 2/15 ---- lib/glob/gmisc.c - fix wmatchlen and umatchlen to avoid going past the end of the string on an incomplete bracket expression that ends with a NUL. Partial fix for bug reported by Clark Wang 2/16 ---- subst.h - new string extract flag value: SX_WORD. Used when calling extract_dollar_brace_string to skip over the word in ${param op word} from parameter_brace_expand subst.c - change parameter_brace_expand to add SX_WORD to flags passed to extract_dollar_brace_string - change parameter_brace_expand to use SX_POSIXEXP for all non-posix word expansion operators that treat single quotes as special, not just % and # - change extract_dollar_brace_string to initialize dolbrace_state to DOLBRACE_WORD if SX_WORD flag supplied and we shouldn't use DOLBRACE_QUOTE. Fixes bug reported by Juergen Daubert doc/{bash.1,bashref.texi} - document the exact expansions here strings undergo 2/17 ---- lib/readline/vi_mode.c - make sure that `dd', `cc', and `yy' call vidomove_dispatch from rl_domove_read_callback. Fixes bug reported by Clark Wang lib/readline/callback.c - make sure _rl_internal_char_cleanup is called after the vi-motion callbacks (rl_vi_domove_callback) in rl_callback_read_char. Companion to above fix doc/{bash.1,bashref.texi} - make sure that the text describing the rhs of the == and =~ operators to [[ states that only the quoted portion of the pattern is matched as a string 2/18 ---- lib/glob/gmisc.c - better fix for umatchlen/wmatchlen: keep track of the number of characters in a bracket expression as the value to increase matchlen by if the bracket expression is not well-formed. Fixes bug reported by Clark Wang subst.c - change expand_string_for_rhs so that it sets the W_NOSPLIT2 flag in the word flags. We will not perform word splitting or quote removal on the result, so we do not want to add quoted nulls if we see "" or ''. Fixes bug reported by Mike Frysinger 2/19 ---- variables.c - new function, int chkexport(name), checks whether variable NAME is exported and remakes the export environment if necessary. Returns 1 if NAME is exported and 0 if not - call chkexport(name) to get tzset to look at the right variable in the environment when modifying TZ in sv_tz. Don't call tzset if chkexport doesn't indicate that the variable is exported variables.h - new extern declaration for chkexport {parse.y,builtins/printf.def} - call sv_tz before calling localtime() when formatting time strings in prompt strings or using printf. Fixes bug reported by Dennis Williamson execute_cmd.c - modify fix of 2/9 to add casts when those variables are passed to functions; some compilers throw errors instead of warnings. Report and fix from Joachim Schmitz support/shobj-conf - add a stanza for nsk on the Tandem from Joachim Schmitz {shell,lib/readline/shell}.c - Tandem systems should use getpwnam (getlogin()); for some reason they don't do well with using getuid(). Fix from Joachim Schmitz 3/1 --- variables.c - make sure that the return value from find_variable is non-null before trying to use it in chkexport. Fixes bug reported by Evangelos Foutras 3/3 --- parse.y - when adding $$ to the current token buffer in read_token_word(), don't xmalloc a buffer for two characters and then strcpy it, just copy the characters directly into the token buffer. Fix from Michael Whitten execute_cmd.c - fix expand_word_unsplit to add the W_NOSPLIT2 flag to the word to be expanded, so "" doesn't add CTLNUL. Similar to fix of 2/18 to expand_string_for_rhs. Fixes bug reported by Nathanael D. Noblet and Matthias Klose parse.y - fix extended_glob case of read_token_word to allocate an extra space in the buffer for the next character read after the extended glob specification if it's a CTLESC or CTLNUL. Report and fix from Michael Witten - fix shell expansions case of read_token_word to allocate an extra space in the buffer for the next character read after the shell expansion if it's a CTLESC or CTLNUL. Report and fix from Michael Witten - TENTATIVE: fix read_token_word to reduce the amount of buffer space required to hold the translated and double-quoted value of $"..." strings. Report and fix from Michael Witten - change code around got_character and got_escaped_character labels to make sure that we call RESIZE_MALLOCED_BUFFER before adding the CTLESC before a CTLESC or CTLNUL, and before adding the character if we're not adding a CTLESC. Report and fix from Michael Witten subst.c - new param flags value, PF_ASSIGNRHS, mirrors W_ASSIGNRHS, noting that parameter expansion is on rhs of assignment statement. That inhibits word splitting - change param_expand to call string_list_dollar_at with quoted == 1 if PF_ASSIGNRHS is set, so it will quote IFS characters in the positional parameter before separating them with the first char of $IFS. This keeps the rhs from being split inappropriately. Fixes bug reported by Andres Perera 3/4 --- lib/readline/bind.c - add a missing free of `names' in rl_function_dumper. Bug report and fix from Michael Snyder 3/5 --- lib/readline/rltty.c - change rl_deprep_terminal so it uses fileno (stdin) for the tty fd if rl_instream is not set, like rl_prep_terminal 3/6 --- lib/readline/display.c - fix rl_message to use a dynamically-allocated buffer instead of a fixed-size buffer of 128 chars for the `local message prompt'. Bug report and fix from Micah Cowan 3/7 --- jobs.c - add sentinel to wait_sigint_handler so it only sets wait_sigint_received if waiting_for_child is non-zero; otherwise, it restores the old SIGINT handler and sends itself the SIGINT - set waiting_for_child around the calls to waitchld that use it to synchronously wait for a process - change logic that decides whether or not the child process blocked or handled SIGINT based on whether or not waitpid returns -1/EINTR and the shell receives a SIGINT and the child does not exit. If the child later exits due to SIGINT, cancel the assumoption that it was handled - instead of testing whether or not the child exited due to SIGINT when deciding whether the shell should act on a SIGINT it received while waiting, test whether or not we think the child caught SIGINT. If it did, we let it go (unless the shell has it trapped); if it did not catch it, the shell acts on the SIGINT. Fix from Linus Torvalds , bug report originally from Oleg Nesterov 3/8 --- shell.c - initialize no_line_editing to 1 if READLINE is not defined -- we can't have line editing without readline 3/12 ---- lib/readline/signals.c - add SIGHUP to the set of signals readline handles lib/readline/doc/rltech.texi - document that SIGHUP is now part of the set of signals readline handles lib/readline/input.c - if _rl_caught_signal indicates that read() was interrupted by a SIGHUP or SIGTERM, return READERR or EOF as appropriate - call rl_event_hook, if it's set, if call to read in rl_getc returns -1/EINTR. If rl_event_hook doesn't do anything, this continues the loop as before. This handles the other fatal signals execute_cmd.c - add a couple of QUIT; calls to execute_disk_command and execute_simple_command to improve responsiveness to interrupts and fatal signals input.c - rearrange getc_with_restart so that the return values from read() are handled right parse.y - don't need to set terminate_immediately in yy_stream_get, since getc_with_restart checks for terminating signals itself - since readline returns READERR on SIGHUP or SIGTERM, don't need to set terminate_immediately. Still doesn't handle other signals well -- will have to check that some more bashline.c - new function, bash_event_hook, for rl_event_hook. Just checks for terminating signals and acts on them using CHECK_TERMSIG. - set rl_event_hook to bash_event_hook builtins/read.def - take out setting terminate_immediately; add calls to CHECK_TERMSIG after read calls doc/{bash.1,bashref.texi} - move the text describing the effect of negative subscripts used to reference indexed array elements to the paragraphs describing ${parameter[subscript]}, since that's where they are implemented. Pointed out by Christopher F. A. Johnson arrayfunc.[ch],subst.c - array_expand_index now takes a new first argument: a SHELL_VAR * of the array variable being subscripted. Can be used later to fully implement negative subscripts 3/14 ---- lib/glob/glob.c - fix mbskipname to not turn the directory entry name into a wide char string if the conversion of the pattern to a wide char string fails - fix mbskipname to call skipname if either the pattern or the filename can't be converted into a wide-char string lib/glob/xmbsrtowcs.c - fix xdupmbstowcs2 to handle return value of 0 from mbsnrtowcs and short-circuit with failure in that case. Fixes bug reported by Roman Rakus 3/15 ---- bashline.c - new variable, bash_filename_quote_characters to store the value assigned to rl_filename_quote_characters so it can be restored if changed. - change bashline_reset and attempt_shell_completion to restore rl_filename_quote_characters if not set to default 3/22 ---- lib/glob/glob.c - wdequote_pathname falls back to udequote_pathname if xdupmbstowcs fails to convert the pathname to a wide-character string lib/glob/xmbsrtowcs.c - xdupmbstowcs2: change to fix problem with leading '\\' (results in nms == 0, which causes it to short-circuit with failure right away). Fixes bug pointed out by Werner Fink - xdupmbstowcs2: compensate for mbsnrtowcs returning 0 by taking the next single-byte character and going on - xdupmbstowcs2: change memory allocation to increase by WSBUF_INC bytes; try to avoid calls to realloc (even if they don't actually result in more memory being allocated) 3/24 ---- doc/{bash.1,bashref.texi} - slightly modify BASH_SUBSHELL description based on complaint from Sam Liddicott 3/25 ---- trap.c - change free_trap_strings to not call free_trap_string for signals that are being ignored, like reset_or_restore_signal_handlers. Fixes bug reported by Satoshi Takahashi 3/26 ---- lib/readline/rltypedefs.h - remove old Function/VFunction/CPFunction/CPPFunction typedefs as suggested by Tom Tromey lib/readline/rlstdc.h - move defines for USE_VARARGS/PREFER_STDARG/PREFER_VARARGS from config.h.in to here because declaration of rl_message in readline.h uses the defines. This makes it hard for another packages to use after the header files are installed, since config.h is not one of the installed files. Suggested by Tom Tromey 3/27 ---- print_cmd.c - change indirection_string from a static buffer to a dynamic one managed by indirection_level_string(), so we don't end up truncating PS4. Suggested by Dennis Williamson lib/readline/shell.c - change sh_set_lines_and_columns to use static buffers instead of allocating the buffers to pass to setenv/putenv lib/readline/terminal.c - change _rl_get_screen_size to not call sh_set_lines_and_columns if ignore_env == 0 - _rl_sigwinch_resize_terminal: new function to just retrieve terminal size, ignoring environment lib/readline/rlprivate.h - new external declaration for _rl_sigwinch_resize_terminal() (currently unused) lib/readline/signals.c - rl_sigwinch_handler: set _rl_caught_signal to SIGWINCH - rl_sigwinch_handler: don't immediately call rl_resize_terminal; just leave _rl_caught_signal set for RL_CHECK_SIGNALS to handle - _rl_signal_handler: call rl_resize_terminal if sig == SIGWINCH. Should fix hang when sending multiple repeated SIGWINCH reported by Henning Bekel 3/29 ---- lib/sh/snprintf.c - include math.h for any defines for isinf/isnan - use code from gnulib documentation to implement isinf/isnan if they are not defined configure.in - don't check for isinf or isnan; c99 says they're macros anyway config.h.in - remove defines for ISINF_IN_LIBC and ISNAN_IN_LIBC, no longer used by snprintf.c 4/2 --- braces.c - brace_gobbler: fix to understand double-quoted command substitution, since the shell understands unquoted comsubs. Fixes bug reported by Michael Whitten lib/readline/display.c - include on MDOS - get and set screen size using DJGPP-specific calls on MSDOS - move cursor up clear screen using DJGPP-specific calls - don't call tputs on DJGPP; there is no good terminfo support lib/readline/terminal.c - include on MDOS - get and set screen size using DJGPP-specific calls on MSDOS - use DJGPP-specific initialization on MSDOS, zeroing all the _rl_term_* variables - don't call tputs on DJGPP; there is no good terminfo support DJGPP support from Eli Zaretskii 4/6 --- config-top.h - change DEFAULT_PATH_VALUE to something more useful and modern 4/8 --- tests/printf2.sub - make sure LC_ALL and LC_CTYPE are set so LANG assignment takes effect. Reported by Cedric Arbogast 4/11 ---- include/chartypes.h - fix a couple of dicey defines (though ones that don't cause any compiler warnings) in IN_CTYPE_DOMAIN doc/{bashref.texi,bash.1} - add note referring to duplicating file descriptors in sections describing redirecting stdout and stderr and appending to stdout and stderr. Suggested by Matthew Dinger pcomplete.c - it_init_helptopics: new function to support completing on help topics, not just builtins - it_helptopics: new programmable completion list of help topics - build list of helptopic completions in gen_action_completions on demand pcomplete.h - new extern declaration for it_helptopics builtins/complete.def - the `helptopic' action now maps to CA_HELPTOPIC intead of CA_BUILTIN, since there are more help topics than just builtins. Suggested by Clark Wang 4/12 ---- print_cmd.c - fix print_arith_for_command to add a call to PRINT_DEFERRED_HEREDOCS before ending the body of the command, so heredocs get attached to the right command instead of to the loop. From gentoo bug 363371 http://bugs.gentoo.org/show_bug.cgi?id=363371 execute_cmd.c - change coproc_pidchk to unset the appropriate shell variables when the (currently single) known coproc pid terminates - cleanup and new functions to fully support multiple coprocesses when and if I decide to go there 4/13 ---- print_cmd.c - fix print_group_command to add a call to PRINT_DEFERRED_HEREDOCS after call to make_command_string_internal before printing closing `}' - fix make_command_string_internal to add a call to PRINT_DEFERRED_HEREDOCS after recursive call to make_command_string_internal in case cm_subshell before printing closing `)' 4/14 ---- print_cmd.c - change overlapping strcpy in named_function_string to memmove sig.h - UNBLOCK_SIGNAL: convenience define, same as UNBLOCK_CHILD, just restores an old signal mask trap.c - set_signal: instead of setting the signal handler to SIG_IGN while installing the new trap handler, block the signal and unblock it after the new handler is installed. Fixes bug reported by Roman Rakus 4/15 ---- doc/{bash.1,bashref.texi} - make it clear that enabling monitor mode means that all jobs run in separate process groups 4/18 ---- builtins/fc.def - update fix of 4/15/2010 to not take saved_command_line_count into account when stepping down the history list to make sure that last_hist indexes something that is valid. Fixes bug reported by 4/19 ---- builtins/fc.def - fc_gethnum: make sure the calculation to decide the last history entry is exactly the same as fc_builtin. Fixes bug uncovered by fix of 4/18 to stop seg fault 4/22 ---- lib/readline/terminal.c - change _rl_enable_meta_key to set a flag indicating that it sent the enable-meta sequence - _rl_disable_meta_key: new function to turn off meta mode after we turned it on with _rl_enable_meta_key lib/readline/rlprivate.h - extern declaration for _rl_disable_meta_key configure.in - if not cross-compiling, set CFLAGS_FOR_BUILD from any CFLAGS inherited from the environment. Fixes HP/UX build problem reported by "Daniel Richard G." 4/26 ---- config-top.h - define MULTIPLE_COPROCS to 0 so the code is still disabled but easy to enable via configure option or editing this file 4/29 ---- lib/sh/eaccess.c - freebsd provides faccessat, with the same misfeature as their eaccess and access implementations (X_OK returns true for uid==0 regardless of the actual file permissions), so reorganize code to check the file permissions as with eaccess. Report and fix from Johan Hattne 5/2 --- doc/{bash.1,bashref.texi} - add forward reference to `Pattern Matching' from `Pathname Expansion', suggested by Greg Wooledge 5/5 --- pcomplib.c - the bash_completion project now distributes over 200 completions for various programs, with no end in sight, so increase the value of COMPLETE_HASH_BUCKETS from 32 to 128 pathexp.c - quote_string_for_globbing: make sure CTLESC quoting CTLESC is translated into \ even if the flags include QGLOB_REGEXP. We don't want to process the second CTLESC as a quote character. Fixes bug reported by Shawn Bohrer 5/6 --- builtins/printf.def - change PRETURN to not call fflush if ferror(stdout) is true - if a call to one of the stdio functions or printstr leaves ferror(stdout) true, and PRETURN is going to be called, let PRETURN print the error message rather than doubling up the messages. Fixes problem reported by Roman Rakus 5/9 --- doc/{bash.1,bashref.texi} - add note to the effect that lists inside compound command can be terminated by newlines as well as semicolons. Suggested by Roman Byshko 5/10 ---- subst.c - remove_quoted_nulls: fix problem that caused it to skip over the character after a CTLNUL, which had the effect of skipping every other of a series of CTLNULs. Fixes bug reported by Marten Wikstrom 5/11 ---- subst.c - extract_process_subst: add SX_COMMAND flag to call to extract_delimited_string, since we're expanding the same sort of command as command substitution. Fixes bug reported in Ubuntu bug 779848 5/12 ---- configure.in - set the prefer_shared and prefer_static variables appropriately depending on the value of $opt_static_link aclocal.m4 - AC_LIB_LINKFLAGS_BODY: change to not prefer shared versions of the libraries it's searching for if the prefer_shared variable is "no". Fixes problem reported by Cedric Arbogast 5/13 ---- lib/readline/readline.c - _rl_internal_teardown: add call to _rl_disable_meta_key to make the meta key active only for the duration of the call to readline() - _rl_internal_setup: move call to _rl_enable_meta_key here from readline_initialize_everything so the meta key is active only for the duration of the call to readline(). Suggestion from Miroslav Lichvar builtins/help.def - help_builtin: change strncmp to strcmp so that `help read' no longer matches `readonly'. Suggested by Clark Wang config.h.in - add define for GLIBC21, checked using jm_GLIBC21 as part of the tests for libintl lib/malloc/malloc.c - internal_free: don't use the cached value of memtop when deciding whether or not to adjust the break and give memory back to the kernel when using the GNU C library, since glibc uses sbrk for its own internal purposes. From Debian bug 614815, reported by Samuel Thibault aclocal.m4 - BASH_STRUCT_WEXITSTATUS_OFFSET: change AC_RUN_IFELSE to AC_TRY_RUN to avoid warning about not using AC_LANG_SOURCE 5/14 ---- bashline.[ch] - two new functions, bashline_set_event_hook and bashline_reset_event_hook, to set rl_event_hook to bash_event_hook and back to NULL, respectively - don't set rl_event_hook unconditionally sig.c - termsig_sighandler: if the shell is currently interactive and readline is active, call bashline_set_event_hook to cause termsig_handler to be called via bash_event_hook when the shell returns from the signal handler 5/15 ---- lib/readline/display.c - _rl_col_width: Mac OS X has a bug in wcwidth: it does not return 0 for UTF-8 combining characters. Added workaround dependent on MACOSX. Fixes problem pointed out by Thomas De Contes 5/16 ---- lib/readline/rlmbutil.h - WCWIDTH: wrapper for wcwidth that returns 0 for Unicode combining characters on systems where wcwidth is broken (e.g., Mac OS X). lib/readline/{complete,display,mbutil}.c - use WCWIDTH instead of wcwidth 5/17 ---- lib/readline/display.c - update_line: after computing ofd and nfd, see whether the next character in ofd is a zero-width combining character. If it is, back ofd and nfd up one, so the base characters no longer compare as equivalent. Fixes problem reported by Keith Winstein lib/readline/nls.c - _rl_utf8locale: new flag variable, set to non-zero if the current locale is UTF-8 - utf8locale(): new function, returns 1 if the passed lspec (or the current locale) indicates that the locale is UTF-8. Called from _rl_init_eightbit lib/readline/rlprivate.h - extern declaration for _rl_utf8locale locale.c - locale_utf8locale: new flag variable, set to non-zero if the current locale is UTF-8 (currently unused) - locale_isutf8(): new function, returns 1 if the passed lspec (or the current locale) indicates that the locale is UTF-8. Should be called whenever the locale or LC_CTYPE value is modified aclocal.m4 - BASH_WCWIDTH_BROKEN: new test for whether or not wcwidth returns zero-width characters like unicode combining characters as having display length 1; define WCWIDTH_BROKEN in this case config.h.in - WCWIDTH_BROKEN: new define lib/readline/rlmbutil.h - change WCWIDTH macro to use _rl_utf8locale and the full range of Unicode combining characters (U+0300-U+036F) 5/19 ---- lib/readline/rlprivate.h - _rl_search_context: new member, prevc, will hold character read prior to lastc lib/readline/isearch.c - _rl_isearch_dispatch: if the character causes us to index into another keymap, save that character in cxt->prevc - _rl_isearch_dispatch: if we index into another keymap, but don't find a function that's special to i-search, and the character that caused us to index into that keymap would have terminated the search, push back cxt->prevc and cxt->lastc to make it appear as if `prevc' terminated the search, and execute lastc as a command. We have to push prevc back so we index into the same keymap before we read lastc. Fixes bug report from Davor Cubranic 5/20 ---- expr.c - expr_bind_variable: pay attention to the return value from bind_variable and check whether or not we should error out due to a readonly or noassign variable. Fixes bug reported by Eric Blake 5/26 ---- lib/readline/search.c - include histlib.h for ANCHORED_SEARCH defines - rl_history_search_flags: new variable, holds ANCHORED_SEARCH flag for the duration of a history search - rl_history_search_reinit: takes a new flags variable, defines whether or not the search is anchored; assigned to rl_history_search_flags - rl_history_serarch_reinit: if ANCHORED_SEARCH flag passed, add ^ to beginning of search string; otherwise search string is unmodified - rl_history_search_internal: set rl_point appropriately based on whether or not rl_history_search_flags includes ANCHORED_SEARCH - rl_history_substr_search_forward: new function, for non-anchored substring search forward through history for string of characters preceding rl_point - rl_history_substr_search_backward: new function, for non-anchored substring search backward through history for string of characters preceding rl_point. Original code from Niraj Kulkarni lib/readline/readline.h - extern declarations for rl_history_substr_search_{for,back}ward lib/readline/funmap.c - history-substring-search-forward: new bindable command, invokes rl_history_substr_search_forward - history-substring-search-backward: new bindable command, invokes rl_history_substr_search_backward lib/readline/doc/{rluser.texi,readline.3} - document history-substring-search-forward and history-substring-search-backward 5/27 ---- {nojobs,jobs}.c - add support for DONT_REPORT_SIGTERM so that the shell doesn't print a message when a job exits due to SIGTERM since that's the default signal sent by the kill builtin. Suggested by Marc Herbert config-top.h - DONT_REPORT_SIGTERM: new user-modifiable setting. Commented out by default 5/28 ---- lib/readline/bind.c - _rl_skip_to_delim: skip to a closing double quote or other delimiter, allowing backslash to quote any character, including the delimiter - rl_parse_and_bind: call _rl_skip_to_delim instead of using inline code - rl_parse_and_bind: allow quoted strings as the values of string variables. Variable values without double quotes have trailing whitespace removed (which still allows embedded whitespace, for better or worse). Fixes problem with string variables not matching in `set' command if values happen to have trailing spaces or tabs (debian bash bug #602762), but introduces slight incompatibility. 5/29 ---- doc/{bash.1,bashref.texi} - clarify unset description to specify that without options, a variable, then a shell function if there is no variable by that name, is unset. Fixes discrepancy reported by Mu Qiao 6/4 ---- doc/{bash.1,bashref.texi} - clarify description of LINES and COLUMNS (and checkwinsize shopt option) to make it clear that only interactive shells set a handler for SIGWINCH and update LINES and COLUMNS. Original report submitted by Jonathan Nieder arrayfunc.c - expand_compound_array_assignment: defer expansion of words between parens when performing compound assignmnt to an associative array variable - assign_compound_array_list: perform the same expansions when doing a compound array assignment to an associative array variable as when doing a straight array index assignment. The idea is that foo=( [ind1]=bar [ind2]=quux) is the same as foo[ind1]=bar ; foo[ind2]=quux This fixes problems with double-expansion and quote removal being performed on the array indices 6/13 ---- doc/{bash.1,bashref.texi} - Add a little text to make it clear that the locale determines how range expressions in glob patterns are handled. 6/21 ---- builtins/read.def - display a message and return error status if -a is used with an existing associative array. Fixes bug reported by Curtis Doty 6/24 ---- {jobs,nojobs}.c - non-interactive shells now react to the setting of checkwinsize and set LINES and COLUMNS after a foreground job exits. From a suggestion by Leslie Rhorer doc/{bash.1,bashref.texi} - checkwinsize: remove language saying that only interactive shells check the window size after each command lib/readline/histfile.c - history_backupfile: new file, creates a backup history file name given a filename (appending `-') - history_do_write: when overwriting the history file, back it up before writing. Restore backup file on a write error. Suggested by chkno@chkno.net bashline.c - find_cmd_name: two new arguments, return the start and end of the actual text string used to find the command name, without taking whitespace into account - attempt_shell_completion: small changes to make sure that completion attempted at the beginning of a non-empty line does not find a programmable completion, even if the command name starts at point - attempt_shell_completion: small change to make sure that completion does not find a progcomp when in whitespace before the command name - attempt_shell_completion: small change to make sure that completion does not find a progcomp when point is at the first character of a command name, even when there is leading whitespace (similar to above). Fixes problems noted by Ville Skytta subst.c - brace_expand_word_list: since the individual strings in the strvec returned by brace_expand are already allocated, don't copy them to newly-allocated memory when building the WORD_LIST, just use them intact locale.c - locale_mb_cur_max: cache value of MB_CUR_MAX when we set or change the locale to avoid a function call every time we need to read it shell.h - new struct to save shell_input_line and associated variables: shell_input_line_state_t - add members of sh_parser_state_t to save and restore token and the size of the token buffer parse.y - {save,restore}_input_line_state: new functions to save and restore shell_input_line and associated variables - {save,restore}_parser_state: add code to save and restore the token and token buffer size - xparse_dolparen: call save_ and restore_input_line_state to avoid problems with overwriting shell_input_line when we recursively call the parser to parse a command substitution. Fixes bug reported by Rui Santos include/shmbutil.h - use locale_mb_cur_max instead of MB_CUR_MAX in ADVANCE_CHAR and similar macros lib/glob/smatch.c - rangecmp,rangecmp_wc: change to take an additional argument, which forces the use of strcoll/wscoll when non-zero. If it's 0, a new variable `glob_asciirange' controls whether or not we use strcoll/ wscoll. If glob_asciirange is non-zero, we use straight C-locale-like ordering. Suggested by Aharon Robbins 6/30 ---- execute_cmd.c - execute_pipeline: make sure the lastpipe code is protected by #ifdef JOB_CONTROL. Fixes problem reported by Thomas Cort 7/2 --- lib/readline/complete.c - EXPERIMENTAL: remove setting of _rl_interrupt_immediately around completion functions that touch the file system. Idea from Jan Kratochvil and the GDB development team lib/readline/signals.c - rl_signal_handler: if we're in callback mode, don't interrupt immediately on a SIGWINCH 7/3 --- bashline.c - set_directory_hook: and its siblings are a new set of functions to set, save, and restore the appropriate directory completion hook - change callers to use {set,save,restore}_directory_hook instead of manipulating rl_directory_rewrite_hook directly - dircomplete_expand: new variable, defaults to 0, if non-zero causes directory names to be word-expanded during word and filename completion - change {set,save,restore}_directory_hook to look at dircomplete_expand and change rl_directory_completion_hook or rl_directory_rewrite_hook appropriately bashline.h - extern declaration for set_directory_hook so shopt code can use it 7/6 --- builtins/shopt.def - globasciiranges: new settable shopt option, makes glob ranges act as if in the C locale (so b no longer comes between A and B). Suggested by Aharon Robbins 7/7 --- doc/{bash.1,bashref.texi} - document new `globasciiranges' shopt option 7/8 --- builtins/shopt.def - direxpand: new settable option, makes filename completion expand variables in directory names like bash-4.1 did. - shopt_set_complete_direxpand: new function, does the work for the above by calling set_directory_hook doc/{bash.1,bashref.texi} - document new `direxpand' shopt option 7/15 ---- lib/readline/isearch.c - _rl_isearch_dispatch: when adding character to search string, use cxt->lastc (which we use in the switch statement) instead of c, since lastc can be modified earlier in the function 7/18 ---- lib/readline/rlprivate.h - _rl_search_context: add another member to save previous value of (multibyte) lastc: pmb is to mb as prevc is to lastc lib/readline/isearch.c: - _rl_isearch_dispatch: if a key sequence indexes into a new keymap, but doesn't find any bound function (k[ind].function == 0) or is bound to self-insert (k[ind].function == rl_insert), back up and insert the previous character (the one that caused the index into a new keymap) and arrange things so the current character is the next one read, so both of them end up in the search string. Fixes bug reported by Clark Wang - _rl_isearch_dispatch: a couple of efficiency improvements when adding characters to the isearch string 7/24 ---- lib/readline/isearch.c - _rl_isearch_dispatch: save and restore cxt->mb and cxt->pmb appropriately when in a multibyte locale doc/{bash.1,bashref.texi} - correct description of {x}>file (and other redirection operators that allocate a file descriptor) to note the the fd range is greater than or equal to 10. Fixes problem reported by Christian Ullrich lib/readline/signals.c - rl_signal_handler: don't interrupt immediately if in callback mode lib/readline/callback.c - rl_callback_read_char: install signal handlers only when readline has control in callback mode, so readline's signal handlers aren't called when the application is active (e.g., between the calls to rl_callback_handler_install and rl_callback_read_char). If the readline signal handlers only set a flag, which the application doesn't know about, the signals will effectively be ignored until the next time the application calls into the readline callback interface. Fixes problem of calling unsafe functions from signal handlers when in callback mode reported by Jan Kratochvil execute_cmd.c - fix_assignment_words: when in Posix mode, the `command' builtin doesn't change whether or not the command name it protects is an assignment builtin. One or more instances of `command' preceding `export', for instance, doesn't make `export' treat its assignment statement arguments differently. Posix interpretation #351 doc/{bash.1,bashref.texi} - document new Posix-mode behavior of `command' when preceding builtins that take assignment statements as arguments builtins/printf.def - printstr: if fieldwidth or precision are < 0 or > INT_MAX when supplied explicitly (since we take care of the `-' separately), clamp at INT_MAX like when using getint(). Fixes issue reported by Ralph Coredroy 7/25 ---- lib/readline/chardefs.h - isxdigit: don't define if compiling with c++; declared as a c++ template function. Fixes bug reported by Miroslav Lichvar builtins/printf.def - getint: if garglist == 0, return whatever getintmax returns (0). Fixes bug reported by Ralph Coredroy 7/28 ---- doc/{bash.1,bashref.texi} - minor changes to the descriptions of the cd and pushd builtins lib/sh/zread.c - zsyncfd: change variable holding return value from lseek to off_t. Bug report and fix from Gregory Margo 8/1 --- expr.c - don't check for division by 0 when in a context where no evaluation is taking place. Fixes bug reported by dnade.ext@orange-ftgroup.com 8/6 --- execute_cmd.c - execute_command_internal: the parent branch of the subshell code (where the child calls execute_in_subshell) should not close all open FIFOs with unlink_fifo_list if it's part of a shell function that's still executing. Fixes bug reported by Maarten Billemont 8/9 --- builtins/common.c - get_exitstat: return EX_BADUSAGE (2) on a non-numeric argument builtins/return.def - return_builtin: just call get_exitstat to get the return status, let it handle proper parsing and handling of arguments. Fixes issue most recently raised by Linda Walsh . Reverses change from 9/11/2008 (see above) 8/16 ---- doc/{bash.1,bashref.texi} - clean up `set -e' language to make it clearer that any failure of a compound command will cause the shell to exit, not just subshells and brace commands 8/17 ---- configure.in - make the various XXX_FOR_BUILD variables `precious' to autoconf to avoid stale data - change how CC_FOR_BUILD is initialized when cross-compiling and not, but do not change behavior - initialize CFLAGS_FOR_BUILD to -g when cross-compiling - initialize LIBS_FOR_BUILD to $(LIBS) when not cross-compiling, empty when cross-compiling - create AUTO_CFLAGS variable to hold basic CFLAGS defaults; used when CFLAGS not inherited from environment (like effect of old auto_cflags variable) - substitute LIBS_FOR_BUILD into output Makefiles [changes inspired by bug report from Nathan Phillip Brink -- gentoo bug 378941] builtins/Makefile.in - substitute LIBS_FOR_BUILD from configure, not strictly initialized to $(LIBS) 8/27 ---- doc/{bash.1,bashref.texi} - minor changes to the here string description to clarify the expansions performed on the word support/shobj-conf - handle compilation on Lion (Mac OS X 10.7/darwin11) with changes to darwin stanzas. Fixes readline bug reported by Vincent Sheffer lib/sh/strtrans.c - ansic_wshouldquote: check a string with multi-byte characters for characters that needs to be backslash-octal escaped for $'...' - ansic_shouldquote: if is_basic fails for one character, let ansic_wshouldquote examine the rest of the string and return what it returns. From a patch sent by Roman Rakus 8/30 ---- lib/sh/strtrans.c - ansic_quote: changes to quote (or not) multibyte characters. New code converts them to wide characters and uses iswprint to check valid wide chars. From a patch sent by Roman Rakus 9/7 --- lib/sh/shquote.c - sh_backslash_quote: change to be table-driven so we can use a different table if we want to - sh_backslash_quote: takes a second char table[256] argument; externs.h - sh_backslash_quote: add second argument to function prototype bashline.c,braces.c,parse.y,builtins/printf.def - change callers of sh_backslash_quote to add second argument bashline.c - filename_bstab: table of characters to pass to sh_backslash_quote; characters with value 1 will be backslash-quoted - set_filename_bstab: turn on characters in filename backslash-quote table according to passed string argument - call set_filename_bstab every time rl_filename_quote_characters is assigned a value - bash_quote_filename: call sh_backslash_quote with filename_bstab as second argument. This allows other characters in filenames to be quoted without quoting, for instance, a dollar sign in a shell variable reference 9/8 --- bashline.c - complete_fullquote: new variable, controls table passed to sh_backslash_quote. If non-zero (the default), the standard set of shell metacharacters -- as in bash versions up to and including bash-4.2 -- gets backslash-quoted by the completion code. If zero, sh_backslash_quote gets the table with the characters in the variable reference removed, which means they are removed from the set of characters to be quoted in filenames 9/10 ---- bashline.c - bash_filename_stat_hook: new function, designed to expand variable references in filenames before readline passes them to stat(2) to determine whether or not they are a directory 9/15 ---- builtins/declare.def - if assign_array_element fails due to a bad (or empty) subscript, mark it as an assignment error and don't attempt any further processing of that declaration. Fixes segfault bug reported by Diego Augusto Molina 9/19 ---- expr.c - exppower: replace the simple exponentiation algorithm with an implementation of exponentiation by squaring. Inspired by report from Nicolas ARGYROU bashline.c - bash_quote_filename: check for rtext being non-null before dereferencing it - set_saved_history: operate_and_get_next assumes that the previous line was added to the history, even when the history is stifled and at the max number of entries. If it wasn't, make sure the history number is incremented properly. Partial fix for bug reported by gregrwm doc/{bash.1,bashref.texi},lib/readline/doc/{hsuser,rluser}.texi - minor editorial changes inspired by suggestions from Roger Zauner 9/20 ---- lib/intl/localealias.c - read_alias_file: close resource leak (fp) when returning on error 9/22 ---- execute_command.c - execute_intern_function: implement Posix interpretation 383 by making it an error to define a function with the same name as a special builtin when in Posix mode. http://austingroupbugs.net/view.php?id=383#c692 9/25 ---- doc/{bash.1,bashref.texi} - formatting and some content changes from Benno Schulenberg - document new posix-mode behavior from interp 383 change of 9/22 9/30 ---- execute_cmd.c - shell_execve: add strerror to error message about executable file that shell can't execute as a shell script. From suggestion by daysleeper 10/1 ---- bashhist.c - maybe_add_history: act as if literal_history is set when parser_state includes PST_HEREDOC, so we save the bodies of here-documents just as they were entered. Fixes bug reported by Jonathan Wakely - bash_add_history: make sure that the second and subsequent lines of a here document don't have extra newlines or other delimiting chars added, since they have the trailing newline preserved, when `lithist' is set and history_delimiting_chars isn't called execute_cmd.c - execute_command_internal: avoid fd exhaustion caused by using process substitution in loops inside shell functions by using copy_fifo_list and close_new_fifos (). Fixes debian bash bug 642504 lib/readline/complete.c - new variable, rl_filename_stat_hook, used by append_to_match. If filename completion is desired, and rl_filename_stat_hook points to a function, call that function to expand the filename in an application-specific way before calling stat. bashline.c - bash_default_completion: if variable completion returns a single match, use bash_filename_stat_hook and file_isdir to determine whether or not the variable name expands to a directory. If it does, set the filename_append_character to `/'. This is not perfect, so we will see how it works out. Adds functionality requested by Peter Toft and Patrick Pfeifer - rl_filename_stat_hook: assigned bash_filename_stat_hook, so things like $HOME/Downloads (after completion) have a slash appended. In general, this causes the stat hook to be called whenever filename completion is appended. Adds functionality requested by Patrick Pfeifer lib/readline/readline.h - new extern declaration for rl_filename_stat_hook lib/readline/doc/rltech.texi - rl_directory_rewrite_hook: now documented - rl_filename_stat_hook: document pcomplete.c - gen_action_completions: in the CA_DIRECTORY case, turn off rl_filename_completion_desired if it was off before we called rl_filename_completion_function and we didn't get any matches. Having it on causes readline to quote the matches as if they were filenames. Adds functionality requested by many, including Clark Wang assoc.[ch] - assoc_replace: new function, takes the same arguments as assoc_insert, but returns the old data instead of freeing it - assoc_insert: if the object returned by hash_insert doesn't have the same value for its key as the key passed as an argument, we are overwriting an existing value. In this case, we can free the key. Fixes bug reported by David Parks 10/5 ---- print_cmd.c - indirection_level_string: small change to only re-enable `x' option after calling decode_prompt_string if it was on before. In normal mode, it will be, but John Reiser has a novel use for that code in conjunction with a pre-loaded shared library that traces system call usage in shell scripts 10/10 ----- Makefile.in - Fix from Mike Frysinger to avoid trying to build y.tab.c and y.tab.h with two separate runs of yacc if parse.y changes. Problem with parallel makes - Fix from Mike Frysinger to avoid subdirectory builds each trying to make version.h (and all its dependencies) lib/sh/Makefile.in - remove some dependencies on version.h where it doesn't make sense variables.c - initialize_shell_variables: while reading the environment, a shell running in posix mode now checks for SHELLOPTS being readonly (it gets set early on in main()) before trying to assign to it. It saves an error message and the variable gets parsed as it should. Fixes bug reported by Len Giambrone 10/14 ----- doc/{bash.1,bashref.texi} - add to the "duplicating file descriptors" description that >&word doesn't redirect stdout and stderr if word expands to `-' - add to the "appending standard output and standard error" description a note that >&word, where word is a number or `-', causes other redirection operators to apply for sh and Posix compatibility reasons. Suggested by Greg Wooledge 10/15 ----- pcomplete.c - change pcomp_filename_completion_function to only run the filename dequoting function in the cases (as best as it can figure) where readline won't do it via rl_filename_completion_function. Based on reports from 10/19 ----- bashline.c - attempt_shell_completion: add call to set_directory_hook() to make sure the rewrite functions are correct. It's cheap and doesn't hurt - command_word_completion_function: if completing a command name that starts with `.' or `..', temporarily suppress the effects of the `direxpand' option and restore the correct value after calling rl_filename_completion_function. If it's enabled, the directory name will be rewritten and no longer match `./' or `../'. Fixes problem reported by Michael Kalisz 10/22 ----- builtins/history.def - push_history: make sure remember_on_history is enabled before we try to delete the last history entry -- the `history -s' command might not have been saved. Fixes bug reported by lester@vmw-les.eng.vmware.com lib/readline/complete.c - rl_callback_read_char: add calls to a macro CALLBACK_READ_RETURN instead of straight return; add same call at end of function. Placeholder for future work in deinstalling signal handlers when readline is not active 10/25 ----- expr.c - exp2: catch arithmetic overflow when val1 == INTMAX_MIN and val2 == -1 for DIV and MOD and avoid SIGFPE. Bug report and pointer to fix from Jaak Ristioja - expassign: same changes for arithmetic overflow for DIV and MOD 10/28 ----- subst.c - parameter_brace_expand: allow pattern substitution when there is an expansion of the form ${var/} as a no-op: replacing nothing with nothing - parameter_brace_patsub: don't need to check for PATSUB being NULL; it never is flags.c - if STRICT_POSIX is defined, initialize history_expansion to 0, since history expansion (and its treatment of ! within double quotes) is not a conforming posix environment. From austin-group issue 500 lib/readline/histexpand.c - history_expand: when processing a string within double quotes (DQUOTE == 1), make the closing double quote inhibit history expansion, as if the word were outside double quotes. In effect, we assume that the double quote is followed by a character in history_no_expand_chars. tcsh and csh seem to do this. This answers a persistent complaint about history expansion 10/29 ----- make_cmd.c - make_arith_for_command: use skip_to_delim to find the next `;' when breaking the string between the double parens into three separate components instead of a simple character loop. Fixes bug reported by Dan Douglas 11/2 ---- Makefile.in - make libbuiltins.a depend on builtext.h to serialize its creation and avoid conflict between multiple invocations of mkbuiltins. Fix from Mike Frysinger 11/5 ---- findcmd.c - user_command_matches: if stat(".", ...) returns -1, set st_dev and st_ino fields in dotinfo to 0 to avoid same_file matches - find_user_command_in_path: check stat(2) return the same way lib/glob/glob.c - glob_vector: don't call strlen(pat) without checking pat == 0 - glob_dir_to_array: make sure to free `result' and all allocated members before returning error due to malloc failure - glob_vector: make sure to free `nextname' and `npat' on errors (mostly when setting lose = 1) - glob_vector: if flags & GX_MATCHDIRS but not GX_ALLDIRS, make sure we free `subdir' - glob_filename: when expanding ** (GX_ALLDIRS), make sure we free temp_results (return value from glob_vector) lib/glob/xmbsrtowcs.c - xdupmbstowcs: fix call to realloc to use sizeof (char *) instead of sizeof (char **) when assigning idxtmp execute_cmd.c - print_index_and_element: return 0 right away if L == 0 - is_dirname: fix memory leak by freeing `temp' - time_command: don't try to deref NULL `command' when assigning to `posix_time' - shell_execve: null-terminate `sample' after READ_SAMPLE_BUF so it's terminated for functions that expect that builtins/read.def - read_builtin: don't call bind_read_variable with a potentially-null string pcomplete.c - gen_command_matches: don't call dispose_word_desc with a NULL arg - gen_compspec_completions: fix memory leak by freeing `ret' before calling gen_action_completions (tcs, ...). happens when performing directory completion as default and no completions have been generated - gen_progcomp_completions: make sure to set foundp to 0 whenever returning NULL - it_init_aliases: fix memory leak by freeing alias_list before returning bashline.c - command_word_completion_function: don't call restore_tilde with a NULL directory_part argument - bash_directory_expansion: bugfix: don't throw away results of rl_directory_rewrite_hook if it's set and returns non-zero - bind_keyseq_to_unix_command: free `kseq' before returning error arrayfunc.c - assign_array_element_internal: make sure `akey' is freed if non-null before returning error - assign_compound_array_list: free `akey' before returning error - array_value_internal: free `akey' before returning error - unbind_array_element: free `akey' before returning error subst.c - array_length_reference: free `akey' before returning error in case of expand_assignment_string_to_string error - array_length_reference: free `akey' after call to assoc_reference - skip_to_delim: if skipping process and command substitution, free return value from extract_process_subst - parameter_brace_substring: free `val' (vtype == VT_VARIABLE) before returning if verify_substring_values fails - parameter_brace_expand: remove two duplicate lines that allocate ret in parameter_brace_substring case - parameter_brace_expand: convert `free (name); name = xmalloc (...)' to use `xrealloc (name, ...)' - parameter_brace_expand: free `name' before returning when handling ${!PREFIX*} expansion - split_at_delims: fix memory leak by freeing `d2' before returning redir.c - redirection_error: free `filename' if the redirection operator is REDIR_VARASSIGN by assigning allocname eval.c - send_pwd_to_eterm: fix memory leak by freeing value returned by get_working_directory() builtins/cd.def - change_to_directory: fix memory leak by freeing return value from resetpwd() - cd_builtin: fix memory leak by freeing value returned by dirspell() - cd_builtin: fix memory leak by freeing `directory' if appropriate before overwriting with return value from resetpwd() builtins/type.def - describe_command: free `full_path' before overwriting it with return value from sh_makepath builtins/complete.def - compgen_builtin: fix memory leak by calling strlist_dispose (sl) before overwriting sl with return value from completions_to_stringlist builtins/hash.def - list_hashed_filename_targets: fix memory leak by freeing `target' make_cmd.c - make_arith_for_command: free `init', `test', and `step' before returning error on parse error jobs.c - initialize_job_control: don't call move_to_high_fd if shell_tty == -1 general.c - check_dev_tty: don't call close with an fd < 0 - legal_number: deal with NULL `string' argument, return invalid lib/sh/fmtulong.c - fmtulong: if the `base' argument is invalid, make sure we index buf by `len-1' at maximum print_cmd.c - print_deferred_heredocs: don't try to dereference a NULL `cstring' - cprintf: make sure to call va_end (args) variables.c - push_dollar_vars: fix call to xrealloc to use sizeof (WORD_LIST *) instead of sizeof (WORD_LIST **) lib/sh/zmapfd.c - zmapfd: if read returns error, free result and return -1 immediately instead of trying to reallocate it 11/6 ---- execute_cmd.c - cpl_reap: rewrote to avoid using pointer after freeing it; now builds new coproc list on the fly while traversing the old one and sets the right values for coproc_list when done 11/12 ----- builtins/set.def - if neither -f nor -v supplied, don't allow a readonly function to be implicitly unset. Fixes bug reported by Jens Schmidt lib/readline/callback.c - change CALLBACK_READ_RETURN to clear signal handlers before returning from rl_callback_read_char so readline's signal handlers aren't installed when readline doesn't have control. Idea from Jan Kratochvil and the GDB development team pcomplete.h - COPT_NOQUOTE: new complete/compgen option value builtins/complete.def - noquote: new complete/compgen option; will be used to disable filename completion quoting pcomplete.c - pcomp_set_readline_variables: pay attention to COPT_NOQUOTE; turns of rl_filename_quoting_desired if set; turns it on if unset (value is inverted, since default is on) doc/bash.1,lib/readline/doc/rluser.texi - document new -o noquote option to complete/compgen/compopt pathexp.c - quote_string_for_globbing: if QGLOB_REGEXP, make sure characters between brackets in an ERE bracket expression are not inappropriately quoted with backslashes. This is a pretty substantial change, should be stressed when opening bash up for alpha and beta tests. Fixes bug pointed out by Stephane Chazleas doc/{bash.1,bashref.texi} - document that regexp matches can be inconsistent when quoting characters in bracket expressions, since usual quoting characters lose their meaning within brackets - note that regular expression matching when the pattern is stored in a shell variable which is quoted for expansion causes string matching redir.h - RX_SAVEFD: new flag value; notes that a redirection denotes an fd used to save another even if it's not >= SHELL_FD_BASE redir.c - do_redirection_internal: when deciding whether or not to reset the close-on-exec flag on a restored file descriptor, trust the value of redirect->flags & RX_SAVCLEXEC even if the fd is < SHELL_FD_BASE if the RX_SAVEFD flag is set - add_undo_redirect: set the RX_SAVEFD flag if the file descriptor limit is such that the shell can't duplicate to a file descriptor >= 10. Fixes a limitation that tripped a coreutils test reported by Paul Eggert 11/19 ----- doc/{bash.1,bashref.texi},lib/readline/doc/hsuser.texi - make it clear that bash runs HISTFILESIZE=$HISTSIZE after reading the startup files - make it clear that bash runs HISTSIZE=500 after reading the startup files - make it clear that setting HISTSIZE=0 causes commands to not be saved in the history list - make it clear that setting HISTFILESIZE=0 causes the history file to be truncated to zero size variables.c - sv_histsize: change so setting HISTSIZE to a value less than 0 causes the history to be `unstifled' - sv_histsize: change so setting HISTFILESIZE to a value less than 0 results in no file truncation - make it clear that numeric values less than 0 for HISTFILESIZE or HISTSIZE inhibit the usual functions 11/23 ----- parse.y - save_input_line_state: add missing `return ls' at the end, since the function is supposed to return its argument. Pointed out by Andreas Schwab builtins/read.def - skip over NUL bytes in input, as most modern shells seem to. Bug report by Matthew Story lib/readline/vi_mode.c - rl_vi_replace: set _rl_vi_last_key_before_insert to invoking key 11/25 ----- builtins/read.def - read_builtin: if xrealloc returns same pointer as first argument, don't bother with the remove_unwind_protect/add_unwind_protect pair - read_builtin: set a flag (`reading') around calls to zread/zreadc and readline() - sigalrm: change to set flag (`sigalrm_seen') and only longjmp if currently in read(2) (reading != 0) - CHECK_ALRM: new macro, checks sigalrm_seen and longjmps if non-zero, behavior of old SIGALRM catching function - read_builtin: call CHECK_ALRM in appropriate places while reading line of input. Fixes bug reported by Pierre Gaston lib/readline/vi_mode.c - rl_vi_replace: initialize characters before printing characters in vi_replace_keymap to their default values in vi_insertion_keymap, since we're supposed to be in insert mode replacing characters - rl_vi_replace: call rl_vi_start_inserting to set last command to `R' for undo - rl_vi_replace: set _rl_vi_last_key_before_insert to `R' for future use by _rl_vi_done_inserting - vi_save_insert_buffer: new function, broke out code that copies text into vi_insert_buffer from _rl_vi_save_insert - _rl_vi_save_replace: new function, saves text modified by rl_vi_replace (using current point and vi_replace_count to figure it out) to vi_replace_buffer - _rl_vi_save_insert: call vi_save_insert_buffer - _rl_vi_done_inserting: if _rl_vi_last_key_before_insert == 'R', call _rl_vi_save_replace to save text modified in replace mode (uses vi_save_insert_buffer) - _rl_vi_replace_insert: new function, replaces the number of chars in vi_insert_buffer after rl_point with contents ov vi_insert_buffer - rl_vi_redo: call _rl_vi_replace_insert if last command == 'R' and there's something in vi_insert_buffer. Fixes bug with `.' not redoing the most recent `R' command, reported by Geoff Clare in readline area on savannah 11/26 ----- lib/readline/rlprivate.h - RL_SIG_RECEIVED(): evaluate to non-zero if there is a pending signal to be handled - RL_SIGINT_RECEIVED(): evaluate to non-zero if there is a pending SIGINT to be handled lib/readline/complete.c - remove all mention of _rl_interrupt_immediately - rl_completion_matches: check RL_SIG_RECEIVED after each call to the entry function, call RL_CHECK_SIGNALS if true to handle the signal - rl_completion_matches: if RL_SIG_RECEIVED evaluates to true, free and zero out the match_list this function allocated - rl_completion_matches: if the completion entry function is rl_filename_completion_function, free the contents of match_list, because that function does not keep state and will not free the entries; avoids possible memory leak pointed out by Garrett Cooper - gen_completion_matches: if RL_SIG_RECEIVED evalutes to true after calling rl_attempted_completion_function, free the returned match list and handle the signal with RL_CHECK_SIGNALS; avoids possible memory leak pointed out by Garrett Cooper - gen_completion_matches: if RL_SIG_RECEIVED evaluates to true after calling rl_completion_matches, free the returned match list and handle the signal with RL_CHECK_SIGNALS lib/readline/util.c - rl_settracefp: new utility function to set the tracing FILE * lib/readline/signals.c - _rl_sigcleanup: pointer to a function that will be called with the signal and a void * argument from _rl_handle_signal - _rl_sigcleanarg: void * that the rest of the code can set to have passed to the signal cleanup function - _rl_handle_signal: if _rl_sigcleanup set, call as (*_rl_sigcleanup) (sig, _rl_sigcleanarg) lib/readline/rlprivate.h - extern declarations for _rl_sigcleanup and _rl_sigcleanarg lib/readline/complete.c - _rl_complete_sigcleanup: signal cleanup function for completion code; calls _rl_free_match_list on _rl_sigcleanarg if signal == SIGINT - rl_complete_internal: before calling display_matches if what_to_do == `?', set _rl_sigcleanup to _rl_complete_sigcleanup so the match list gets freed on SIGINT; avoids possible memory leak pointed out by Garrett Cooper - rl_complete_internal: in default switch case, call _rl_free_match_list before returning to avoid memory leak doc/bashref.texi - start at a set of examples for the =~ regular expression matching operator, touching on keeping the pattern in a shell variable and quoting portions of the pattern to remove their special meaning 12/1 ---- lib/glob/gmisc.c - extglob_pattern: new function, returns 1 if pattern passed as an argument looks like an extended globbing pattern lib/glob/glob.c - skipname: return 0 immediately if extglob_pattern returns non-zero, let the extended globbing code do the right thing with skipping names beginning with a `.' - mbskipname: return 0 immediately if extglob_pattern returns non-zero, let the extended globbing code do the right thing with skipping names beginning with a `.'. Fixes bug reported by Yongzhi Pan 12/2 ---- lib/glob/smatch.c - patscan, patscan_wc: no longer static so other parts of the glob library can use them, renamed to glob_patscan, glob_patscan_wc lib/glob/glob.c - extern declarations for glob_patscan, glob_patscan_wc - wchkname: new function, does skipname on wchar_t pattern and dname, old body of mbskipname after converting to wide chars - extglob_skipname: new function, checks all subpatterns in an extglob pattern to determine whether or not a filename should be skipped. Calls skipname for each subpattern. Dname is only skipped if all subpatterns indicate it should be. Better fix for bug reported by Yongzhi Pan - wextglob_skipname: wide-char version of extglob_skipname, calls wchkname instead of calling back into mbskipname for each subpattern to avoid problems with char/wchar_t mismatch - skipname: call extglob_skipname if extglob_pattern returns non-zero - mbskipname: call wextglob_skipname if extglob_pattern returns non-zero - mbskipname: short-circuit immediately if no multibyte chars in pattern or filename execute_cmd.c - execute_cond_node: added parens to patmatch assignment statement to make intent clearer 12/3 ---- configure.in,config.h.in - check for imaxdiv, define HAVE_IMAXDIV if present expr.c - expassign, exp2: use imaxdiv if available. Doesn't help with checks for overflow from 10/25 12/6 ---- lib/readline/complete.c - compute_lcd_of_matches: if we're ignoring case in the matches, only use what the user typed as the lcd if it matches the first match (after sorting) up to the length of what was typed (if what the user typed is longer than the shortest of the possible matches, use the shortest common length of the matches instead). If it doesn't match, use the first of the list of matches, as if case were not being ignored. Fixes bug reported by Clark Wang 12/7 ---- builtins/cd.def - cd_builtin: add code to return error in case cd has more than one non-option argument, conditional on CD_COMPLAINS define (which is not defined anywhere) doc/{bash.1,bashref.texi} - note that additional arguments to cd following the directory name are ignored. Suggested by Vaclav Hanzl 12/10 ----- lib/readline/input.c - rl_read_key: don't need to increment key sequence length here; doing it leads to an off-by-one error lib/readline/macro.c - rl_end_kbd_macro: after off-by-one error with rl_key_sequence_length fixed, can decrement current_macro_index by rl_key_sequence_length (length of key sequence that closes keyboard macro) lib/readline/readline.c - _rl_dispatch_subseq: fix extra increment of rl_key_sequence_length when ESC maps to a new keymap and we're converting meta characters to ESC+key - _rl_dispatch_subseq: better increment of rl_key_sequence_length before we dispatch to a function in the ISFUNC case (where the second increment above should have happened) - rl_executing_keyseq: the full key sequence that ended up executing a readline command. Available to the calling application, maintained by _rl_dispatch_subseq, indexed by rl_key_sequence_length - rl_executing_key: the key that was bound to the currently-executing readline command. Same as the `key' argument to the function lib/readline/readline.h - rl_executing_keyseq: extern declaration - rl_executing_key: extern declaration - rl_key_sequence_length: declaration moved here from rlprivate.h, now part of public interface lib/readline/rlprivate.h - new extern declaration for _rl_executing_keyseq_size, buffer size for rl_executing_keyseq lib/readline/doc/rltech.texi - documented new variables: rl_executing_key, rl_executing_keyseq, rl_key_sequence_length 12/13 ----- bashline.c - bash_execute_unix_command: replace ad-hoc code that searches cmd_xmap for correct command with call to rl_function_of_keyseq using rl_executing_keyseq; now supports key sequences longer than two characters. Fixes bug reported by Michael Kazior 12/15 ----- make_cmd.c - make_function_def: don't null out source_file before calling make_command so it can be used later on when the function definition is executed execute_cmd.c - execute_intern_function: second argument is now FUNCTION_DEF * instead of COMMAND * - execute_command_internal: call execute_intern_function with the new second argument (the entire FUNCTION_DEF instead of just the command member) - execute_intern_function: if DEBUGGER is defined, call bind_function_def before calling bind_function, just like make_function_def does (might be able to take out the call in make_function_def depending on what the debugger does with it). Fixes bug reported by expr.c - more minor changes to cases of INTMAX_MIN % -1 and INTMAX_MIN / 1; fix typos and logic errors 12/16 ----- bashline.c - find_cmd_start: change flags to remove SD_NOSKIPCMD so it skips over command substitutions and doesn't treat them as command separators - attempt_shell_completion: instead of taking first return from find_cmd_name as command name to use for programmable completion, use loop to skip over assignment statements. Fixes problem reported by Raphael Droz - attempt_shell_completion: if we don't find a command name but the command line is non-empty, assume the other words are all assignment statements and flag that point is in a command position so we can do command name completion - attempt_shell_completion: if the word being completed is the first word following a series of assignment statements, and the command line is non-empty, flag that point is in a command position so we can do command name completion lib/readline/history.c - history_get_time: atol -> strtol 12/18 ----- parse.y - parser_in_command_position: external interface to the command_token_position macro for use by other parts of the shell, like the completion mechanism externs.h - extern declaration for parser_in_command_position 12/19 ----- builtins/read.def - read_builtin: make sure all calls to bind_read_variable are passed a non-null string. Fixes bug reported by Dan Douglas bashline.c - attempt_shell_completion: mark that we're in a command position if we're at the start of the line and the parser is ready to accept a reserved word or command name. Feature most recently suggested by Peng Yu 12/21 ----- lib/readline/bind.c - _rl_escchar: return the character that would be backslash-escaped to denote the control character passed as an argument ('\n' -> 'n') - _rl_isescape: return 1 if character passed is one that has a backslash escape - _rl_untranslate_macro_value: new second argument: use_escapes, if non-zero translate to backslash escapes where possible instead of using straight \C-x for control character `x'. Change callers - _rl_untranslate_macro_value: now global lib/readline/rlprivate.h - _rl_untranslate_macro_value: extern declaration lib/readline/{macro.c,readline.h} - rl_print_last_kbd_macro: new bindable function, inspired by patch from Mitchel Humpherys lib/readline/funmap.c - print-last-kbd-macro: new bindable command, bound to rl_print_last_kbd_macro lib/readline/doc/{rluser.texi,readline.3},doc/bash.1 - print-last-kbd-macro: document. lib/readline/text.c - _rl_insert_next: if we're defining a macro, make sure the key gets added to the macro text (should really audit calls to rl_read_key() and make sure the right thing is happening for all of them) bashline.[ch] - print_unix_command_map: new function, prints all bound commands in cmd_xmap using rl_macro_dumper in a reusable format builtins/bind.def - new -X option: print all keysequences bound to Unix commands using print_unix_command_map. Feature suggested by Dennis Williamson (2/2011) doc/{bash.1,bashref.texi} - document new `bind -X' option 12/24 ----- doc/{bash.1,bashref.texi} - add a couple of sentences to the description of the case modification operators making it clearer that each character of parameter is tested against the pattern, and that the pattern should only attempt to match a single character. Suggested by Bill Gradwohl 12/28 ----- shell.c - init_noninteractive: instead of calling set_job_control(0) to unconditionally turn off job control, turn on job control if forced_interactive or jobs_m_flag is set - shell_initialize: call initialize_job_control with jobs_m_flag as argument so `bash -m script' enables job control while running the script jobs.c - initialize_job_control: if the `force' argument is non-zero, turn on job control even if the shell is not currently interactive (interactive == 0) 12/29 ----- flags.h - new extern declaration for jobs_m_flag builtins/{cd,set}.def,doc/{bash.1,bashref.texi} - added text clarifying the descriptions of cd -L and -P, suggested by Padraig Brady - slight change to the description of `set -P' about resolving symbolic links lib/readline/doc/rluser.texi - Added an example to the programmable completion section: _comp_cd, a completion function for cd, with additional verbiage. Text includes a reference to the bash_completion project 1/1/2012 -------- jobs.c - set_job_status_and_cleanup: note that a job is stopped due to SIGTSTP (any_tstped) if job_control is set; there's no need to test interactive 1/5 --- quit.h - LASTSIG(): new macro, expands to signal number of last terminating signal received (terminating_signal or SIGINT) trap.c - first_pending_trap: returns lowest signal number with a trap pending - trapped_signal_received: set to the last trapped signal the shell received in trap_handler(); reset to 0 in run_pending_traps builtins/read.def - read_builtin: changes to posix-mode (posixly_correct != 0) to make `read' interruptible by a trapped signal. After the trap runs, read returns 128+sig and does not assign the partially-read line to the named variable(s). From an austin-group discussion started by David Korn 1/11 ---- doc/{bash.1,bashref.texi} - slight changes to the descriptions of the compat32 and compat40 shell options to clarify their meaning 1/12 ---- lib/readline/{colors.[ch],parse-colors.[ch]} - new files, part of color infrastructure support Makefile.in,lib/readline/Makefile.in - arrange to have colors.o and parse-colors.o added to readline library {configure,config.h}.in - check for stdbool.h, define HAVE_STDBOOL_H if found 1/14 ---- lib/readline/bind.c - colored_stats: new bindable variable, enables using colors to indicate file type when listing completions lib/readline/complete.c - _rl_colored_stats: new variable, controlled by colored-stats bindable variable - colored_stat_start, colored_stat_end: new functions to set and reset the terminal color appropriately depending on the type of the filename to be printed - print_filename: changes to print colors if `colored-stats' variable set. Changes contributed by Raphael Droz lib/readline/readline.c - rl_initialize_everything: add call to _rl_parse_colors to parse color values out of $LS_COLORS. May have to add to rl_initialize to make more dynamic if LS_COLORS changes (which doesn't happen very often, if at all) lib/readline/rlprivate.h - _rl_colored_stats: new extern declaration lib/readline/doc/{readline.3,rluser.texi},doc/bash.1 - colored-stats: document new bindable readline variable lib/readline/colors.c - _rl_print_color_indicator: call rl_filename_stat_hook before calling lstat/stat so we can get color indicators for stuff like $HOME/Applications lib/readline/complete.c - stat_char: call rl_filename_stat_hook before calling lstat/stat findcmd.[ch],execute_cmd.c - search_for_command: now takes a second `flags' argument; changed header function prototype and callers - search_for_command: if (flags & 1), put the command found in $PATH into the command hash table (previous default behavior) execute_cmd.c - is_dirname: call search_for_command with flags argument of 0 so it doesn't try to put something in the command hash table bashline.c - bash_command_name_stat_hook: a hook function for readline's filename_stat_hook that does $PATH searching the same way that execute_cmd.c:execute_disk_command() does it, and rewrites the passed filename if found. Does not put names into command hash table. This allows command name completion to take advantage of `visible-stats' and `colored-stats' settings. - executable_completion: new function, calls the directory completion hook to expand the filename before calling executable_file or executable_or_directory; change command_word_completion_function to call executable_completion. This allows $HOME/bin/[TAB] to do command completion and display alternatives 1/17 ---- pcomplete.c - gen_command_matches: now takes a new second argument: the command name as deciphered by the programmable completion code and used to look up the compspec; changed callers (gen_compspec_completions) - gen_shell_function_matches: now takes a new second argument: the command that originally caused the completion function to be invoked; changed callers (gen_compspec_completions)) - build_arg_list: now takes a new second argument: the command name corresponding to the current compspec; changed callers (gen_command_matches, gen_shell_function_matches) - build_arg_list: now uses `cmd' argument to create $1 passed to invoked command or shell function - gen_compspec_completions: if we skipped a null command at the beginning of the line (e.g., for completing `>'), add a new word for it at the beginning of the word list and increment nw and cw appropriately. This is all a partial fix for the shortcoming pointed out by Sung Pae 1/18 ---- {configure,config.h}.in - new check: check for AUDIT_USER_TTY defined in , define HAVE_DECL_AUDIT_USER_TTY if both are found lib/readline/rlconf.h - ENABLE_TTY_AUDIT_SUPPORT: new define, allows use of the Linux kernel tty auditing system if it's available and enabled lib/readline/util.c - _rl_audit_tty: new function, send a string to the kernel tty audit system lib/readline/rlprivate.h - _rl_audit_tty: new extern declaration lib/readline/readline.c - readline: call _rl_audit_tty with line to be returned before returning it if the Linux tty audit system is available and it's been enabled in rlconf.h Original patch from Miroslav Trmac; recent request from Miroslav Lichvar 1/21 ---- lib/readline/readline.c: - _rl_dispatch_subseq: add an inter-character timeout for multi-char key sequences. Suggested by . Still needs work to make a user-settable variable parse.y - shell_getc: make code that uses the pop_alias dependent on ALIAS define variables.h - sv_tz: extern define should only depend on HAVE_TZSET expr.c - expr_streval: if ARRAY_VARS is not defined, set lvalue->ind to -1; move assignment to `ind' inside define - expr_bind_array_element: declaration and uses need to be #ifdef ARRAY_VARS arrayfunc.h - AV_ALLOWALL, AV_QUOTED, AV_USEIND: define to 0 if ARRAY_VARS not defined; used in subst.c unconditionally sig.h - make the signal blocking functions not dependent on JOB_CONTROL sig.c - sigprocmask: make the replacement definition not dependent on JOB_CONTROL trap.c - use BLOCK_SIGNAL/UNBLOCK_SIGNAL instead of code dependent on HAVE_POSIX_SIGNALS and BSD signals 1/24 ---- print_cmd.c - print_redirection_list: change the conditions under which r_duplicating_output_word is mapped to r_err_and_out to more or less match those used in redir.c. Fixes bug pointed out by Dan Douglas 1/29 ---- lib/readline/signals.c - _rl_block_sigwinch,_rl_release_sigwinch: don't compile in bodies unless SIGWINCH is defined. Fixes bug reported by Pierre Muller doc/{bash.1,bashref.texi} - small modifications to the introduction to the REDIRECTION section to describe how redirections can modify file handles - small modification to the section describing base#n to make it clearer that n can be denoted using non-numerics. From a posting by Linda Walsh 2/2 --- builtins/printf.def - printf_builtin: make sure vbuf is intialized and non-null when -v is supplied, since other parts of the code assume that it's not null (e.g., bind_printf_variable()). Fixes bug reported by Jim Avera 2/4 --- lib/readline/undo.c - _rl_free_undo_list: new function, old body of rl_free_undo_list, frees undo entries in UNDO_LIST * passed as argument - rl_free_undo_list: call _rl_free_undo_list lib/readline/rlprivate.h - _rl_free_undo_list: new extern declaration - _rl_keyseq_timeout: new extern declaration (see below) lib/readline/misc.c - rl_clear_history: new function. Clears the history list and frees all associated data similar to history.c:clear_history(), but takes rl_undo_list into account and frees and UNDO_LISTs saved as `data' members of a history list entry lib/readline/doc/rltech.texi - rl_clear_history: documented lib/readline/readline.c - _rl_keyseq_timeout: new variable to hold intra-key timeout value from 1/21 fix; specified in milliseconds. Default value is 500 - _rl_dispatch_subseq: change to use _rl_keyseq_timeout as intra-key timeout if it's greater than 0; no timeout if <= 0 - _rl_dispatch_subseq: don't check for queued keyboard input if we have pushed or pending input, or if we're reading input from a macro lib/readline/bind.c - keyseq-timeout: new bindable variable, shadows _rl_keyseq_timeout - string_varlist: add keyseq-timeout - sv_seqtimeout: new function to modify value of _rl_keyseq_timeout; clamps negative values at 0 for now - _rl_get_string_variable_value: return value for keyseq-timeout doc/bash.1,lib/readline/doc/{rluser.texi,readline.3} - keyseq-timeout: documented lib/readline/isearch.c - _rl_isearch_dispatch: modification to fix from 7/18 to not use cxt->keymap and cxt->okeymap, since by the time this code is executed, they are equal. Use `f' to check for rl_insert or unbound func - _rl_isearch_dispatch: if we're switching keymaps, not in callback mode, and don't have pending or pushed input, use _rl_input_queued to resolve a potentially ambiguous key sequence. Suggested by Roger Zauner - _rl_isearch_dispatch: if we have changed keymaps and resolved to an editing function (not self-insert), make sure we stuff the right characters back onto the input after changing the keymap back so the right editing function is executed after the search is terminated. Rest of fix for bug reported by Roger Zauner 2/5 --- builtins/gen-helpfiles.c - new file: reads struct builtin and writes the long docs to files in the `helpdirs' subdirectory. The filename is given in the previously-unused `handle' member of the struct builtin. Links with `tmpbuiltins.o', which is created by Makefile to have the right long documentation. When not cross-compiling, gets the right #defines based on configuration options from config.h instead of trying to parse conditional parts of def files. Fixes shortcoming pointed out by Andreas Schwab builtins/Makefile.in - tmpbuiltins.c: new generated file, created to enable creation of separate helpfiles based on correct #defines instead of trying to parse conditional parts of def files - gen-helpfiles: new program to generate helpfiles, links with tmpbuiltins.o - HELPFILES_TARGET: new target, substituted by configure to `helpdoc' if separate helpfiles requested - targets: new target, libbuiltins.a and $(HELPFILES_TARGET) - CREATED_OBJECTS: new variable, holds created object files for make clean; changed make clean to remove created objects - helpdoc: changed to call gen-helpfiles instead of mkbuiltins Makefile.in - when building libbuiltins.a, recursively call make with `targets' argument to make sure separate helpfiles get built configure.in - substitute `helpdoc' as value of HELPFILES_TARGET if --enable-separate-helpfiles supplied as configure argument builtins/mkbuiltins.c - `-nofunctions': new argument, causes mkbuiltins to not write value for function implementing a particular builtin to struct builtin and to write document file name to `handle' member of struct builtin - no longer writes separate helpfiles; that is left to gen-helpfiles 2/8 --- subst.c - make sure last_command_exit_value is set to a non-zero value before any calls to report_error, since `-e' set will short-circuit report_error. Fixes bug reported by Ewan Mellor variables.c - make_local_array_variable: added second argument; if non-zero, function will return an existing local associative array variable instead of insisting on an indexed array variable.h,subst.c - make_local_array_variable: changed prototype and caller builtins/declare.def - declare_internal: add second arg to call to make_local_array_variable; making_array_special, which indicates we're processing an assignment like declare a[b]=c. Fixes seg fault resulting from a being an already-declared local associative array variable in a function. Ubuntu bash bug 928900. 2/14 ---- execute_cmd.c - execute_command_internal: if redirections into or out of a loop fail, don't try to free ofifo_list unless saved_fifo is non-zero. It's only valid if saved_fifo is set 2/15 ---- {arrayfunc,braces,variables}.c - last_command_exit_value: make sure it's set before any calls to report_error, since -e will cause that to exit the shell builtins/common.c - get_job_by_name: call internal_error instead of report_error so this doesn't exit the shell 2/18 ---- builtins/evalstring.c - parse_and_execute: make sure the file descriptor to be redirected to is 1 before calling cat_file. One fix for bug reported by Dan Douglas parse.y - read_token_word: don't return NUMBER if a string of all digits resolves to a number that overflows the bounds of an intmax_t. Other fix for bug reported by Dan Douglas 2/19 ---- lib/sh/strtrans.c - ansicstr: use 0x7f as the boundary for characters that translate directly from ASCII to unicode (\u and \U escapes) instead of UCHAR_MAX, since everything >= 0x80 requires more than one byte. Bug and fix from John Kearney builtins/printf.def - tescape: ditto for printf \u and \U escape sequences 2/20 ---- lib/sh/unicode.c - u32toutf8: fix to handle encodings up to six bytes long correctly (though technically UTF-8 only has characters up to 4 bytes long). Report and fix from John Kearney - u32toutf8: first argument is now an unsigned 32-bit quantity, changed callers (u32cconv) to pass c instead of wc - u32reset: new function, resets local static state to uninitialized (locale information, currently) locale.c - call u32reset whenever LC_CTYPE/LC_ALL/LANG is changed to reset the cached locale information used by u32cconv. From a report from John Kearney 2/21 ---- doc/{bash,builtins}.1 - minor changes from Bjarni Ingi Gislason lib/sh/unicode.c - u32cconv: only assume you can directly call wctomb on the passed value if __STDC_ISO_10646__ is defined and the value is <= 0x7fffffff - stub_charset: return locale as default instead of "ASCII", let rest of code decide what to do with it lib/readline/parens.c - _rl_enable_paren_matching: make paren matching work in vi insert mode. Bug report from 2/22 ---- lib/sh/shquote.c - sh_backslash_quote: quote tilde in places where it would be expanded. From a report from John Kearney 2/23 ---- execute_cmd.c - execute_pipeline: wrap the discard_unwind_frame call in #ifdef JOB_CONTROL, since the frame is only created if JOB_CONTROL is defined. Bug and fix from Doug Kehn 2/25 ---- error.c - report_error: make sure last_command_exit_value is non-zero before we call exit_shell, since the exit trap may reference it. Call exit_shell with last_command_exit_value to allow exit statuses other than 1 unicode.c - stub_charset: use local static buffer to hold charset, don't change value returned by get_locale_var. Based on idea and code from John Kearney - u32toutf16: function to convert unsigned 32-bit value (unicode) to UTF-16. From John Kearney - u32cconv: call u32toutf16 if __STDC_ISO_10646__ defined and wchar_t is two bytes, send result to wcstombs, return if not encoding error. From John Kearney - u32cconv: return UTF-8 conversion if iconv conversion to local charset is unsupported 3/2 --- lib/readline/complete.c - print_filename: if there is no directory hook, but there is a stat hook, and we want to append a slash to directories, call the stat hook before calling path_isdir on the expanded directory name. Report and pointer to fix from Steve Rago 3/3 --- builtins/evalstring.c - parse_and_execute: fix to change of 2/18: make sure the file descriptor being redirected to is 0 before calling cat_file when we see something like $(< file). Real fix for bug reported by Dan Douglas subst.c - parameter_brace_patsub: run the replacement string through quote removal even if the expansion is within double quotes, because the parser and string extract functions treat the quotes and backslashes as special. If they're treated as special, quote removal should remove them (this is the Posix position and compatible with ksh93). THIS IS NOT BACKWARDS COMPATIBLE. 3/4 --- lib/readline/complete.c - rl_menu_complete: fix to make show-all-if-ambiguous and menu-complete-display-prefix work together if both are set. Fix from Sami Pietila 3/5 --- bashline.c - dircomplete_expand_relpath: new variable, if non-zero, means that `shopt -s direxpand' should expand relative pathnames. Zero by default, not user-settable yet - bash_directory_completion_hook: if we have a relative pathname that isn't changed by canonicalization or spell checking after being appended to $PWD, then don't change what the user typed. Controlled by dircomplete_expand_relpath 3/7 --- m4/timespec.m4 - new macros, cribbed from gnulib and coreutils: find out whether we have `struct timespec' and what file includes it m4/stat-time.m4 - new macros, cribbed from gnulib and coreutils: find out whether the mtime/atime/ctime/etctime fields of struct stat are of type struct timespec, and what the name is include/stat-time.h - new file, cribbed from gnulib, with additions from coreutils: include the right file to get the struct timespec define, or provide our own replacement. Provides a bunch of inline functions to turn the appropriate members of struct stat into `struct timespec' values, zeroing out the tv_nsec field if necessary test.c - include "stat-time.h" for the nanosecond timestamp resolution stuff - stat_mtime: new function, returns struct stat and the mod time normalized into a `struct timespec' for the filename passed as the first argument - filecomp: call stat_mtime instead of sh_stat for each filename argument to get the mtime as a struct timespec - filecomp: call timespec_cmp instead of using a straight arithmetic comparison for the -nt and -ot operators, using timespec returned by stat_mtime. Added functionality requested by by Werner Fink for systems that can support it 3/10 ---- include/posixdir.h - REAL_DIR_ENTRY: remove dependency on _POSIX_SOURCE, only use feature test macros to decide whether dirent.d_ino is present and usable; define D_INO_AVAILABLE. Report and fix from Fabrizion Gennari - D_FILENO_AVAILABLE: define if we can use dirent.d_fileno lib/sh/getcwd.c - use D_FILENO_AVAILABLE to decide whether or not to compile in _path_checkino and whether or not to call it. Report and initial fix from Fabrizion Gennari lib/readline/signals.c - make sure all occurrences of SIGWINCH are protected by #ifdef sig.c - make sure all occurrences of SIGCHLD are protected by #ifdef nojobs.c - make sure SA_RESTART is defined to 0 if the OS doesn't define it version.c - show_shell_version: don't use string literals in printf, use %s. Has added benefit of removing newline from string to be translated trap.c - queue_sigchld_trap: new function, increments the number of pending SIGCHLD signals by the argument, which is by convention the number of children reaped in a call to waitchld() trap.h - queue_sigchld_trap: new extern declaration jobs.c - waitchld: if called from the SIGCHLD signal handler (sigchld > 0), then call queue_sigchld_trap to avoid running the trap in a signal handler context. Report and original fix from Siddhesh Poyarekar lib/sh/unicode.c - u32tocesc: take an unsigned 32-bit quantity and encode it using ISO C99 string notation (\u/\U) - u32cconv: call u32tocesc as a fallback instead of u32cchar - u32cconv: call u32tocesc if iconv cannot convert the character. Maybe do the same thing if iconv_open fails - u32reset: call iconv_close on localconv if u32init == 1 3/11 ---- config-top.h - CHECKWINSIZE_DEFAULT: new define, set to initial value of check_window_size (shopt checkwinsize): 0 for off, 1 for on. Default is 0 {jobs,nojobs}.c - check_window_size: default initial value to CHECKWINSIZE_DEFAULT 3/13 ---- doc/bashref.texi - change text referring to the copying restrictions to that recommended by the FSF (no Front-Cover Texts and no Back-Cover Texts) lib/readline/doc/{history,rlman,rluserman}.texi - change text referring to the copying restrictions to that recommended by the FSF (no Front-Cover Texts and no Back-Cover Texts) 3/15 ---- array.c - LASTREF_START: new macro to set the starting position for an array traversal to `lastref' if that's valid, and to the start of the array if not. Used in array_reference, array_insert, array_remove - array_remove: try to be a little smarter with lastref instead of unconditionally invalidating it 3/16 ---- array.c - array_insert: fix memory leak by deleting element to be added in the case of an error 3/18 ---- lib/sh/mbschr.c - mbschr: don't call mbrlen unless is_basic is false; devolves to a straight character-by-character run through the string 3/19 ---- stringlib.c - substring: use memcpy instead of strncpy, since we know the length and are going to add our own NUL terminator 3/20 ---- subst.c - parameter_brace_expand_rhs: if expand_string_for_rhs returns a quoted null string (a list with one element for which QUOTED_NULL(list->word->word) returns true), return the quoted null and set the flags in the returned word to indicate it. Fixes bug reported by Mark Edgar lib/sh/tmpfile.c - use random(3) instead of get_random_number to avoid perturbing the random sequence you get using $RANDOM. Bug report and fix from Jurij Mihelic 3/21 ---- config-top.h - OPTIMIZE_SEQUENTIAL_ARRAY_ASSIGNMENT: define to 1 to optimize sequential indexed array assignment patterns. Defined to 1 by default array.c - array_insert: if OPTIMIZE_SEQUENTIAL_ARRAY_ASSIGNMENT is defined, start the search at lastref (see change from 3/15) 3/27 ---- print_cmd.c - debug_print_word_list: new debugging function, prints a word list preceded by an optional string and using a caller-specified separator 4/1 --- command.h - W_ASSNGLOBAL: new flag, set to indicate declare -g execute_cmd.c - fix_assignment_words: note that we have a -g argument to an assignment builtin and set the W_ASSNGLOBAL flag in the variable word subst.c - dump_word_flags: print out W_ASSNGLOBAL if present - do_assignment_internal: only set ASS_MKLOCAL if W_ASSIGNARG is set and W_ASSNGLOBAL is not. Don't want to create a local variable even if variable_context is non-zero if ASSNGLOBAL is set. Fixes bug reported by Bill Gradwohl 4/7 --- lib/readline/readline.c - _rl_dispatch_subseq: make the `keyseq-timeout' variable apply to ESC processing when in vi mode. After hitting ESC, readline will wait up to _rl_keyseq_timeout*1000 microseconds (if set) for additional input before dispatching on the ESC and switching to command/movement mode. Completes timeout work suggested by ; this prompted by report from Barry Downes lib/sh/shmbchar.c - sh_mbsnlen: new function, returns the number of (possibly multibyte) characters in a passed string with a passed length, examining at most maxlen (third argument) bytes externs.h - sh_mbsnlen: extern declaration for new function shell.c - exit_shell: call maybe_save_shell_history if remember_on_history is set, not just in interactive shells. That means the history is saved if history is enabled, regardless of whether or not the shell is interactive doc/{bash.1,bashref.texi} - TMOUT: fix description to make it explicit that TMOUT is the timeout period for a complete line of input, not just any input. Fixes problem reported in Ubuntu bug 957303: https://bugs.launchpad.net/ubuntu/+source/bash/+bug/957303 - HISTFILE: document change to write history list to history file in any shell with history enabled, not just interactive shells. This seems to be more logical behavior. Suggested by Greg Wooledge 4/12 ---- lib/readline/colors.h - only include stdbool.h if HAVE_STDBOOL_H is defined - if HAVE_STDBOOL_H is not defined, provide enough definition for the library to use `bool', `true', and `false' lib/readline/parse-colors.[ch] - don't try to include at all; rely on colors.h to do it lib/sh/snprintf.c - vsnprintf_internal: only treat '0' as a flag to indicate zero padding if `.' hasn't been encountered ((flags&PF_DOT) == 0); otherwise treat it as the first digit of a precision specifier. Fixes bug reported by Petr Sumbera 4/15 ---- lib/sh/snprintf.c - vsnprintf_internal: if the '0' and '-' flags both occur, the '0' flag is ignored -- Posix. Start of a series of fixes based on tests and patches from Petr Sumbera - PUT_PLUS: make sure PF_PLUS flag is specified before putting the `+' - vsnprintf_internal: when '+' is read as a flag, don't set right- justify flag if the LADJUST (`-') flag has already been supplied - floating: make sure to output space padding before the `+', zero padding after - exponent: make sure to output space padding before the `+', zero padding after - exponent: only subtract one from the width for the decimal point if we're really going to print one - floating: use presence of PF_PLUS flag to decide whether to account for the `+' in the padded field width. Ditto for exponent() 4/16 ---- lib/sh/snprintf.c - vsnprint_internal: only reduce precision by 1 when processing the `g' format if it's > 0. A precision of 0 should stay 0; otherwise it gets set to -1 (NOT_FOUND) and converted to the default - number, lnumber: if an explicit precision is supplied, turn off the zero-padding flag and set the pad character back to space - number, lnumber: only account for a `+' when performing the field width calculation if the coversion is base 10; we don't add a `+' for other bases 4/18 ---- tests/printf3.sub - try using "perl -e 'print time'" to get the current time in seconds since the epoch if "date +%s" is not available (solaris 8-10) 4/19 ---- tests/run-printf - use cat -v instead of relying on diff -a being available to convert control characters to ascii and avoid the dreaded "Binary files /tmp/xx and printf.right differ" 4/20 ---- lib/sh/strftime.c - incoporated new version from Aharon Robbins 4/22 ---- doc/{bash.1,bashref.texi} - slight change to the description of /dev/tcp and /dev/udp subst.c - match_wpattern: logic fix to the calculation of `simple' (was |=, needs to be &=). Bug report from Mike Frysinger , fix from Andreas Schwab bashline.c - bash_filename_stat_hook: add code from bash_directory_completion_hook that performs pathname canonicalization in the same way that cd and other builtins will do 4/25 ---- execute_cmd.c - execute_pipeline: change the call to move_to_high_fd to make it use getdtablesize() and to not stomp on existing open file descriptors, like the fd the shell is using to read a script. Bug report from Greg Wooledge 5/6 --- subst.c - expand_word_internal: case '$': after calling param_expand and setting had_quoted_null, set TEMP to null. The code that builds the returned string at the end of the function will take care of making and returning a quoted null string if there's nothing else in ISTRING. If there is, the quoted null should just go away. Part of fix for bug reported by Ruediger Kuhlmann - expand_word_internal: when processing ISTRING to build return value, only set W_HASQUOTEDNULL in the returned word flags if the word is a quoted null string AND had_quoted_null is set. Rest of fix 5/9 --- variables.c - bind_variable_internal: if we get an array variable here (implicit assignment to index 0), call make_array_variable_value, which dummies up a fake SHELL_VAR * from array[0]. This matters when we're appending and have to use the current value - bind_variable_internal: after computing the new value, treat assoc variables with higher precedence than simple array variables; it might be that a variable has both attributes set arrayfunc.c - bind_array_var_internal: break code out that handles creating the new value to be assigned to an array variable index into a new function, make_array_variable_value. This handles creating a dummy SHELL_VAR * for implicit array[0] assignment. Fixes bug reported by Dan Douglas arrayfunc.h - make_array_variable_value: new extern declaration 5/19 ---- variables.c - bind_int_variable: if an assignment statement like x=y comes in from the expression evaluator, and x is an array, handle it like x[0]=y. Fixes bug reported by Dan Douglas 5/24 ---- braces.c - mkseq: handle possible overflow and break the sequence generating loop if it occurs. Fixes OpenSUSE bug 763591: https://bugzilla.novell.com/show_bug.cgi?id=763591 5/25 ---- Makefile.in - LDFLAGS_FOR_BUILD: add to compilation recipes for build tools buildversion, mksignames, mksyntax - LDFLAGS_FOR_BUILD: add to compilation recipes for test tools recho, zecho, printenv, xcase builtins/Makefile.in - LDFLAGS_FOR_BUILD: add to compilation recipes for build tools gen-helpfiles, psize.aux variables.c - bind_int_variable: if LHS is a simple variable name without an array reference, but resolves to an array variable, call bind_array_variable with index 0 to make x=1 equivalent to x[0]=1. Fixes bug reported by Dan Douglas 5/27 ---- subst.c - expand_word_internal: make sure has_dollar_at doesn't get reset before recursive calls to param_expand or expand_word_internal, since it has to save state of what came before. Use temp variable and make sure has_dollar_at is incremented if recursive call processes "$@". Fixes bug reported by gregrwm and supplemented by Dan Douglas doc/{bash.1,bashref.texi} - changes to the description of substring expansion inspired by suggestions from Bill Gradwohl doc/bashref.texi - added substring expansion examples inspired by suggestions from Bill Gradwohl variables.c - find_shell_variable: search for a variable in the list of shell contexts, ignore the temporary environment - find_variable_tempenv: search for a variable in the list of shell contexts, force search of the temporary environment - find_variable_notempenv: search for a variable in the list of shell contexts, don't force search of the temporary environment variables.h - find_shell_variable: extern declaration - find_variable_tempenv: extern declaration - find_variable_notempenv: extern declaration arrayfunc.c - bind_array_variable: call find_shell_variable instead of calling var_lookup directly findcmd.c - search_for_command: call find_variable_tempenv instead of find_variable_internal directly - _find_user_command_internal: call find_variable_tempenv instead of find_variable_internal directly builtins/setattr.def - set_var_attribute: call find_variable_notempenv instead of find_variable_internal directly - show_name_attributes: call find_variable_tempenv instead of find_variable_internal directly 6/1 --- sig.c - termsig_handler: don't try to save the shell history on a terminating signal any more, since it just causes too many problems on Linux systems using glibc and glibc malloc lib/readline/vi_mode.c - rl_vi_change_to: change to correctly redo `cc', since `c' is not a vi motion character. From Red Hat bug 813289 - rl_vi_delete_to: change to correctly redo `dd', since `d' is not a vi motion character - rl_vi_yank_to: change to correctly redo `yy', since `y' is not a vi motion character 6/4 --- lib/sh/mktime.c - current versions of VMS do not need to include . Fix from John E. Malmberg 6/5 --- lib/sh/eaccess.c - sh_stat: instead of using a static buffer to do the DEV_FD_PREFIX translation, use a dynamically-allocated buffer that we keep resizing. Fixes potential security hole reported by David Leverton 6/5 --- braces.c - expand_seqterm: check errno == ERANGE after calling strtoimax for rhs and incr. Part of a set of fixes from Scott McMillan - expand_seqterm: incr now of type `intmax_t', which changes arguments to mkseq - mkseq: a better fix for detecting overflow and underflow since it's undefined in C and compilers `optimize' out overflow checks. Uses ADDOVERFLOW and SUBOVERFLOW macros - mkseq: use sh_imaxabs (new macro) instead of abs() for intmax_t variables - mkseq: don't allow incr to be converted to -INTMAX_MIN - mkseq: make sure that strvec_create isn't called with a size argument greater than INT_MAX, since it only takes an int 6/6 --- braces.c - mkseq: try and be smarter about not overallocating elements in the return array if the increment is not 1 or -1 6/7 --- parse.y - history_delimiting_chars: if the parser says we're in the middle of a compound assignment (PST_COMPASSIGN), just return a space to avoid adding a stray semicolon to the history entry. Fixes bug reported by "Davide Brini" 6/8 --- bashline.c - bash_directory_completion_hook: don't attempt spelling correction on the directory name unless the direxpand option is set and we are going to replace the directory name with the corrected one in the readline line. Suggested by Linda Walsh lib/sh/shquote.c - sh_backslash_quote: now takes a third argument: flags. If non-zero, tildes are not backslash-escaped. Have to handle both printf %q, where they should be escaped, and filename completion, where they should not when used as usernames externs.h - sh_backslash_quote: declaration now takes a third argument builtins/printf.def - printf_builtin: call sh_backslash_quote with 1 as third argument so tildes get escaped {bashline,bracecomp}.c - call sh_backslash_quote with 0 as third argument so tildes are not escaped in completed words doc/bash.1 - add `coproc' to the list of reserved words. From a report by Jens Schweikhardt 6/10 ---- execute_cmd.c - line_number_for_err_trap: now global, so parse_and_execute can save and restore it with unwind-protect builtins/evalstring.c - parse_prologue: save and restore line_number_for_err_trap along with line_number - restore_lastcom: new function, unwind-protect to restore the_printed_command_except_trap - parse_prologue: use restore_lastcom to save and restore the value of the_printed_command_except_trap around calls to parse_and_execute (eval/source/.) 6/15 ---- lib/readline/complete.c - complete_fncmp: change filename comparison code to understand multibyte characters, even when doing case-sensitive or case-mapping comparisons. Fixes problem reported by Nikolay Shirokovskiy 6/20 ---- builtins/mapfile.def - mapfile: move the line count increment and check for having read the specified number of lines to the end of the loop to avoid reading an additional line with zgetline. Fixes bug reported by Dan Douglas 6/21 ---- execute_cmd.c - execute_pipeline: make sure `lastpipe_flag' is initialized to 0 on all systems, since it's tested later in the function. Fixes bug reported by John E. Malmberg 6/22 ---- mailcheck.c - file_mod_date_changed: return 0 right away if mailstat() does not return success. Fixes bug with using uninitialized values reported by szymon.kalasz@uj.edu.pl builtins/set.def - the `monitor' option is not available when the shell is compiled without job control, since the underlying `m' flag is not available nojobs.c - job_control: now declared as int variable, initialized to 0, never modified jobs.h - job_control: extern declaration no longer dependent on JOB_CONTROL execute_cmd.c - execute_pipeline: made necessary changes so `lastpipe' shell option is now available in all shells, even those compiled without JOB_CONTROL defined 6/23 ---- lib/glob/glob.c - glob_filename: check for interrupts before returning if glob_vector returns NULL or an error. Bug reported by Serge van den Boom , fix from Andreas Schwab - call run_pending_traps after each call to QUIT or test of interrupt_state, like we do in mainline shell code - glob_vector: don't call QUIT; in `if (lose)' code block; just free memory, return NULL, and let callers deal with interrupt_state or other signals and traps 6/25 ---- lib/readline/input.c - rl_read_key: restructure the loop that calls the event hook a little, so that the hook is called only after rl_gather_tyi returns no input, and any pending input is returned first. This results in better efficiency for processing pending input without calling the hook on every input character as bash-4.1 did. From a report from Max Horn 6/26 ---- trap.c - signal_is_pending: return TRUE if SIG argument has been received and a trap is waiting to execute trap.h - signal_is_pending: extern declaration lib/glob/glob.c - glob_vector: check for pending SIGINT trap each time through the loop, just like we check for interrupt_state or terminating_signal, and set `lose = 1' so we clean up after ourselves and interrupt the operation before running the trap. This may require a change later, maybe call run_pending_traps and do that if run_pending_traps returns? variables.c - sv_histtimefmt: set history_comment_character to default (`#') if it's 0 when we're turning on history timestamps. The history code uses the history comment character to prefix timestamps, and leaving it at 0 effectively removes them from the history. From a report to help-bash by Dennis Williamson 6/27 ---- lib/readline/signals.c - rl_maybe_restore_sighandler: new function, sets handler for SIG to HANDLER->sa_handler only if it's not SIG_IGN. Needs to be called on same signals set using rl_maybe_set_sighandler, which does not override an existing SIG_IGN handler (SIGALRM is ok since it does the check inline; doesn't mess with SIGWINCH) 6/30 ---- variables.h - additional defines for the new `nameref' variable attribute (att_nameref): nameref_p, nameref_cell, var_setref variables.c - find_variable_nameref: resolve SHELL_VAR V through chain of namerefs - find_variable_last_nameref: resolve variable NAME until last in a chain of possibly more than one nameref starting at shell_variables - find_global_variable_last_nameref: resolve variable NAME until last in a chain of possibly more than one nameref starting at global_variables - find_nameref_at_context: resolve SHELL_VAR V through chain of namerefs in a specific variable context (usually a local variable hash table) - find_variable_nameref_context: resolve SHELL_VAR V through chain of namerefs following a chain of varible contexts - find_variable_last_nameref_context: resolve SHELL_VAR V as in find_variable_last_context, but return the final nameref instead of what the final nameref resolves to - find_variable_tempenv, find_variable_notempenv, find_global_variable, find_shell_variable, find_variable: modified to follow namerefs - find_global_variable_noref: look up a global variable without following any namerefs - find_variable_noref: look up a shell variable without following any namerefs - bind_variable_internal: modify to follow a chain of namerefs in the global variables table; change to handle assignments to a nameref by following nameref chain - bind_variable: modify to follow chain of namerefs when binding to a local variable - unbind_variable: changes to unset nameref variables (unsets both nameref and variable it resolves to) subst.c - parameter_brace_expand_word: change to handle expanding nameref whose value is x[n] - parameter_brace_expand_indir: change to expand in ksh93-compatible way if variable to be indirected is nameref and a simple (non-array) expansion - param_expand: change to expand $foo where foo is a nameref whose value is x[n] execute_cmd.c - execute_for_command: changes to implement ksh93 semantics when index variable is a nameref builtins/setattr.def - show_var_attributes: change to add `n' to flags list if att_nameref is set builtins/set.def - unset_builtin: changes to error messages to follow nameref variables builtins/declare.def - document new -n option - declare_internal: new `-n' and `+n' options - declare_internal: handle declare -n var[=value] and declare +n var[=value] for existing and non-existant variables. Enforce restriction that nameref variables cannot be arrays. Implement semi-peculiar ksh93 semantics for typeset +n ref=value 7/5 --- variables.c - unbind_variable: unset whatever a nameref resolves to, leaving the nameref variable itself alone - unbind_nameref: new function, unsets a nameref variable, not the variable it references variables.h - unbind_nameref: extern declaration builtins/set.def - unset_builtin: modify to add -n option, which calls unbind_nameref leaving unbind_variable for the usual case. This required slight changes and additions to the test suite doc/{bash.1,bashref.texi} - document namerefs and typeset/declare/local/unset -n 7/13 ---- lib/sh/casemod.c - include shmbchar.h for is_basic and supporting pieces - sh_casemod: use _to_wupper and _to_wlower to convert wide character case instead of TOUPPER and TOLOWER. Fixes bug reported by Dennis Williamson , fix from Andreas Schwab - cval: short-circuit and return ascii value if is_basic tests true - sh_casemod: short-circuit and use non-multibyte case modification and toggling code if is_basic tests true lib/readline/signals.c - _rl_{block,release}_sigint: remove the code that actually blocks and releases the signals, since we defer signal handling until calls to RL_CHECK_SIGNALS() lib/readline/{callback,readline,util}.c - if HAVE_POSIX_SIGSETJMP is defined, use sigsetjmp/siglongjmp without saving and restoring the signal mask instead of setjmp/longjmp lib/readline/rltty.c - prepare_terminal_settings: don't mess with IXOFF setting if USE_XON_XOFF defined doc/{bash.1,bashref.texi} - add some text to the description of set -e clarifying its effect on shell functions and shell function execution. Suggested by Rainer Blome bashline.c - edit_and_execute_command: increment current_command_line_count before adding partial line to command history (for command-oriented-history because of rl_newline at beginning of function), then reset it to 0 before adding the dummy history entry to make sure the dummy entry doesn't get added to previous incomplete command. Partial fix for problem reported by Peng Yu 7/24 ---- configure.in - interix: define RECYCLES_PIDS. Based on a report from Michael Haubenwallner 7/26 ---- jobs.c - make_child: call bgp_delete on the newly-created pid unconditionally. Some systems reuse pids before cycling through an entire set of CHILD_MAX/_SC_CHILD_MAX unique pids. This is no longer dependent on RECYCLES_PIDS. Based on a report from Michael Haubenwallner support/shobj-conf - Mac OS X: drop MACOSX_DEPLOYMENT_TARGET=10.3 from the LDFLAGS. We can finally kill Panther 7/28 ---- subst.c - command_substitute: make sure last_made_pid gets reset if make_child fails execute_cmd.c - execute_command_internal: case cm_simple: decide whether or not to wait_for a child if already_making_children is non-zero, indicates that there is an unwaited-for child. More of fix for bug report from Michael Haubenwallner jobs.c - make_child: call delete_old_job (new_pid) unconditionally, don't bother to check whether or not pid wrap occurred. Rest of fix for bug report from Michael Haubenwallner 7/29 ---- shell.c - subshell_exit: new function, exits the shell (via call to sh_exit()) after calling any defined exit trap externs.h - subshell_exit: new extern declaration execute_cmd.c - execute_command_internal: make sure to call subshell_exit for {} group commands executed asynchronously (&). Part of fix for EXIT trap bug reported by Maarten Billemont sig.c - reset_terminating_signals: make sure to set termsigs_initialized back to 0, so a subsequent call to initialize_terminating_signals works right. Rest of fix for bug reported by Maarten Billemont {execute_cmd,general,jobs,mailcheck,mksyntax,test}.c builtins/{cd,fc,pushd,ulimit}.def lib/malloc/getpagesize.h lib/sh/{clktck,fpurge,inet_aton,mailstat,oslib,pathcanon,pathphys,spell,strerror}.c - make inclusion of dependent on HAVE_SYS_PARAM_H consistently 8/6 --- lib/readline/histexpand.c - history_expand_internal: now takes an additional argument saying whether the history expansion occurs within a quoted string, set to the open quote character - history_expand_internal: use new argument instead of checking prev char and initializing quoted_search_delimiter, pass qc directly to get_history_event, where it allows a matching quote to terminate a string defining an event - history_expand: change single-quote handling code so that if history_quotes_inhibit_expansion is 0, single quotes are treated like double quotes - history_expand: change call to history_expand_internal to pass new argument of `"' if double-quoted string, `'' if single-quoted string; this lets history_expand decide what is a quoted string and what is not 8/7 --- configure.in - AC_CANONICAL_BUILD: invoke for later use lib/readline/macro.c - _rl_prev_macro_key: new function, inverse of _rl_next_macro_key: backs up the index into the current macro by 1 lib/readline/rlprivate.h - _rl_prev_macro_key: extern declaration lib/readline/readline.c - _rl_dispatch_subseq, _rl_subseq_result: don't call _rl_unget_char if we're currently reading from a macro; call _rl_prev_macro_key instead. Fixes bug reported by Clark Wang 8/13 ---- builtins/evalstring.c - evalstring(): new function, wrapper around parse_and_execute. make sure we handle cases where parse_and_execute can call `return' and short-circuit without cleaning up properly. We call parse_and_execute_cleanup() then jump to the previous-saved return location builtins/common.h - extern declaration for evalstring() builtins/eval.def - eval_builtin: make sure we handle `eval " ... return"' in contexts where `return' is valid by calling evalstring(). Fixes bug with `eval return' in sourced files reported by Clark Wang trap.c - run_pending_traps: call evalstring instead of parse_and_execute. XXX - still needs to handle saving and restoring token state in the presence of `return'; could use unwind_protects for that builtins/mapfile.def - run_callback: call evalstring instead of parse_and_execute 8/15 ---- bashline.c - bash_filename_stat_hook: make sure we don't free local_dirname before using it to canonicalize any expanded filename. Make sure it always points to *dirname and only free it if we're replacing it. lib/readline/complete.c - append_to_match: make sure we call rl_filename_stat_hook with newly-allocated memory to avoid problems with freeing it twice 8/17 ---- variables.c,config-top.h - if ARRAY_EXPORT is defined to 1 when variables.c is compiled, the code that allows indexed arrays to be exported is enabled and included 8/19 ---- shell.c - call start_debugger from main() only if dollar_vars[1] != 0 (close enough to a non-interactive shell, since we can be interactive with -i while running a shell script). Fixes oddity reported by Techlive Zheng 8/20 ---- arrayfunc.c - quote_array_assignment_chars: don't bother quoting if the word has not been marked as an assignment (W_ASSIGNMENT) - quote_array_assignment_chars: turn on W_NOGLOB in the word flags so assignment statements don't undergo globbing. Partial fix for problems reported by Dan Douglas 8/21 ---- command.h - W_NOBRACE: new word flag that means to inhibit brace expansion subst.c - brace_expand_word_list: suppress brace expansion for words with W_NOBRACE flag 8/22 ---- builtins/read.def - read_builtin: don't call dequote_string on what we've read, even if we saw an escape character, unless (input_string && *input_string). We may have escaped an IFS whitespace character. Fixes seg fault reported by execute_cmd.c - execute_command_internal: set the_printed_command_except trap when about to execute a ( ... ) user subshell. For now, set it only if ERR is trapped; can relax that later. Fixes bug reported by Mike Frysinger 8/23 ---- jobs.c - remove references to first_pid and pid_wrap, since we're not using them for anything anymore 8/24 ---- subst.c - changes for W_NOBRACE everywhere appropriate: so it can be displayed for debugging, and passed out of expand_word_internal doc/{bash.1,bashref.texi} - small changes to make it clearer that the = and == operators are equivalent, and will cause pattern matching when used with [[. From a question from Michal Soltys doc/bashref.texi - some small formatting changes from Karl Berry 8/27 ---- lib/readline/doc/{history,rlman,rluserman}.texi - some small formatting changes from Karl Berry arrayfunc.c - assign_array_element_internal, assign_compound_array_list, unbind_array_element, array_value_internal: changes to make assignment statements to negative indices (a[-1]=2) and unsetting array elements using negative indices (unset 'a[-1]') work. From suggestions by Dennis Williamson and Chris F. A. Johnson subst.c - array_length_reference: changes to make length references to array elements using negative indices (${#a[-1]}) work 8/28 ---- doc/{bash.1,bashref.texi} - document new treatment of negative indices to indexed arrays when assigning, referencing, calculating length, and unsetting 8/29 ---- shell.c - show_shell_usage: add -l to list of shell invocation options (short for --login). From Red Hat bug 852469 configure.ac - renamed from configure.in, as latest autoconf versions want. Patches Stefano Lattarini MANIFEST,Makefile.in,doc/bashref.texi,support/mkconffiles - configure.in -> configure.ac 9/1 --- parse.y - read_token_word: allow words like {array[ind]} to be valid redirection words for constructs like {x} lib/readline/display.c - update_line: if the first difference between the old and new lines is completely before any invisible characters in the prompt, we should not adjust _rl_last_c_pos, since it's before any invisible characters. Fixed in two places - prompt_modechar: return a character indicating the editing mode: emacs (@), vi command (:), or vi insert (+) - _rl_reset_prompt: new function, just calls rl_expand_prompt. Will be inlined, placeholder for more changes - expand_prompt: if show-mode-in-prompt is enabled, add a character to the front of the prompt indicating the editing mode, adjusting the various variables as appropriate to keep track of the number of visible characters and number of screen positions lib/readline/bind.c - show-mode-in-prompt: new bindable boolean variable, shadowed by _rl_show_mode_in_prompt variable - hack_special_boolean_var: call _rl_reset_prompt when toggling or setting show-mode-in-prompt lib/readline/readline.c - readline_internal_setup: make sure the correct vi mode keymap is set before expanding the prompt string for the first time lib/readline/misc.c - rl_emacs_editing_mode: make sure to call _rl_reset_prompt if we're showing the editing mode in the prompt lib/readline/rlprivate.h - _rl_reset_prompt, _rl_show_mode_in_prompt: extern declarations lib/readline/vi_mode.c - rl_vi_insertion_mode: call _rl_reset_prompt - rl_vi_movement_mode: call _rl_reset_prompt. Finishes changes for showing mode in prompt string, originally requested by Miroslav Koskar and most recently by Jordan Michael Ziegler doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - document new show-mode-in-prompt variable, off by default 9/3 --- jobs.c - set_childmax: new function, external mechanism for other parts of the shell to set js.c_childmax, the number of saved exited child statuses to remember jobs.h - set_childmax: extern declaration variables.c - CHILD_MAX: new special variable, with sv_childmax function to run when it changes. Setting CHILD_MAX to a value greater than zero but less than some maximum (currently 8192) sets the number of exited child statuses to remember. set_childmax (jobs.c) ensures that the number does not drop below the posix-mandated minimum (CHILD_MAX) doc/{bash.1,bashref.texi} - CHILD_MAX: document new meaning and action when variable is set 9/5 --- redir.c - redir_varassign: call stupidly_hack_special_variables after assigning fd number to specified variable, so we can use constructs like {BASH_XTRACEFD}>foo. Suggested by Pierre Gaston 9/8 --- expr.c - readtok: invalidate previous contents of `curlval' before freeing and reallocating tokstr (which, chances are, will get the same pointer as before and render curlval inconsistent). Fixes other bug reported by Dan Douglas 9/9 --- lib/readline/complete.c - rl_username_completion_function: protect call to setpwent() with #ifdef (HAVE_GETPWENT)/#endif. Fixes bug reported by Gerd Hofmann lib/readline/display.c - rl_message: second and subsequent calls to rl_message can result in local_prompt being overwritten with new values (e.g., from the successive calls displaying the incremental search string). Need to free before overwriting if it's not the same as the value saved in saved_local_prompt. Fixes memory leak reported by Wouter Vermaelen lib/readline/{terminal.c,rlprivate.h} - move CUSTOM_REDISPLAY_FUNC and CUSTOM_INPUT_FUNC defines from terminal.c to rlprivate.h so other files can use them expr.c - expr_streval: if noeval is non-zero, just return 0 right away, short-circuiting evaluation completely. readtok will leave curtok set correctly without re-entering the evaluator at all. Rest of fix for bug reported by Dan Douglas 9/11 ---- parse.y - parse_comsub: make sure the `reserved word ok in this context' flag is preserved after we read `do' followed by whitespace. Fixes bug reported by Benoit Vaugon 9/13 ---- configure.ac,config.h.in - enable-direxpand-default: new configure option, turns the `direxpand' shell option on by default bashline.c - dircomplete_expand, dircomplete_expand_relpath: initialize to 1 if DIRCOMPLETE_EXPAND_DEFAULT is defined and non-zero doc/bashref.texi - enable-direxpand-default: document new configure option 9/14 ---- shell.c - --protected: make option valid only when wordexp is compiled into the shell. Fix from Roman Rakus configure.ac - HP NonStop (*-nsk*): compile --without-bash-malloc. Change from Joachim Schmitz 9/16 ---- subst.c,execute_cmd.c,lib/glob/sm_loop.c,lib/sh/shquote.c - minor code cleanups from Joachim Schmitz lib/readline/colors.h - workaround for HP NonStop compiler issue with from Joachim Schmitz 9/17 ---- builtins/printf.def - printf_builtin: handle localtime returning NULL, as can happen when encountering overflow. Bug report and initial fix from Eduardo A. Bustamante López doc/{bash.1,bashref.texi} - emphasize that brace expansion using character ranges ({a..c}) acts as if the C locale were in use. Prompted by message from Marcel Giannelia 9/20 ---- lib/sh/wcsnwidth.c - wcsnwidth: new function, variant of wcwidth, returns the number of wide characters from a string that will be displayed to not exceed a specified max column position 9/21 ---- builtins/help.def - show_builtin_command_help: break code that displays the short-doc for each builtin in two columns into a new function: dispcolumn - wdispcolumn: multibyte-char version of dispcolumn; uses wide chars and printf "%ls" format. Fixes problem reported by Nguyá»n Thái Ngá»c Duy 9/22 ---- execute_cmd.c - execute_disk_command: before running the command-not-found hook, call kill_current_pipeline() to make sure we don't add processes to an existing pipeline or wait for processes erroneously 9/23 ---- lib/readline/input.c - rl_input_available_hook: new hook function, called from _rl_input_available (or _rl_input_queued) to return whether or not input is available wherever the input source is lib/readline/doc/rltech.texi - rl_input_available_hook: document 9/27 ---- lib/glob/sm_loop.c: - GMATCH: after one or more `*', an instance of ?(x) can match zero or 1 times (unlike ?, which has to match one character). The old code failed if it didn't match at least once. Fixes `a*?(x)' bug. - GMATCH: if we hit the end of the search string, but not the end of the pattern, and the rest of the pattern is something that can match the NUL at the end of the search string, we should successfully match. Fixes `a*!(x)' bug reported by 10/2 ---- command.h - add c_lock member to coproc structure for future use to tell who is manipulating it execute_cmd.c - execute_coproc: block SIGCHLD while parent is forking coproc process and adding pid to sh_coproc struct to avoid race condition where child is reaped before the pid is assigned and the coproc is never marked as having died. Fixes race condition identified by Davide Baldini - add assignments to c_lock member of struct coproc in various functions that manipulate it; was used to identify race condition - coproc_pidchk: don't call coproc_dispose to avoid using malloc and other functions in a signal handler context - coproc_dispose: call BLOCK_SIGNAL/UNBLOCK_SIGNAL for SIGCHLD while manipulating the sh_coproc struct 10/6 ---- lib/readline/complete.c - rl_display_match_list: if printing completions horizontally, don't bother with spacing calculations if limit == 1, which means we are printing one completion per line no matter what. Fixes bug reported by David Kaasen 10/7 ---- builtins/declare.def - declare_internal: add error checking for nameref attribute and variable assignments: self-references, attempts to make an array variable a nameref subst.c - parameter_brace_expand: handle parameter_brace_expand_word returning &expand_param_fatal or &expand_param_error and return the appropriate error value - parameter_brace_expand_word: if a nameref variable's value is not a valid identifier, return an error - param_expand: if a nameref variable's value is not a valid identifier, return an error test.c - unary_operator: add new -R variable, returns true if variable is set and has the nameref attribute. From ksh93 builtins/test.def - add -R to description of conditional commands for help test doc/{bash.1,bashref.texi} - document new -R unary conditional operator 10/13 ----- trap.c - check_signals_and_traps: new function, convenience function for the rest of the shell to check for pending terminating and interrupt signals, and to check for and process any pending traps - any_signals_trapped: new function, returns non-zero if any signals are trapped and -1 if not trap.h - extern declaration for check_signals_and_traps bashline.c - bashline_reset: make sure we reset the event hook - bash_event_hook: call check_signals_and_traps instead of just checking for terminating signals so we can run pending traps and react to interrupts, and reset the event hook when we're done 10/14 ----- trap.c - trap_handler: if executing in a readline signal handler context, call bashline_set_event_hook to install bash_event_hook to process the signal (if bash cares about it) sig.c - sigint_sighandler: call bashline_set_event_hook to set the event hook if we're executing in a readline signal handler context lib/readline/input.c - rl_getc: call RL_CHECK_SIGNALS if read returns -1/EINTR and the caught signal is SIGINT or SIGQUIT rather than waiting until the next time around the loop - rl_getc: call rl_event_hook after calling RL_CHECK_SIGNALS to allow an application signal handler to set the event hook in its own signal handler (e.g., like bash trap_handler or sigint_sighandler) parse.y - yy_readline_get: don't set interrupt_immediately before we call readline(). Inspired by report from lanshun zhou input.c - getc_with_restart: add call to run_pending_traps after call to CHECK_TERMSIG lib/sh/zread.c - zread: call check_signals_and_traps if read() returns -1/EINTR instead of just ignoring the EINTR and deferring handling any signal that generated it builtins/mapfile.def - mapfile: don't set interrupt_immediately before calling zgetline() (which uses zread internally) builtins/read.def - read_builtin: don't set interrupt_immediately before calling zread (moved code around so that it was only being set right around calls to zread to avoid signal handler conflicts). Inspired by report from lanshun zhou - edit_line: don't set interrupt_immediately around call to readline() - include shmbutil.h - read_builtin: don't call read_mbchar unless is_basic(c) returns false for the character we just read 10/15 ----- sig.c - throw_to_top_level: if interrupt_state is non-zero, make sure that last_command_exit_value reflects 128+SIGINT if it's not already greater than 128 10/20 ----- builtins/wait.def - WAIT_RETURN: set wait_signal_received back to 0 for the potential next call to wait quit.h - CHECK_WAIT_INTR: macro to check whether trap_handler handled a signal and set wait_signal_received; longjmp to wait_intr_buf in that case jobs.c - wait_for, waitchld: call CHECK_WAIT_INTR at the same places we call CHECK_TERMSIG to check for terminating signals - wait_sigint_handler: don't longjmp out of the wait builtin unless interrupt_immediately is set; otherwise just SIGRETURN from the handler - wait_sigint_handler: if interrupt_immediately not set, but we are executing in the wait builtin and SIGINT is not trapped, treat it as a `normally received' SIGINT: restore the signal handler and send SIGINT to ourselves - waitchld: when in posix mode and running SIGCHLD traps, don't longjmp to wait_intr_buf (and let wait be interrupted) if we're running from a signal handler. Wait for CHECK_WAIT_INTR to do the longjmp. run_pending_traps will run the SIGCHLD trap later nojobs.c - reap_zombie_children, wait_for_single_pid, wait_for: call CHECK_WAIT_INTR where we call CHECK_TERMSIG - wait_sigint_handler: don't longjmp out of the wait builtin unless interrupt_immediately is set; otherwise just SIGRETURN from the handler trap.c - trap_handler: make sure wait_signal_received is set if the wait builtin is executing, and only longjmp if interrupt_immediately is set. This whole set of fixes was prompted by report from lanshun zhou 10/24 ----- lib/glob/glob.c - glob_filename: only check directory_name for globbing chars if it's of non-zero length lib/sh/strchrnul.c - new simpler implementation subst.c - command_substitute: call set_shellopts after turning off errexit in subshells so it's reflected in $SHELLOPTS 11/7 ---- builtins/evalstring.c - parse_and_execute: treat ERREXIT case like reader_loop does: set variable_context to 0 before longjmping back to top_level. Don't run the unwind-protect context to avoid side effects from popping function contexts. Part of fix for problem reported by Nikolai Kondrashov execute_cmd.c - execute_simple_command: call unlink_fifo_list only if this is the last element of a pipeline (or not in a pipeline), rather than for every child. Fixes difference in behavior between /dev/fd and FIFOs reported by Zev Weiss - execute_null_command: do the same thing in the parent branch after make_child 11/14 ----- subst.c - parameter_brace_expand: a variable is null if it's special ($@, $*), the expansion occurs within double quotes, and the expansion turns into a quoted null. Fixes debian bug 692447 reported by Matrosov Dmitriy jobs.c - run_sigchld_trap: make sure `running_trap' sentinel is set appropriately - waitchld: only run the sigchld trap if we're not in a signal handler, not running a trap, and executing the wait builtin. Otherwise, queue for later handling. We still run one instance of the trap handler per exited child. Bulk of fix for bug reported by Elliott Forney trap.c - queue_sigchld_trap: set catch_flag so run_pending_traps notices, and set trapped_signal_received for completeness. Rest of fix for bug reported by Elliott Forney lib/malloc/malloc.c - block_signals: renamed to _malloc_block_signals, made public - unblock_signals: renamed to _malloc_unblock_signals, made public lib/malloc/imalloc.h - extern declarations for _malloc_{un,}block_signals lib/malloc/table.c - mregister_alloc, mregister_free: block signals around table manipulation 11/15 ----- trap.c - run_pending_traps: set SIG_INPROGRESS flag around calls to run_sigchld_handler so other parts of the shell know that the SIGCHLD trap handler is executing - run_pending_traps: if we get a situation where we are looking at running a SIGCHLD trap but the trap string is IMPOSSIBLE_TRAP_HANDLER and the SIG_INPROGRESS flag is set, just skip it. This is possible if run_pending_traps is called from a SIGCHLD trap handler run by run_sigchld_trap doc/bash.1,lib/readline/doc/{rluser.texi,readline.3} - corrected description of the effect of `set history-size 0'. Report from Vesa-Matti J Kari include/stdc.h - CPP_STRING: new define, replaces __STRING lib/malloc/{malloc.c,imalloc.h} - replace __STRING with CPP_STRING 11/16 ----- lib/readline/bind.c - sv_histsize: if argument evaluates to a value < 0, unstifle the history 11/22 ----- redir.c - do_redirection_internal: if we have REDIR_VARASSIGN set in the redirection flags and we set up `redirector' using fcntl or dup2, don't add a redirect to make sure it stays open. Let the script programmer manage the file handle. Fixes bug reported by Sam Liddicott 11/24 ----- jobs.c - wait_for_any_job: new function, waits for an unspecified background job to exit and returns its exit status. Returns -1 on no background jobs or no children or other errors. Calls wait_for with new sentinel value ANY_PID - wait_for: changes to handle argument of ANY_PID: don't look up or try to modify the child struct, only go through the wait loop once. Return -1 if waitpid returns no children jobs.h - ANY_PID: new define builtins/wait.def - new option: -n. Means to wait for the next job and return its exit status. Returns 127 if there are no background jobs (or no children). Feature most recently requested by Elliott Forney doc/{bash.1,bashref.texi} - document new `wait -n' option execute_cmd.c - execute_command_internal: save make_command_string () result in a temp variable before calling savestring() on it; avoids evaluating make_command_string() result twice. Fix from John E. Malmberg 11/28 ----- builtins/declare.def - declare_internal: if an array variable is declared using `declare -a' or `declare -A', but not assigned a value, set the `invisible' attribute so the variable does not show up as set. Fix for bug about variable initialization reported by Tim Friske builtins/{mapfile,read}.def - after calling find_or_make_array_variable, make sure the invisible flag is turned off, in case the variable was declared previously using `declare -a' or `declare -A'. Side effect of above change to declare_internal subst.c - shell_expand_word_list: handle the W_ASSNGLOBAL flag and put -g into the list of options passed to make_internal_declare as appropriate. Fix for bug reported by Tim Friske 11/30 ----- test.c - unary_op: make sure -v and -n check that the variable is not marked as invisible before calling var_isset. Fix for bug reported by Tim Friske 12/2 ---- subst.c - process_substitute: turn off the `expanding_redir' flag, which controls whether or not variables.c:find_variable_internal uses the temporary environment to find variables. We want to use the temp environment, since we don't have to worry about order of evaluation in a subshell. Fixes bug reported by Andrey Borzenkov 12/4 ---- lib/glob/glob.c - glob_filename: changes to avoid null filenames and multiple entries returned for patterns like **/** (globstar enabled). Fixes bug reported by Ulf Magnusson 12/10 ----- lib/glob/glob.c - glob_filename: finish up a series of changes to make globstar-style globbing more efficient, avoid more duplicate filenames, and be more compatible with other shells that implement it o collapse a sequence of **/**/** to one ** o note when the directory name is all ** or ends in ** so we can treat it specially when the filename is ** All inspired by report from Andrey Borzenkov lib/sh/zread.c - zreadn: new function, like zread, but takes an additional argument saying how many bytes to read into the local buffer. Can be used to implement `read -N' without so many one-byte calls to zreadc. Code from Mike Frysinger 12/12 ----- lib/glob/sm_loop.c - PATSCAN (glob_patscan): if passed string already points to end of pattern, return NULL immediately. Fixes problem with extglob_skipname reported by Raphaël Droz 12/13 ----- execute_cmd.c - execute_coproc: handle the command's exit status being inverted (an oversight). Fixes bug reported by DJ Mills and Andreas Schwab 12/14 ----- lib/readline/readline.c - bind_arrow_keys_internal: add MINGW key bindings for Home, End, Delete, and Insert keys. Fix from Pierre Muller builtins/printf.def - printf_builtin: '%()T' conversion: if there is no argument supplied, behave as if -1 had been supplied (current time). ksh93-like feature suggested by Clark Wang doc/{bash.1,bashref.texi} - document new printf %()T default argument behavior 12/15 ----- lib/readline/display.c - displaying_prompt_first_line: new variable, indicates whether or not the first line of output is displaying the prompt. Always true in normal mode, sometimes false in horizontal scrolling mode - rl_redisplay: set displaying_prompt_first_line to true unless we are in horizontal mode; set to false in horizontal mode if the left margin of the displayed line is greater than the end of the prompt string - rl_redisplay: when in horizontal scroll mode, don't adjust _rl_last_c_pos by the wrap offset unless the line is displaying a prompt containing invisible chars - update line: don't adjust _rl_last_c_pos by the wrap offset unless the line is displaying a prompt containing invisible chars - update_line: if shrinking the line by reducing the number of displayed characters, but we have already moved the cursor to the beginning of the line where the first difference starts, don't try to delete characters builtins/read.def - unbuffered_read: set to 2 if invoked as `read -N' - if unbuffered_read is set to 2, compute the number of chars we need to read and read that many with zreadn. Posix mode still uses zreadintr. Code from Mike Frysinger doc/{bash.1,bashref.texi} - read: make it clear that if read times out, it saves any input read to that point into the variable arguments. Report from Fiedler Roman subst.c - command_substitute: change direct assignment of exit_immediately_on_error to use change_flag ('e', FLAG_OFF) instead flags.c - use errexit_flag as the variable modified by changes to the -e option, reflect those changes to exit_immediately_on_error execute_cmd.c - execute_builtin: new global variable, builtin_ignoring_errexit, set to 0 by default and set to 1 if eval/source/command executing in a context where -e should be ignored - execute_builtin: set exit_immediately_on_error to errextit_flag after executing eval/source/command in a context where -e should be ignored flags.c - if builtin_ignoring_errexit is set, changes to errexit_flag are not reflected in the setting of exit_immediately_on_error. Fixes bug reported by Robert Schiele 12/23 ----- include/posixjmp.h - setjmp_nosigs: new define, call setjmp in such a way that it will not manipulate the signal mask {expr,test,trap}.c - setjmp_nosigs: call instead of setjmp; don't need to manipulate signal mask builtins/read.def - read_builtin: setjmp_nosigs: call instead of setjmp; don't need to manipulate signal mask builtins/evalstring.c: - parse_and_execute: setjmp_nosigs: call instead of setjmp; don't need to manipulate signal mask - parse_string: setjmp_nosigs: call instead of setjmp; don't need to manipulate signal mask - parse_and_execute: save and restore the signal mask if we get a longjmp that doesn't cause us to return or exit (case DISCARD) 12/24 ----- general.c - bash_tilde_expand: only set interrupt_immediately if there are no signals trapped; we want to jump to top level if interrupted but not run any trap commands 12/25 ----- jobs.c - run_sigchld_trap: no longer set interrupt_immediately before calling parse_and_execute, even if this is no longer run in a signal handler context input.c - getc_with_restart: add call to QUIT instead of CHECK_TERMSIG parse.y - yy_stream_get: now that getc_with_restart calls QUIT, don't need to set interrupt_immediately (already had call to run_pending_traps) execute_cmd.c - execute_subshell_builtin_or_function,execute_function,execute_in_subshell: setjmp_nosigs: call instead of setjmp when saving return_catch; don't need to manipulate signal mask - execute_subshell_builtin_or_function,execute_in_subshell: setjmp_nosigs: call instead of setjmp where appropriate when saving top_level; don't need to manipulate signal mask if we're going to exit right away subst.c - command_substitute: setjmp_nosigs: call instead of setjmp when saving return_catch; don't need to manipulate signal mask - command_substitute: setjmp_nosigs: call instead of setjmp where appropriate when saving top_level; don't need to manipulate signal mask if we're going to exit right away trap.c - run_exit_trap: setjmp_nosigs: call instead of setjmp when saving return_catch; don't need to manipulate signal mask - run_exit_trap: setjmp_nosigs: call instead of setjmp where appropriate when saving top_level; don't need to manipulate signal mask if we're going to exit right away - _run_trap_internal: setjmp_nosigs: call instead of setjmp when saving return_catch; don't need to manipulate signal mask builtins/evalfile.c - _evalfile: setjmp_nosigs: call instead of setjmp when saving return_catch; don't need to manipulate signal mask builtins/evalstring.c - evalstring: setjmp_nosigs: call instead of setjmp when saving return_catch; don't need to manipulate signal mask shell.c - main: setjmp_nosigs: call instead of setjmp where appropriate when saving top_level; don't need to manipulate signal mask if we're going to exit right away - run_one_command: setjmp_nosigs: call instead of setjmp where appropriate when saving top_level; don't need to manipulate signal mask if we're going to exit right away - run_wordexp: setjmp_nosigs: call instead of setjmp where appropriate when saving top_level; don't need to manipulate signal mask if we're going to exit right away eval.c - reader_loop: save and restore the signal mask if we get a longjmp that doesn't cause us to return or exit (case DISCARD) 12/26 ----- parse.y - shell_input_line_{index,size,len}: now of type size_t; in some cases the unsigned property makes a difference - STRING_SAVER: saved_line_{size,index} now of type size_t - shell_getc: don't allow shell_input_line to grow larger than SIZE_MAX; lines longer than that are truncated until read sees a newline; addresses theoretical buffer overflow described by Paul Eggert - set_line_mbstate: size_t changes like shell_getc - shell_getc: if shell_input_line is larger than 32K, free it and start over to avoid large memory allocations sticking around variables.c - bind_global_variable: new function, binds value to a variable in the global shell_variables table variables.h - bind_global_variable: new extern declaration builtins/declare.def - declare_internal: if -g given with name=value, but variable is not found in the global variable table, make sure to call bind_global_variable so the variable is created and modified at global scope. Fixes a bug where declare -g x=y could modify `x' at a previous function scope command.h - W_ASSIGNARRAY: new word flag, compound indexed array assignment subst.h - ASS_MKGLOBAL: new assignment flag, forcing global assignment even in a function context, used by declare -g execute_cmd.c - fix_assignment_words: set W_ASSIGNARRAY flag if -a option given to declaration builtin subst.c - do_assignment_internal: explicitly handle case where we are executing in a function and we want to create a global array or assoc variable - shell_expand_word_list: call make_internal_declare if -a option given to declaration builtin (W_ASSIGNARRAY); handle -g option with it (W_ASSNGLOBAL). Fixes inconsistency noticed by Vicente Couce Diaz , where declare -ag foo=(bar) could modify array variable foo at previous function scope, not global scope 12/27 ----- bashline.c - Minix needs the third argument to tputs to be a void funtion taking an int argument, not an int-returning function. Fix from John E. Malmberg as part of VMS bash port 12/29 ----- configure.ac,version.c,patchlevel.h - bash-4.3-devel: new version, new shell compatibility level (43) subst.c - parameter_brace_patsub: put the bash-4.2 code back in from the change of 3/3 that runs the replacement string through quote removal, make it dependent on shell_compatibility_level <= 42 builtins/shopt.def - compat42: new shopt option - set_compatibility_level: change logic to set and unset various compat variables and shell_compatibility_level COMPAT - new documentation for bash-4.3 compatibility changes doc/{bash.1,bashref.texi} - compat42: document new shopt option builtins/shopt.def - set_compatibility_opts: new function, sets the various shopt compat variables based on the value of shell_compatibility_level builtins/common.h - set_compatibility_opts: new extern declaration variables.c - BASH_COMPAT: new special variable; sets the shell compatibility level. Accepts values in decimal (4.2) or integer (42) form; Unsetting variable, setting it to empty string, or setting it to out-of-range value sets the shell's compatibility level to the default for the current version. Valid values are 3.1/31 through the current version - sv_shcompat: new function implementing logic for BASH_COMPAT variables.h - sv_shcompat: new extern declaration doc/{bash.1,bashref.texi} - BASH_COMPAT: description of new variable lib/readline/complete.c - _rl_colored_stats: default back to 0 for 4.3 release branch 1/5/2013 -------- quit.h - remove spurious call to itrace in CHECK_WAIT_INTR bashline.c - bash_event_hook: if we're going to jump to top_level, make sure we clean up after readline() by calling rl_cleanup_after_signal(). Fixes bug reported against devel branch by Raphaël Droz - bash_event_hook: reset the event hook before checking for signals or traps in case we longjmp doc/{bash.1,bashref.texi} - small additions to the set -e section to make it more clear that contexts where -e is ignored extend to compound commands as well as shell functions lib/readline/readline.h - rl_signal_event_hook: new extern declaration lib/readline/input.c - rl_signal_event_hook: new variable, hook function to call when a function (currently just read(2)) is interrupted by a signal and not restarted - rl_getc: call rl_signal_event_hook instead of rl_event_hook lib/readline/doc/rltech.texi - rl_signal_event_hook: document new function bashline.c - changes to set rl_signal_event_hook instead of rl_event_hook lib/readline/readline.h - change readline version numbers to 6.3 1/6 --- doc/{bash.1,bashref.texi} - a couple of changes to the descriptions of the ERR trap and its effects based on a message from Rob Nagler 1/9 --- expr.c - expassign: invalidate curlval before freeing and NULLing tokstr to avoid aliasing issues. Fixes bug reported by Eduardo A. Bustamante López and Dan Douglas braces.c - array_concat: don't be so aggressive in trying to short-circuit. We can only short-circuit if we have a single-element array where the element is an empty string (array[0] == "" array[1] = 0x0). Existing practice requires us to replicate arrays and prefix or append empty strings. Fixes bug reported by Eduardo A. Bustamante López 1/11 ---- execute_cmd.c - execute_builtin: since mapfile uses evalstring() to run its callbacks internally, just like eval, so it needs to handle the case where the temp environment given to mapfile persists throughout the entire set of callback commands. This might be a problem with trap also, but trap isn't run in the same way. Fixes bug reported by Dan Douglas 1/13 ---- redir.c - redirection_error: before expanding the redirection word (if expandable_redirection_filename returns true), disable command substitution during expansion. Fixes bug reported by Dan Douglas subst.c - expand_word_internal: case '\\': if the next character is an IFS character, and the expansion occurs within double quotes, and the character is not one for which backslash retains its meaning, add the (escaped) '\' and the (escaped) character. Fixes bug reported by Dan Douglas 1/15 ---- builtins/cd.def - cd_builtin: make sure call to internal_getopt handles -e option. Fixes bug reported by 1/17 ---- subst.c - expand_word_list_internal: make sure tempenv_assign_error is initialized to 0 execute_cmd.c - execute_simple_command: make sure tempenv_assign_error is reset to 0 after it's tested to see if an error should force the shell to exit. Fixes problem where a the failure of a tempenv assignment preceding a non-special builtin `sticks' and causes the next special builtin to exit the shell. From a discussion on bug-bash started by douxin 1/20 ---- subst.c - parameter_brace_expand_rhs: call stupidly_hack_special_variables after assigning with ${param[:]=word} even if IFS is changing. Suggested by Dan Douglas [TENTATIVE, needs work on IFS side effects] command.h - W_GLOBEXP (which was unused) is now W_SPLITSPACE (which isn't used yet) {execute_cmd,subst,variables}.c - removed all code that mentioned W_GLOBEXP - removed mention of gnu_argv_flags and code that set it 1/22 ---- subst.c - param_expand: set W_SPLITSPACE if we expand (unquoted) $* and IFS is unset or null so we can be sure to split this on spaces no matter what happens with IFS later - expand_word_internal: note that param_expand returns W_SPLITSPACE in the returned word flags and keep track of that state with `split_on_spaces' 1/23 ---- subst.c - expand_word_internal: if split_on_spaces is non-zero, make sure we split `istring' on spaces and return the resultant word. The previous expansions should have quoted spaces in the positional parameters where necessary. Suggested by Dan Douglas execute_cmd.c - execute_command_internal: make sure any subshell forked to run a group command or user subshell at the end of a pipeline runs any EXIT trap it sets. Fixes debian bash bug 698411 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=698411 subst.c - shell_expand_word_list: fix code that creates args for and calls make_internal_declare to avoid calling it twice (missing `else' in 12/26 change) - do_assignment_internal: fix code from 12/26 change to fix problem where an existing assoc variable could be converted to an array without checking `mkassoc' 1/24 ---- builtins/evalfile.c - _evalfile: add missing `close (fd)' calls before returning to avoid fd leaks. Bug and fix from Roman Rakus 1/25 ---- builtins/read.def - read_builtin: don't try to play tricks with the top of the unwind- protect stack after read gets a SIGALRM; save input_string to new memory, run the stack, then restore input_string and assign the variables. Part of fix for bug reported by konsolebox ; the rest of the fix is with the changes in trap and signal handling and doing away with interrupt_immediately 1/26 ---- redir.c - redirection_expand, write_here_string, write_here_document: before calling any of the word expansion functions, after setting expanding_redir to 1 (which bypasses the temp environment in the variable lookup functions), call sv_ifs to reset the cached IFS- related variables set by subst.c:setifs(). This ensures that redirections will not get any IFS values that are set in the temporary environment, as Posix specifies. Then, after the word expansions, after resetting expanding_redir to 0, call sv_ifs again to make sure the cached IFS values are set from any assignments in the temporary environment. We force executing_builtin to 1 to `fool' the variable lookup functions into using any temp environment, then reset it to its old value after sv_ifs returns. This is what allows read() to use the (cached) IFS variables set in the temp environment. Fixes inconsistency reported by Dan Douglas 1/29 ---- lib/readline/display.c - update_line: fix off-by-one error when updating vis_lbreaks array in a multibyte locale that occurs when moving multibyte chars from one line down to another. Bug report and fix from Egmont Koblinger 1/30 ---- configure.ac - changed version to 4.3-alpha redir.c - redir_open: handle open returning -1/EINTR, which seems to happen a lot with FIFOs and SIGCHLD, and call QUIT to handle other signals that can interrupt open(2). Bug report and initial fix from Mike Frysinger 1/31 ---- subst.c - parameter_brace_expand: make sure to propagate the PF_ASSIGNRHS flag to parameter_brace_expand_word - parameter_brace_expand_word: make sure that if the PF_ASSIGNRHS flag is set and we are expanding ${a[@]} or ${a[*]} we set quoted to include Q_DOUBLE_QUOTES before calling array_value_internal, mirroring what we do for $@ and $*. Fixes inconsistency reported by Dan Douglas configure.ac - use AC_CHECK_TOOL instead of AC_CHECK_PROG to check for ar, since it will find $host-prefixed versions of utilities. Report and fix from Mike Frysinger builtins/setattr.def - set_var_attribute: check whether bind_variable (called when the variable whose attributes are being modified is found in the temp environment) just modified a read-only global variable, and don't bother marking the temporary variable for propagation if so. The propagation is superfluous and will result in a strange error message 2/2 --- variables.c - initialize_shell_variables: don't try to import function definitions with invalid names from the environment if already in posix mode, but create them as (invisible) exported variables so they pass through the environment. Print an error message so user knows what's wrong. Fixes bug reported by Tomas Trnka 2/9 --- builtins/read.def - sigalrm_seen, alrmbuf: now global so the rest of the shell (trap.c) can use them - sigalrm: just sets flag, no longer longjmps to alrmbuf; problem was longjmp without manipulating signal mask, leaving SIGALRM blocked quit.h - move CHECK_ALRM macro here from builtins/read.def so trap.c: check_signals() can call it trap.c - check_signals: add call to CHECK_ALRM before QUIT - check_signals_and_traps: call check_signals() instead of including CHECK_ALRM and QUIT inline. Integrating check for read builtin's SIGALRM (where zread call to check_signals_and_traps can see it) fixes problem reported by Mike Frysinger 2/12 ---- lib/glob/xmbsrtowcs.c - xdupmbstowcs2: fixed but where end of string was not handled correctly, causing loop to go past end of string in a bunch of cases. Fixes bug reported by "Dashing" 2/13 ---- builtins/pushd.def - popd_builtin: treat any argument that isn't -n or of the form [-+][[:digit:]]* as an error. Fixes problem reported by Bruce Korb 2/14 ---- configure.ac - add check for sig_atomic_t; already a placeholder for it in config.h.in 2/15 ---- subst.c - do_compound_assignment: don't call assign_compound_array_list with a NULL variable in case make_local_xxx_variable returns NULL (it will if you try to shadow a readonly or noassign variable). Fixes bug reported by Richard Tollerton 2/16 ---- variables.c - make_local_variable: print error messager if an attempt is made to create a local variable shadowing a `noassign' variable. Previously we just silently refused to do it trap.[ch] - get_original_signal: now global so rest of the shell can use it sig.c - initialize_shell_signals: install a signal handler for SIGTERM that does nothing except set a sigterm_received flag instead of ignoring it with SIG_IGN, as long as SIGTERM is not ignored when the shell is started. Use get_original_signal early to get the original handler, since we will do that later anyway - set_signal_handler: if installing sigterm_sighandler as the SIGTERM handler, make sure to add SA_RESTART flag to make it as close to SIG_IGN as possible sig.h - sigterm_sighandler: new extern declaration quit.h - RESET_SIGTERM: set sigterm_receved to 0 - CHECK_SIGTERM: check sigterm_received; if it's non-zero, treat it as a fatal signal and call termsig_handler to exit the shell jobs.c - make_child: call RESET_SIGTERM just before fork() so we can detect if the child process received a SIGTERM before it's able to change the signal handler back to what it was when the shell started (presumably SIG_DFL). Only has effect if the shell installed sigterm_sighandler for SIGTERM, interactive shells that were not started with SIG_IGN as the SIGTERM handler - make_child: call RESET_SIGTERM in the parent after fork() so the rest of the shell won't react to it execute_cmd.c - execute_simple_command: call CHECK_SIGTERM after make_child in child to catch SIGTERM received after fork() and before restoring old signal handlers - execute_disk_command: call CHECK_SIGTERM after make_child in child process after restoring old signal handlers and again just before calling shell_execve. Fixes race condition observed by Padraig Brady when testing with his `timeout' program lib/readline/display.c - open_some_spaces: new function, subset of insert_some_chars that just opens up a specified number of spaces to be overwritten - insert_some_spaces: now just calls to open_some_spaces followed by _rl_output_some_chars - update_line: use col_temp instead of recalculating it using _rl_col_width in the case where we use more columns with fewer bytes - update_line: use open_some_spaces and then output the right number of chars instead of trying to print new characters then overwrite existing characters in two separate calls. This includes removing some dodgy code and making things simpler. Fix from Egmont Koblinger - use new variable `bytes_to_insert' instead of overloading temp in some code blocks (nls - nfd, bytes that comprise the characters different in the new line from the old) 2/18 ---- redir.c - do_redirection_internal: add undoable redirection for the implicit close performed by the <&n- and >&n- redirections. Fixes bug reported by Stephane Chazelas 2/19 ---- sig.c - termsig_handler: an interactive shell killed by SIGHUP and keeping command history will try to save the shell history before exiting. This is an attempt to preserve the save-history-when-the-terminal- window-is-closed behavior 2/21 ---- braces.c - brace_expand: if a sequence expansion fails (e.g. because the integers overflow), treat that expansion as a simple string, including the braces, and try to process any remainder of the string. The remainder may include brace expansions. Derived from SuSE bug 804551 example (https://bugzilla.novell.com/show_bug.cgi?id=804551) 2/23 ---- {quit,sig}.h,sig.c - sigterm_received declaration now in sig.h; type is sig_atomic_t - sigwinch_received type now sig_atomic_t - sig.h includes bashtypes.h and if SIG_DFL not defined (same logic as trap.h) to pick up sig_atomic_t unwind_prot.c - include sig.h before quit.h (reverse order) 2/27 ---- builtins/shopt.def - reset_shopt_options: make sure check_window_size is reset to the default from config.h, not unconditionally to 0 jobs.[ch] - last_made_pid, last_asynchronous_pid: now volatile. Change from SuSE jobs.c - wait_for: if we're using sigaction to install a handler for SIGCHLD, make sure we specify SA_RESTART lib/{tilde,readline}/shell.c - get_home_dir: instead of looking in the password file every time, look once and cache the result sig.[ch] - sigwinch_received, sigterm_received: now `volatile' qualified sig.c,quit.h - interrupt_state,terminating_signal: now sig_atomic_t 3/1 --- MANIFEST,examples/* - removed around 120 files without FSF copyrights; requested by Karl Berry in early January 3/2 --- lib/malloc/malloc.c - morecore: only check whether SIGCHLD is trapped if SIGCHLD is defined doc/bashref.texi - Fixed most of the examples in the GNU Parallel section to use better shell idioms following complaints on bug-bash; added a couple of examples and smoothed out the text quit.h - include "sig.h" for sig_atomic_t lib/readline/display.c - update_line: when inserting one or more characters at the end of the display line in a non-multibyte environment, just write from the first difference to the end of the line and return. We don't have to adjust _rl_last_c_pos. This is needed to adjust from the old two-part copy to a single call to _rl_output_some_chars (change of 2/16) 3/4 --- Makefile.in,doc/Makefile.in - PACKAGE_TARNAME, docdir: new variables substituted by autoconf - OTHER_DOCS,OTHER_INSTALLED_DOCS: new variables with auxiliary documentation files to be installed into $(docdir) - install: add new rule to install $(OTHER_DOCS) - uninstall: add new rule to uninstall $(docdir)/$(OTHER_INSTALLED_DOCS) doc/bash.1 - add URL to `POSIX' file in `SEE ALSO' section; put pointer to that section in --posix and set -o posix descriptions examples/ - removed around 110 examples at the request of the FSF due to copyright issues 3/5 --- builtins/setattr.def - readonly: modified help text slightly to make it clearer that functions aren't changed or displayed unless the -f option is given. Report from 3/9 --- include/typemax.h - SIZE_MAX: define to 65535 (Posix minimum maximum) if not defined parse.y - include "typemax.h" for possible SIZE_MAX definition, make sure we include it after shell.h {braces,expr}.c - include "typemax.h" for possible INTMAX_MIN and INTMAX_MAX definitions 3/10 ---- bashline.c - bash_default_completion: make sure completion type of `!' (same as TAB but with show-all-if-ambiguous set) and glob-word-completion sets rl_filename_completion_desired to 0 so extra backslashes don't get inserted by `quoting' the completion. We can't kill all the matches because show-all-if-ambiguous needs them. Bug report from Marcel (Felix) Giannelia [bash-4.3-alpha frozen] 3/14 ---- general.c - trim_pathname: use memmove instead of memcpy since the source and destination pathnames may overlap. Report and fix from Matthew Riley 3/18 ---- configure.ac - socklen_t is defined as `unsigned int' if configure can't find it 3/20 ---- lib/readline/complete.c - S_ISVTX: since it's not defined on all platforms (Minix), make sure its use is protected with #ifdef 3/21 ---- doc/{bash.1,bashref.texi} - Added mention of ${!name[@]} and ${!name[*]} expansions to get all indices of an array. Suggested by Jonathan Leffler 3/24 ---- subst.h - SD_IGNOREQUOTE: new define for skip_to_delim; if set, means that single quotes (for now) will be treated as ordinary characters subst.c - skip_to_delim: handle SD_IGNOREQUOTE. no callers use it for now 3/25 ---- support/config.{guess,sub} - updated to versions from autoconf-2.69 3/31 ---- lib/sh/shquote.c - sh_single_quote: short-circuit quoting a single "'" instead of creating a long string with empty single-quoted strings parser.h - DOLBRACE_QUOTE2: new define, like DOLBRACE_QUOTE, but need to single- quote results of $'...' expansion because quote removal will be done later. Right now this is only done for ${word/pat/rep} parse.y - parse_matched_pair: set state to DOLBRACE_QUOTE2 for pattern substitution word expansion so we don't treat single quote specially in the pattern or replacement string - parse_matched_pair: if we're parsing a dollar-brace word expansion (${...}) and we're not treating single quote specially within double quotes, single-quote the translation of $'...' ansi-c escaped strings. Original report and fix from Eduardo A. Bustamante López subst.c - extract_dollar_brace_string: ${word/pat/rep} scanning now sets the DOLBRACE_QUOTE2 flag instead of DOLBRACE_QUOTE so we don't treat single quotes specially within a double-quoted string execute_cmd.c - fix_assignment_words: skip over assignment statements preceding a command word before trying to figure out whether or not assignment statements following a possible declaration command should be treated specially. Fixes bug reported by Dan Douglas 4/4 --- lib/readline/readline.c - _rl_dispatch_subseq: only call _rl_vi_set_last (and check whether the key is a text modification command) if the key sequence length is 1. That keeps the arrow keys from setting the last command when called in vi command mode. Fixes bug reported by Ian A. Watson 4/6 --- lib/readline/bind.c - rl_parse_and_bind: when parsing a double-quoted string as the value of a variable, make sure we skip past the leading double quote. Fix from Andreas Schwab variables.c - hash_lookup: set new local variable last_table_searched to the table a successful lookup appears in; tested in make_local_variable to solve the problem below - make_local_variable: if we find a variable with the tempenv flag set at the same `level' as variable_context', but not found in the temporary_env (temp environment preceding the builtin), return it. The temp environment preceding the function call has already been merged (in execute_function) into the list of variable contexts the function sees as shell_variables by the time this is called. Fixes inconsistency pointed out by Dan Douglas subst.c - expand_arith_string: expanded out contents of expand_string, expand_string_internal, expand_string_if_necessary to create a WORD_DESC and call call_expand_word_internal() on it directly. We don't want process substitution to be performed ( 1<(2) ) should mean something different in an arithmetic expression context. It doesn't work to just turn on the DQUOTE flag, since that means that things like ${x["expression"]} are not expanded correctly. Fixes problem pointed out by Dan Douglas 4/13 ---- subst.c - process_substitute: run the EXIT trap before exiting, as other shells seem to. Fixes problem pointed out by Dan Douglas lib/readline/readline.c - readline_internal_setup: call rl_vi_insertion_mode to enter vi mode instead of rl_vi_insert_mode to avoid resetting the saved last command information. Posix says that `.' can repeat a command that was entered on a previous line so we need to save the info. Fixes bug reported by Ian A. Watson 4/14 ---- lib/readline/complete.c - rl_completion_matches: make sure xrealloc returns something non-null (can happen when interrupted by a signal) before trying to add matches to match_list subst.c - array_remove_pattern: return NULL right away if array_variable_part returns an invisible variable - array_length_reference: handle array_variable_part returning an invisible variable - get_var_and_type: handle array_variable_part returning an invisible variable 4/15 ---- execute_cmd.c - execute_command_internal: make sure to run the EXIT trap for group commands anywhere in pipelines, not just at the end. From a point raised by Andreas Schwab variables.c - bind_int_variable: make sure invisible flag is unset. Fixes problems like "declare -ai a; : $(( a[4]=4 ));" arrayfunc.c - array_variable_part: return variable even if invisible flag set, callers must handle invisible vars 4/18 ---- builtins/set.def - unset_builtin: if -n flag given, call unset_nameref instead of unset_variable variables.c - find_variable_nameref: print warning message if nameref circular reference detected, return NULL and let caller deal with it builtins/declare.def - declare_builtin: only disallow global references at this point if we are at the global scope 5/16 ---- configure.ac - update release status to beta 5/23 ---- trap.c - run_pending_traps: save and restore pipeline around calls to evalstring() in case we get a trap while running a trap. Have to figure out the recursive running traps issue elsewhere. Fixes bug reported by Roman Rakus - run_pending_traps: make sure to set running_trap to the appropriate signal value when running a trap command - run_pending_traps: short-circuit immediately if running_trap set when invoked. Could change this later to only skip if it would run the same trap as currently being run (running_trap == sig + 1) configure.ac - add warning if bison not found lib/readline/doc/rltech.texi - new section with an example program illustrating the callback interface. Suggested by Peng Yu examples/loadables/Makefile.in - remove references to `cut' and `getconf', which were removed in early March 5/28 ---- lib/sh/pathphys.c - sh_realpath: correct inverted two arguments to call to sh_makepath. Report and fix from Julien Thomas 6/7 --- execute_cmd.c - executing_line_number: the else clauses that are conditional on various options being defined can simply be if clauses -- they are mutually exclusive and all have `return' in the body. Fixes bug reported by Flavio Medeiros 6/25 ---- lib/readline/readline.c - readline_internal_setup: only sent the meta-key enable string to the terminal if we've been told to use one and the terminal has been successfully initialized (RL_ISSTATE (RL_STATE_TERMPREPPED) != 0). Suggested by Dan Mick lib/readline/signals.c - _rl_signal_handler: call any defined signal hook after calling rl_resize_terminal when handling a SIGWINCH. We already have called the original SIGWINCH handler but will not be resending the signal to ourselves 6/27 ---- lib/readline/doc/history.3, doc/bash.1 - fix description of the `$' modifier to note that it expands to the last *word*, which is not always the last argument. Report from ariyetz@gmail.com via gnu.org RT 6/29 ---- lib/glob/smatch.c - glob_asciiranges: initialize to value of GLOBASCII_DEFAULT instead of 0 (0 if not defined) configure.ac,config.h.in - --enable-glob-asciiranges-default: new option, controls the value of GLOBASCII_DEFAULT; use it to turn globasciiranges shopt option on by default doc/bashref.texi - document new --enable-glob-asciiranges-default configure option variables.c - assign_in_env: implement += value appending semantics for assignments preceding command names 7/4 --- expr.c - set lasttok = NUM in all of the functions that result in a number, even if it's a boolean, to avoid errors with constructs like 1 * x = 1, which should be an asignment error. Fixes problem pointed out by Dan Douglas parse.y - decode_prompt_string: don't bother to call strcpy if polite_directory_format returns its argument unchanged. It's not necessary and Mac OS X 10.9 aborts because of a supposed overlapping string copy. Bug and fix from simon@hitzemann.org subst.c - parameter_brace_find_indir: new function, code from parameter_brace_expand_indir that looks up the indirectly-referenced variable, but does not expand it - parameter_brace_expand_indir: call parameter_brace_find_indir to look up indirected variable reference - get_var_and_type: call parameter_brace_find_indir if it looks like we are trying to manipulate an indirect variable reference like ${!b%%foo}. This makes a difference if !b references an array variable. Bug report from Dan Douglas 7/6 --- lib/sh/casemod.c - sh_modcase: make sure argument passed to is_basic is <= UCHAR_MAX, since cval can convert something to a wchar_t greater than UCHAR_MAX. Fixes bug reported by Tomasz Tomasik 7/8 --- lib/readline/history.c - add_history_time: if history_length == 0, referencing history_length - 1 will result in an array bounds error, so make history_length be at least 1 before going on. Fixes bug reported by Geng Sheng Liu builtins/setattr.def - show_func_attributes: display definition (if NODEFS argument is 0) and attributes for a particular function; used by `declare -fp name' builtins/declare.def - declare_internal: call show_func_attributes if -f supplied with -p. Fixes inconsistency observed by Linda Walsh builtins/common.h - new extern declaration for show_func_attributes builtins/read.def - read_builtin: check the first supplied variable name for validity before attempting to read any input, since we know we will have to at least use that one. Don't check any other names yet. Suggested by jidanni@jidanni.org 7/10 ---- redir.c - do_redirection_internal: when closing a file descriptor with r_close_this ([n]<&-) count close errors as redirection errors if errno ends up as EIO or ENOSPC. Originally reported back in April 2012 by Andrey Zaitsev 7/11 ---- redir.c - do_redirection_internal: before calling check_bash_input, make sure that we don't call check_bash_input for an asynchronous process that is replacing stdin with something else. The seek backwards affects the parent process as well, since parents and children share the file pointer. Fixes problem originally reported in March 2013 by Martin Jackson 7/13 ---- doc/{bash.1,bashref.texi} - slight change to add a description of `shopt -o' suggested by Bruce Korb 7/19 ---- lib/readline/histfile.c - history_do_write: if close returns < 0, make sure we restore the backup history file and return a non-zero value - history_truncate_file: if write or close return < 0, make sure we return a non-zero value [bash-4.3-beta frozen] 7/21 ---- lib/readline/isearch.c - rl_display_search: now takes an entire search context flags word as the second argument, instead of just reverse flag; changed callers - rl_display_search: if the search has failed, add `failed ' to the beginning of the search prompt - _rl_isearch_dispatch: if the search has failed, display the entire search string with an indication that the search failed but with the last matching line. Suggested by jidanni@jidanni.org command.h - W_ASSIGNINT: new word flag; used internally for make_internal_declare and set by fix_assignment_words execute_cmd.c - fix_assignment_words: set W_ASSIGNINT if compound assignment and -i given as option. We don't do anything with the value yet subst.c - shell_expand_word_list: rework the way the option list that is passed to make_internal_declare is created 8/1 --- doc/{bash.1,bashref.texi} - minor changes to description of $! based on a report from Chris Down arrayfunc.c - assign_array_element_internal: before trying to get an array's max index to process a negative subscript, make sure the array exists. Bug report from Geir Hauge 8/2 --- arrayfunc.c - assign_array_element_internal: before using array_max_index() when processing a negative subscript, make sure the variable is an array. if it's not, use 0 as array_max_index assuming it's a string. Fixes bug report from Geir Hauge 8/3 --- Makefile.in - pcomplete.o: add dependency on $(DEFDIR)/builtext.h. Suggested by Curtis Doty 8/5 --- lib/glob/sm_loop.c - strcompare: short-circuit and return FNM_NOMATCH if the lengths of the pattern and string (pe - p and se - s, respectively) are not equal - strcompare: don't bother trying to set *pe or *se to '\0' if that's what they already are. Fixes bug reported by Geir Hauge 8/6 --- doc/{bash.1,bashref.texi},builtins/hash.def,lib/readline/doc/rluser.texi - minor typo changes from Geir Hauge bultins/help.def - show_longdoc: avoid trying to translate the empty string because it often translates to some boilerplate about the project and translation. Report and fix from Geir Hauge 8/8 --- builtins/help.def - help_builtin: try two passes through the list of help topics for each argument: one doing exact string matching and one, if the first pass fails to find a match, doing string prefix matching like previous versions. This prevents `help read' from matching both `read' and `readonly', but allows `help r' to match everything beginning with `r'. Inspired by report from Geir Hauge 8/13 ---- builtins/fc.def - fc_builtin,fc_gethnum: calculate `real' end of the history list and use it if -0 is specified as the beginning or end of the history range to list. Doesn't work for fc -e or fc -s by design. Feature requested by Mike Fied 8/16 ---- trap.c - _run_trap_internal: use {save,restore}_parser_state instead of {save,restore}_token_state. It's more comprehensive 8/23 ---- doc/bash.1 - disown: remove repeated text. Report and fix from Thomas Hood 8/25 ---- lib/readline/rltty.c - set_special_char: fix prototype (last arg is rl_command_func_t *) sig.c - set_signal_handler: return oact.sa_handler only if sigaction succeeds; if it doesn't, return SIG_DFL (reasonable default). From https://bugzilla.redhat.com/show_bug.cgi?id=911404 bashline.c - attempt_shell_completion: fix to skip assignment statements preceding command name even if there are no programmable completions defined. From https://bugzilla.redhat.com/show_bug.cgi?id=994659 - attempt_shell_completion: if still completing command word following assignment statements, do command completion even if programmable completion defined for partial command name entered so far 8/26 ---- pcomplete.c - pcomp_filename_completion_function: make sure rl_filename_dequoting_function is non-NULL before trying to call it. Bug and fix from Andreas Schwab bashline.c - bash_command_name_stat_hook: if *name is not something we're going to look up in $PATH (absolute_program(*name) != 0), just call the usual bash_filename_stat_hook and return those results. This makes completions like $PWD/exam[TAB] add a trailing slash 9/2 --- builtins/read.def - read_builtin: before comparing what we read to the delim, make sure we are not supposed to be ignoring the delimiter (read -N). We set the delim to -1, but it's possible to read a character whose int value ends up being between -1 and -128. Fixes bug reported by Stephane Chazelas doc/{bash.1,bashref.texi} - word splitting: crib some language from Posix to make it clear that characters in IFS are treated as field *terminators*, not field *separators*. Addresses issue raised by DJ Mills lib/readline/{util.c,rldefs.h} - _rl_stricmp,_rl_strnicmp: now take const char * string arguments; changed prototype declarations 9/5 --- doc/{bash.1,bashref.texi} - [[: modify description of pattern matching to make it clear that the match is performed as if the extglob option were enabled. From Red Hat bug https://bugzilla.redhat.com/show_bug.cgi?id=1002078 9/12 ---- lib/readline/isearch.c - _rl_isearch_dispatch: if we read an ESC and it's supposed to terminate the search, make sure we check for typeahead with _rl_pushed_input_available, since installing a hook function causes typeahead to be collected in `ibuffer' (input.c). If there is any, make sure we still use the ESC as a prefix character. Bug and fix from Mike Miller 9/16 ---- builtins/{caller,cd,kill,pushd,wait}.def - builtin_usage(): make sure call to this sets return status to EX_USAGE 9/18 ---- terminal.c - rl_change_environment: new application-settable variable; if non- zero (the default), readline will modify LINES and COLUMNS in the environment when it handles SIGWINCH - _rl_get_screen_size: if rl_change_environment is non-zero, use setenv to modify LINES and COLUMNS environment variables readline.h - rl_change_environment: new extern declaration for applications 9/22 ---- configure.ac - relstatus: bumped version to bash-4.3-beta2 9/24 ---- lib/readline/readline.c - bind_arrow_keys_internal: added more key bindings for the numeric key pad arrow keys on mingw32. Patch from Pierre Muller 10/19 ----- bashline.c - maybe_restore_tilde: version of restore_tilde that honors `direxpand'; calls restore_tilde after saving directory expansion hook if necessary. Report from Andreas Schwab builtins/cd.def - -@: new option, allows cd to use `extended attributes' present in NFSv4, ZFS; idea taken from ksh93. Attributes associated with a file are presented as a directory containing the attributes as individual files. Original patch contributed by Cedric Blancher 10/20 ----- aclocal.m4 - BASH_CHECK_MULTIBYTE: check for wcwidth being broken with unicode combining characters needs a value to use when cross-compiling. Bug report from Bert Sutherland doc/{bash.1,bashref.texi} - document new -@ option to cd builtin 10/28 ----- lib/glob/{{gmisc,glob}.c,glob.h} - extglob_pattern renamed to extglob_pattern_p, declared in glob.h subst.c - expand_word_internal: typo fix: case to fix " $@\ " bug in bash-4.2 had a typo (& isexp instead of &&) 10/29 ----- input.c - getc_with_restart: make sure local_index and local_bufused are reset to 0 before returning EOF, in case we are running an interactive shell without line editing and ignoreeof is set. Report and fix from Yong Zhang lib/readline/search.c - _rl_nsearch_init: take out extra third argument to rl_message; it only matches prototype (and maybe format) in cases where PREFER_STDARG and USE_VARARGS are both undefined, which is rare 10/31 ----- subst.c - process_substitute: when opening the named pipe in the child, open without O_NONBLOCK to avoid race conditions. Happens often on AIX. Bug report and fix from Michael Haubenwallner builtins/ulimit.def - RLIMIT_NTHR: if RLIMIT_PTHREAD is not defined, but RLIMIT_NTHR is, use RLIMIT_NTHR (NetBSD) 11/5 ---- locale.c - set_default_locale_vars,set_locale_var: if TEXTDOMAINDIR has been set, and default_dir has a non-null value, call bindtextdomain(3) when TEXTDOMAIN is assigned a value. Fixes problem reported by Michael Arlt 11/6 ---- builtins/cd.def - cdxattr: only create synthetic pathname in `buf' if NDIRP argument is non-null - change_to_directory: if we have specified -@ and cdxattr returns failure, fail immediately. Fixes bug reported by Joshuah Hurst 11/12 ----- redir.c - print_redirection: change r_err_and_out (&>) and its append form, r_append_err_and_out (&>>) cases to separate redirection operator from filename by a space, in case we have a process substitution. Fixes bug reported by admn ombres 11/15 ----- execute_cmd.c - execute_simple_command: don't close process substitution fds until we are finished executing any current shell function. Partial fix for bug reported by John Dawson support/shobj-conf - add support for Darwin 13 (Mac OS X 10.9, Mavericks). Based on a report by Ludwig Schwardt 11/20 ----- [bash-4.3-rc1 frozen] 11/24 ----- builtins/printf.def - bind_printf_variable: make sure that the variable assigned to is no longer marked as invisible. Fixes bug reported by NBaH 11/28 ----- jobs.c - delete_old_job: fix off-by-one error in job index in call to internal_warning. Bug report from Peter Cordes 11/30 ----- doc/bashref.texi - add string to description of special parameters with name of special parameter prefixed by a $, so you can search for $#, for instance 12/2 ---- lib/readline/{histexpand.c - get_history_event: account for current_history() possibly returning NULL. Report and fix from Pankaj Sharma 12/11 ----- lib/readline/parse-colors.c - get_funky_string: don't call abort if we see something we can't parse; just return an error - _rl_parse_colors: if we encounter an error while parsing $LS_COLORS we need to leave _rl_color_ext_list as NULL after freeing its elements, then turn off _rl_colored_stats. Report and fix from Martin Wesdorp 12/13 ----- lib/readline/parse-colors.c - _rl_parse_colors: if we encounter an unrecognized prefix, throw an error but try to recover and go on to the next specification variables.c - make_local_variable: for new variables this function creates, set the att_invisible attribute. All callers from declare_internal. Indirectly, this is a fix for bug with `declare -n var; var=foo;' reported by Pierre Gaston - bind_variable: if assigning to nameref variable that doesn't have a value yet (e.g., with `declare -n var; var=foo'), don't try to use the unset name. Fixes a segfault reported by Pierre Gaston execute_cmd.c - execute_command_internal: make sure last_command_exit_value is set to 0 after any command executed in the background. Fixes bug reported by Martin Kealey 12/17 ----- support/config.{guess,sub} - updated to latest versions from git 12/19 ----- parse.y - struct STRING_SAVER: now has a new `flags' element, to identify the caller: alias expansion, double-paren parsing, or parse_and_execute - push_string: now sets flags to PSH_ALIAS if `ap' argument is non-NULL - push_string: now doesn't attempt to call strlen on a NULL string to set shell_input_line_size - parser_expanding_alias, parser_save_alias, parser_restore_alias: new functions to provide an external interface to push_string and pop_string; parser_save_alias sets flags element to PSH_SOURCE (could be renamed PSH_EXTERN someday) - shell_getc: when yy_getc returns '\0', instead of just testing whether the pushed_string_list is not-empty before popping it, don't pop if if the saved string has flags PSH_SOURCE, indicating that parse_and_execute set it before setting bash_input to the string. We should continue reading to the end of that string before popping back to a potential alias. Partial solution for the problem of aliases with embedded newlines containing `.' commands being executed out of order reported by Andrew Martin - shell_getc: when yy_getc returns '\0' and there is a saved string of type PSH_SOURCE, restart the read without popping the string stack if we have not read to the end of bash_input.location.string. Rest of fix for out-of-order execution problem externs.h - parser_expanding_alias, parser_save_alias, parser_restore_alias: new extern function declarations builtins/evalstring.c - pe_prologue: if the parser is expanding an alias, make sure to add an unwind-protect to restore the alias; undoes the work that will be performed by parse_and_execute/parse_string - parse_and_execute,parse_string: after calling push_stream to save bash_input, check whether or not the parser is currently expanding an alias (parser_expanding_alias() != 0). If it is, we want to save that string in the pushed_string_list, which we do with parser_save_alias. 12/23 ----- execute_cmd.c - execute_for_command: make sure to set line_number before expanding the word list, so expansion errors have the right line number. From a report from Ben Okopnik expr.c - exp2: save token pointer before calling readtok(), arrange to use saved token pointer when printing error token on a division by 0 error 12/27 ----- lib/readline/display.c - rl_redisplay: when calculating effects of invisible characters in a prompt that is split across physical screen lines to set the indices of linebreaks, don't bother testing local_prompt_prefix (line 751). That prefix doesn't matter when calculating prompt visible and invisible characters. Fixes problem reported by Jinesh Choksi Makefile.in - install: make sure to use $(DESTDIR) when installing OTHER_DOCS. Report and fix from Matthias Klose doc/texinfo.tex - updated to version of 2013-09-11 12/28 ----- lib/readline/undo.c - rl_do_undo: if we are undoing from a history entry (rl_undo_list == current_history()->data), make sure the change to rl_line_buffer is reflected in the history entry. We use the guts of rl_maybe_replace_line to do the work. Fixes problem reported by gregrwm 12/30 ----- sig.c - sigint_sighandler: if we get a SIGINT (and this signal handler is installed) while the wait builtin is running, note that we received it in the same way as jobs.c:wait_sigint_handler and return. The various wait_for functions will look for that with CHECK_WAIT_INTR. This fixes the wait builtin not being interruptible in an interactive job control shell 12/31 ----- trap.c - set_signal_hard_ignored: rename set_signal_ignored to this, since it both sets original_signals[sig] and sets the HARD_IGNORE flag - set_signal_ignored: new function, now just sets original_signals[sig] trap.h - set_signal_hard_ignored: new external declaration sig.c - initialize_terminating_signals: call set_signal_hard_ignored instead of set_signal_ignored for signals with disposition SIG_IGN when the shell starts execute_cmd.c - setup_async_signals: make sure we get the original dispositions for SIGINT and SIGQUIT before starting the subshell, and don't call set_signal_ignored because that sets original_signals[sig]. If we don't, subsequent attempts to reset handling using trap will fail because it thinks the original dispositions were SIG_IGN. Posix interpretation 751 (http://austingroupbugs.net/view.php?id=751) 1/2/2014 -------- lib/sh/stringvec.c - strvec_mcreate, strvec_mresize: versions of create and resize that use malloc and realloc, respectively, instead of xmalloc/xrealloc braces.c - expand_amble,mkseq: use strvec_mcreate/strvec_mresize so we can catch and handle memory allocation failures instead of aborting with the xmalloc/xrealloc interface lib/sh/strdup.c - strdup replacement function for ancient systems that don't have it lib/sh/itos.c - mitos: new function, itos that uses strdup instead of savestring externs.h - strvec_mcreate/strvec_mresize: new extern declarations - mitos: new extern declaration configure.ac - bash version moved to 4.3-rc2 1/6 --- doc/bash.1,lib/readline/doc/{rluser.texi,readline.3} - separate the description of what happens when readline reads the tty EOF character from the description of delete-char, leaving a note in the delete-char description about common binding for ^D. From suggestion by Parke lib/readline/doc/{version.texi,history.3,*.texi} - updated email addresses and copyright dates 1/7 --- variables.c - delete_var: new function, just removes a variable from a hash table and frees it, without doing anything else - make_variable_value: if we are trying to assign to a nameref variable, return NULL if the value is null or the empty string or not a valid identifier variables.h - delete_var: new extern declaration subst.h - ASS_NAMEREF: new define for assignments, means assigning to a nameref variable builtins/declare.def - declare_internal: if we are creating and assigning to a nameref variable, make sure the value is a valid variable name (checks done by make_variable_value via bind_variable_value) and display an error message, deleting the variable we just created, if it is not. Fixes bug reported by Peggy Russell 1/9 --- builtins/declare.def - declare_internal: turning on nameref attribute for an existing variable turns off -i/-l/-u/-c attributes (essentially the ones that cause evaluation at assignment time) for ksh93 compat builtins/setattr.def - show_name_attributes: if asked to display attributes and values for a nameref variable, don't follow the nameref chain to the end. More ksh93 compat 1/10 ---- trap.c - _run_trap_internal: use {save,restore}_parser_state instead of {save,restore}_token_state, like in run_pending_traps(); don't need to save and restore last_command_exit_value as a result - _run_trap_internal: call {save,restore}_pipeline like in run_pending_traps() - run_pending_traps: since we no longer run traps in a signal handler context, do not block and unblock the trapped signal while the trap is executing - run_pending_traps: allow recursive invocations (basically, running traps from a trap handler) with only a warning if the shell is compiled in debug mode. If a caller doesn't want this to happen, it should test running_trap > 0. signal_in_progress (sig) only works for the signals the shell handles specially bashline.c - bash_event_hook: make sure we clean up readline if interrupt_state is set, not only when SIGINT is not trapped. check_signals_and_traps will call check_signals, which calls QUIT, which will longjmp back to top_level, running the interrupt trap along the way. Fixes the problem of signal handlers being reset out from under readline, and not being set properly the next time readline is called, because signals_set_flag is still set to 1. XXX - might need to do this for other signals too? 1/11 ---- subst.h - SD_GLOB: new define for skip_to_delim; means we are scanning a glob pattern. subst.c - skip_to_delim: if flags include SD_GLOB, assume we are scanning a glob pattern. Currently only used to skip bracket expressions which may contain one of the delimiters 1/12 ---- subst.c - parameter_brace_expand: when expanding $@ as part of substring expansion, pattern substitution, or case modification, don't turn on the QUOTED_NULL flag. The code that constructs the word to be returned from expand_word_internal expects a different code path when $@ is being expanded. Fixes bug reported by Theodoros V. Kalamatianos 1/19 ---- subst.c - list_dequote_escapes: new function; analogue of list_quote_escapes pathexp.c - quote_string_for_globbing: fix case where unescaped ^A is last char in string; need to pass it through unaltered instead of turning it into a bare backslash - quote_string_for_globbing: when quoting for regexp matching in [[, don't treat backslash as a quote character; quote the backslash as any other character. Part of investigation into reports from Eduardo A. Bustamante López 1/25 ---- builtins/gen-helpfiles.c - write_helpfiles: add prototype - make sure to #undef xmalloc/xfree/xrealloc/free if USING_BASH_MALLOC is defined. the code does not use them, and we don't link against xmalloc.o. Report from Linda Walsh Makefile.in - variables.o: add dependency on builtins/builtext.h; helps with parallel builds. Report from Linda Walsh support/shobj-conf - darwin: combine the stanzas into one that will not require them to be updated on each Mac OS X release. Report and fix from Max Horn 1/27 ---- support/shobj-conf - darwin: changed the install_name embedded into the shared library to contain only the major version number, not the minor one. The idea is that the minor versions should all be API/ABI compatible, and it is better to link automatically with the latest one. Idea from Max Horn 1/29 ---- [bash-4.3-rc2 released] 1/30 ---- lib/readline/readline.h - rl_clear_history, rl_free_keymap: add extern declarations. Report from Hiroo Hayashi general.c - include trap.h for any_signals_trapped() prototype lib/sh/unicode.c - include for sprintf prototype 1/31 ---- execute_cmd.c - execute_simple_command: only posix-mode shells should exit on an assignment failure in the temporary environment preceding a special builtin. This is what the documentation and code comments have always said - execute_simple_command: make sure redirection errors, word expansion errors, and assignment errors to Posix special builtins cause a non-interactive posix mode shell to exit. Previously the shell would not exit if the failed special builtin was on the LHS of || or && pathexp.c - quote_string_for_globbing: when quoting a regular expression (QGLOB_REGEXP), allow an unquoted backslash to pass through unaltered. Don't use it as a quote character or quote it. More investigation from 1/24 and report by Mike Frysinger - quote_string_for_globbing: when quoting a regular expression (QGLOB_REGEXP), turn CTLESC CTLESC into CTLESC without adding a backslash to quote it. We should not have to quote it because it is not a character special to EREs. More investigation from 1/24 lib/glob/glob.c - glob_testdir: now takes a second flags argument (currently unused); changed prototype and callers 2/1 --- lib/glob/glob.c - glob_testdir: if flags argument includes GX_ALLDIRS (globstar), use lstat so we skip symlinks when traversing the directory tree. Originally reported by Chris Down 2/2 --- lib/readline/undo.c - rl_do_undo: make sure CUR is non-zero before dereferencing it to check cur->data against rl_undo_list. Report and fix from Andreas Schwab doc/{bash.1,bashref.texi} - added slight clarifying language to the description of $*, describing what happens when the expansion is not within double quotes 2/4 --- test.c - unary_test: add code to -v case so that it interprets `bare' array references (foo[1]) and returns true if that index has a value 2/5 --- trap.c - restore_default_signal: fix SIGCHLD special case for SIG_TRAPPED flag off but SIG_INPROGRESS mode set and handler IMPOSSIBLE_TRAP_HANDLER; continue with resetting handler in this case. maybe_set_sigchld_trap will check these things before resetting sigchld trap from run_sigchld_trap. Fixes (apparently long-standing?) problem reported by Alexandru Damian 2/6 --- lib/sh/strtrans.c - ansic_quote: fixed a bug when copying a printable character that consumes more than one byte; byte counter was not being incremented. Bug report from jidanni@jidanni.org 2/7 --- input.c - getc_with_restart: if read(2) returns -1/EINTR and interrupt_state or terminating_signal is set (which means QUIT; will longjmp out of this function), make sure the local buffer variables are zeroed out to avoid reading past the end of the buffer on the next call. Bug report from Dan Jacobson 2/9 --- bashline.c - command_word_completion_function: if a directory in $PATH contains quote characters, we need to quote them before passing the candidate path to rl_filename_completion_function, which performs dequoting on the pathname it's passed. Fixes bug reported by Ilyushkin Nikita 2/11 ---- parse.y - xparse_dolparen: save and restore shell_eof_token around call to parse_string, intead of just leaving it set to ')' - shell_getc: when -v is set, only print the command line when shell_eof_token is 0, so we don't print it multiple times when recursively entering the parser to parse $(...) commands. Fixes bug reported by Greg Wooledge [changed release status to 4.3-release] 2/13 ---- lib/sh/strtrans.c - ansic_quote: handle case where mbrtowc reports that the multibyte sequence is incomplete or invalid. Fixes bug reported by Eduardo A. Bustamante López 2/14 ---- variables.c - find_variable_nameref_context: fix a problem that caused the loop to go one context too close to the global context. In some cases, simple variable assignment would set a variable in the global context instead of a local context. Bug report from Geir Hauge 2/26 ---- [bash-4.3 released] 2/27 ---- aclocal.m4 - broken wcwidth check: fix typo reported by David Michael 2/28 ---- support/bashbug.sh - add ${BUGADDR} to error message printed if sending mail fails trap.c - _run_trap_internal: don't call {save,restore}_pipeline if running DEBUG trap; run_debug_trap calls them itself. Fixes bug reported by Moe Tunes test.c - unary_test: fix 'R' case by using find_variable_noref instead of find_variable - test_unop: add back missing 'R' case. Fixes bug reported by NBaH 3/2 --- jobs.c - end_job_control: if job control is active, we changed the terminal's process group, so make sure we restore it. Fixes bug reported by Eduardo A. Bustamante López 3/7 --- pcomplete.c - pcomp_curtxt: new variable, holds the original text to be completed as passed to the programmable completion code - pcomp_filename_completion_function: if we are running compgen (presumably in a shell function completion) and performing readline completion, check the word being completed. If it's not empty, but the original word passed to the programmable completion code is an empty string (""), call a dequoting function if one is available. This compensates for an assumption in bash-completion. Reported by Albert Shih lib/readline/readline.c - _rl_dispatch_subseq: when deciding whether or not to set vi mode's idea of the last command, use whether or not the dispatching keymap is vi_movement_keymap instead of the key sequence length. The `c', `d', and `y' commands all take motion commands as `arguments' and will produce key sequences longer than 1 character. The arrow keys will end up dispatching out of a different keymap, so the test will prevent arrow keys from setting the last command (the problem in bash-4.2). Bug report from Daan van Rossum lib/readline/vi_mode.c - _rl_vi_motion_command: convenience function to test whether a key is a vi-mode motion command lib/readline/rlprivate.h - _rl_vi_motion_command: extern declaration parse.y - parse_matched_pair: we should not skip processing single quotes in posix mode if dolbrace_state == DOLBRACE_QUOTE2 (pattern substitution). Fixes bug reported by David Sines 3/10 ---- lib/readline/readline.c - _rl_dispatch_callback: treat a return value of -1 as the end of a command dispatch sequence if the current context doesn't indicate that we're reading a multi-key sequence ((cxt->flags & KSEQ_SUBSEQ) == 0). Turn off the multikey flag and free the context chain in this case. Fixes one bug reported by Felix Yan to bug-readline list - _rl_dispatch_callback: treat a return value of > 0 the same as 0 and return from the function, since only values < 0 cause us to simulate recursion. Rest of fix for bug tracked down by Anatol Pomozov 3/11 ---- execute_cmd.c - execute_in_subshell: if a longjmp occurs, set result to EXECUTION_FAILURE only if last_command_exit_value == EXECUTION_SUCCESS; use value of last_command_exit_value otherwise. Fixes cosmetic issue reported by Dennis Lambe Jr. doc/bash.1 - shell-kill-word and shell-backward-kill-word should be documented as unbound by default. Report from Oliver Hartley trap.c - run_pending_traps: save value of $? before running trap commands in trap_saved_exit_value, like run_exit_trap - _run_trap_internal: save value of $? before running trap commands in trap_saved_exit_value, like run_exit_trap builtins/common.c - get_exitstat: when running `return' in a trap action, and it is not supplied an argument, use the saved exit status in trap_saved_exit_value. Fixes Posix problem reported by Eduardo A. Bustamante López 3/13 ---- lib/sh/shquote.c - sh_contains_quotes: new function, returns true if a given string contains any of the shell quote characters (single quote, double quote, or backslash) externs.h - sh_contains_quotes: new extern declaration pcomplete.c - pcomp_filename_completion_function: more changes for the benefit of bash-completion: if the argument is not the same as the original argument passed to the programmable completion code (pcomp_curtxt), and we are being run by compgen as part of a completion, dequote the argument as bash-completion expects. Fix for the complete-word- with-quoted-special-chars problem with bash-completion 3/17 ---- execute_cmd.c - execute_intern_function: when in posix mode, make defining a function with the same name as a special builtin a fatal error only when the shell is not interactive. Interactive shells display an error message and go on. From a discussion with Doug McIlroy 3/18 ---- arrayfunc.c - assign_compound_array_list: when using expand_assignment_string_to_string to expand the value in a ( [x]=y ) compound assignment, make sure that we convert 0x0 to "" when expanding [x]= so it doesn't appear as if the index is unset. Fixes bug reported by Geir Hauge builtins/common.c - get_exitstat: update fix of 3/11 to allow the DEBUG trap to use the current value of $? instead of the value it had before the trap action was run. This is one reason the DEBUG trap exists, and extended debug mode uses it. Might want to do this only in Posix mode doc/{bash.1,bashref.texi} - return: add language cribbed from Posix saying what happens when return is run without an argument from a trap, including the DEBUG trap exception 3/19 ---- lib/glob/gmisc.c - extglob_pattern_p: make sure ?(patlist) is flagged as an extglob pattern lib/glob/glob.c - extglob_skipname: rewrite to handle patterns that begin but do not end with an extglob pattern; change test for easy case and loop through patterns accordingly. Fixes problem with matching filenames with a leading dot reported by Stephane Chazelas - wextglob_skipname: make analogous changes 3/20 ---- Makefile.in - pass -DDEBUG down to builds in readline and history directories lib/readline/util.c - _rl_trace and related functions are now only compiled in if DEBUG is defined lib/readline/Makefile.in - substitute @DEBUG@ and pass -DDEBUG, if necessary, to compilation in LOCAL_CFLAGS 3/21 ---- parse.y - shell_getc: when checking whether or not to reallocate shell_input_line to add trailing newline, don't try to subtract from shell_input_line_size. size_t is unsigned, so if its value is less than 3 (like, say, 2), size-3 is a very large number and the string will not be reallocated. Use len+3 > size instead of len > size-3. Fixes bug reported in https://bugs.launchpad.net/ubuntu/+source/bash/+bug/1295467 3/27 ---- lib/readline/display.c - _rl_clean_up_for_exit: don't bother to call _rl_move_vert to whatever readline thinks the last displayed line is if it's 0. Two reasons: a minor optimization, and it protects against unwanted moving if this function is called twice, as it is when ^C is pressed. Fixes bug reported by Egmont Koblinger 3/28 ---- bashline.c - invalid_completion: new function, used to identify attempts to complete words that are syntax errors - attempt_shell_completion: if invalid_completion returns true for a word in a command position, punt on all completions. Fixes cosmetic issue reported by Uwe Storbeck - attempt_shell_completion: add clause so that in_command_position remains set to 1 for an empty word following a command separator like (, &, or | lib/readline/kill.c - rl_yank, rl_yank_nth_arg_internal: don't return -1 from bindable functions, return 1 instead lib/readline/text.c - rl_rubout, _rl_rubout_char, rl_delete, rl_change_case, rl_transpose_chars, rl_transpose_words, _rl_set_mark_at_pos, rl_exchange_point_and_mark, _rl_insert_next, _rl_char_search, _rl_char_search_internal: don't return -1 from bindable functions, return 1 instead lib/readline/vi_mode.c - rl_vi_end_word, rl_vi_rubout, rl_vi_delete, rl_vi_char_search, rl_vi_match, _rl_vi_set_mark, _rl_vi_goto_mark: don't return -1 from bindable functions, return 1 instead lib/readline/macro.c - rl_start_kbd_macro, rl_end_kbd_macro: don't return -1 from bindable functions, return 1 instead builtins/setattr.def - set_var_attribute: honor setting of no_invisible_vars when setting att_invisible on a variable - include "../flags.h" for no_invisible_vars builtins/declare.def - declare_internal: honor setting of no_invisible_vars when setting att_invisible on a variable - include "../flags.h" for no_invisible_vars Makefile.in,builtins/Makefile.in - make sure declare.o and setattr.o depend on flags.h execute_cmd.c - decpoint: new function, returns locale's decimal point or `.' default - mkfmt: use decpoint() to get decimal point instead of unconditionally using `.'. Fixes bug reported by Andrey Tataranovich in debian bug 741669 4/10 ---- lib/readline/rltypedefs.h - add back old Function/VFunction/etc typedefs, since other packages (python, samba) use them. Mark as deprecated using gcc and clang attributes. Report and fix from Max Horn 4/14 ---- jobs.c - run_sigchld_trap: unwind-protect value of this_shell_builtin, since it matters in some cases whether or not we are running `wait' or `eval'. Fixes bug reported by Eduardo A. Bustamante López 4/18 ---- shell.h - sh_parser_state_t: add `need_here_doc' flags member, since xparse_dolparen (via parse_command) sets it to 0 parse.y - gather_here_documents: make sure need_here_doc is > 0, since we don't want to just decrement it forever if it ends up < 0. Partial fix for bug reported by Jared Yanovich - {save,restore}_parser_state: save and restore need_here_doc flag. Rest of fix for bug reported by Jared Yanovich 4/19 ---- subst.c - cond_expand_word: since we are not supposed to be performing word splitting here, set expand_no_split_dollar_star to 1 in addition to setting W_NOSPLIT2 - expand_word_internal: if we have a case where we have an unquoted $@ but we are in a case where we don't want to split (W_NOSPLIT2), make sure we return a list consisting of a single word with the arguments separated by spaces and don't do word splitting. Fixes bug reported by Greg Wooledge from an IRC discussion builtins/hash.def - print_portable_hash_info: single-quote pathnames and hashed filenames that contain shell metacharacters. Fixes bug reported by in debian bash bug #739853 4/20 ---- lib/readline/display.c - When using horizontal scrolling, the redisplay code erases too much of the line containing successful results, so make sure we only erase to the end of the line after making sure we move the cursor to the end. Fixes bug reported by 4/23 ---- {bashhist,bashline}.c builtins{bind,help,type}.def lib/glob/glob.c, lib/intl/{loadmsgcat,localealias}.c,lib/sh/mktime.c - fixes to memory leaks uncovered by coverity scan 4/24 ---- {bashhist,subst,redir,assoc,jobs,array,trap}.c lib/intl/l10flist.c builtins/complete.def - fixes to memory leaks and other resource usage problems uncovered by coverity scan redir.c - do_redirection_internal: if dup2 fails (presumably because of a resource limit), close the file descriptor we opened before returning error 4/25 ---- config-top.h - DEFAULT_BASHRC: new define with the name of the default shell startup file bashline.c - bash_directory_completion_matches: don't dequote the directory name. If rl_completion_found_quote is non-zero, readline will dequote the filename itself. Fixes bug reported by Clark Wang 4/27 ---- subst.c - parameter_brace_expand_rhs: if parameter_brace_find_indir returns NULL or "", or if it returns something that is not a valid identifier, report an error and return &expand_wdesc_error so the error can propagate up. Fixes bug reported by Andre Holzhey 4/29 ---- subst.c - parameter_brace_substring: don't short-circuit right away if the value is NULL but we are looking at the positional parameters. Part of fix for bug reported by Pierre Gaston - pos_params: if there are no positional parameters, only short-circuit if we are looking for $1 and above. Rest of fix for bug reported by Pierre Gaston subst.h - SD_NOPROCSUB: new flag for skip_to_delim, means to not allow any process subsitutions (should not have overloaded SD_NOSKIPCMD) subst.c - skip_to_delim: honor SD_NOPROCSUB flag make_cmd.c - make_arith_for_expr: set W_NOPROCSUB flag in the created word - make_arith_for_command: set SD_NOPROCSUB in the flags argument to skip_to_delim so we don't treat <( or >( as a process substitution (we won't evaluate them in eval_arith_for_expr anyway). Fixes bug reported by Pierre Gaston 5/1 --- lib/glob/gmisc.c - glob_dirscan: new function, takes a pattern and a directory separator argument and advances the pattern to the last occurrence of the separator. Like strrchr, but understands extended glob patterns and uses glob_patscan to skip over them lib/glob/glob.c - extglob_skipname: if the extended globbing pattern is invalid, don't skip the name - glob_filename: if there is a slash in the pattern to be matched, and extglob is enabled, use glob_dirscan to find the real last occurrence of `/' to avoid being confused by slashes in extglob patterns. Fix for bug reported by Pierre Gaston 5/6 --- variables.c - make_local_variable: only set the att_invisible attribute if no_invisible_vars isn't set - find_variable_for_assignment: new function, intended to be called by code that eventually wants to assign a value to the variable; will not skip invisible variables; currently identical to find_variable - find_variable_no_invisible: new function, finds the first visible instance of variable with a given name in the variable context chain; eventually will be used to replace find_variable; separate right now for testing variables.h - find_variable_for_assignment: extern declaration - find_variable_no_invisible: extern declaration 5/7 --- variables.c - make_local_variable: don't clear `invisible' attribute if we are returning an existing local variable at the right context. Let the upper layers do that. Fixes bug reported by Dan Douglas 5/8 --- lib/readline/input.c - rl_getc: call RL_CHECK_SIGNALS if a read(2) is interrupted (-1/EINTR) by SIGALRM or SIGVTALRM (placeholder for non-keyboard-generated signals of interest) builtins/read.def - edit_line: call bashline_set_event_hook and bashline_reset_event_hook around call to readline(), so the right signal handling happens - read_builtin: make sure we add an unwind_protect call to bashline_reset_event_hook. These changes fix bug reported in https://bugs.launchpad.net/ubuntu/+source/bash/+bug/1317476 bashline.c - bash_event_hook: make sure we clean up the readline state by calling rl_cleanup_after_signal if sigalrm_seen is non-zero. The read builtin sets this when it times out 5/12 ---- doc/{bash.1,bashref.texi} - clarify language to make it clear that changing attributes of a nameref variable (e.g., export), actually changes the attributes of the referenced variable. Fixes omission noted by Jeff Haemer arrayfunc.c - bind_array_var_internal: make sure ENTRY no longer has invisible attribute before returning. Fixes bug reported by Geir Hauge 5/22 ---- execute_cmd.c - shell_execve: if execve fails and we return 127 or 126, make sure to set last_command_exit_value if a call to file_error or report_error causes the shell to exit. This ensures that the shell exits with the right value. 6/6 --- shell.c - drop_priv_mode: print an error message on setuid() failure, optionally exit if errno == EAGAIN, as it can be on Linux when RLIMIT_NPROC for the target user is exceeded. config-top.h - EXIT_ON_SETUID_FAILURE: new settable define, will cause the shell to exit if setuid fails with errno == EAGAIN 6/10 ---- parse.y - time_command_acceptable: fix so time is accepted everywhere the grammar is looking for a `compound_list'. Fixes bug reported by Dale Worley 6/12 ---- subst.c - clear_fifo_list: new function, clears FDs associated with open pipes in current FIFO list without closing the file descriptors. Can possibly be used when shell_execve fails and the shell jumps back to top_level and we don't want the shell to close the open FIFOs each time through the read-execute loop. Bug reported by Harald Koenig 6/16 ---- builtins/shopt.def - compat42: make sure the `compat42' option sets the correct variable for compatibility level. Fixes bug reported by Ondrej Oprala support/bashbug.sh - fix typo when echoing $USAGE. Report from Shantanu Kulkarni execute_cmd.c - shell_execve: before longjmp back to subshell_top_level, clear out the FIFO fd list by calling clear_fifo_list so the FDs (which we inherited from our parent) aren't closed every time through the read-eval loop. Fix for bug reported by Harald Koenig 6/18 ---- subst.c - extract_process_subst: add additional argument: xflags, allow callers to pass flags like extract_command_subst - extract_process_subst: call xparse_dolparen like command substitution to avoid problems when parsing commands constructs with embedded open parens. Fixes bug reported by Tim Friske subst.h - extract_process_subst: modified prototype for extern declaration 6/19 ---- execute_cmd.c - execute_pipeline: if running with lastpipe enabled, make sure that we check whether or not the job id is valid using INVALID_JOB before calling job_exit_status. the jobs list can get frozen and unfrozen in the presence of nested pipelines and loops and wait_for can clear a job table entry. Fixes bug reported by jobs.c - freeze_jobs_list: now returns old value of jobs_list_frozen; unused at current time jobs.h - freeze_jobs_list: change return value 6/20 ---- lib/glob/smatch.c - MEMCHR: single-byte and wide character defines (memchr/wmemchr) lib/glob/sm_loop.c - GMATCH: when the wildcards are the last element of the pattern, make sure they do not match a string containing a `/' if FNM_PATHNAME is set in FLAGS - GMATCH: when recursively calling GMATCH after we see a `*', don't try to consume the rest of the pattern with `*' if FNM_PATHNAME is set in FLAGS, just consume up to the next slash and then see whether or not the rest of the pattern matches. Fixes bug reported by Ian Kelling - GMATCH: when processing `*' in the pattern, after skipping consecutive wildcards, if we hit a literal `/' in the pattern and we're looking for a pathname, skip characters in the string until we find a `/' (no slash means the match fails), and try to match the rest of the pattern against the portion of the string after the next `/'. Picked up from gnulib/glibc pathexp.c - split_ignorespec: since split_ignorespec gets globbing patterns, make sure we call skip_to_delim with the SD_GLOB flag so delimiters that occur within bracket expressions don't delimit the pattern. Fixes problem with [[:digit:]] in GLOBIGNORE reported by Ian Kelling unwind_prot.c - unwind_protect_tag_on_stack: new function, returns 1 if unwind-protect frame corresponding to `tag' argument is on unwind-protect stack unwind_prot.h - unwind_protect_tag_on_stack: extern declaration 6/30 ---- lib/readline/misc.c - _rl_revert_all_lines: set entry->data to 0 after assigning it to rl_undo_list to avoid pointer aliasing problems that would result in entry->line being freed by an undo. The subsequent free would be a double free. Report and fix from Jared Yanovich subst.c - command_substitute: other shells do not appear to inherit the -v option when reading and executing command substitutions. Reported by Ondrej Oprala 7/1 --- config-top.h - CHECKHASH_DEFAULT: new define that supplies the default value for check_hashed_filenames (`checkhash' shopt option); still 0 by default findcmd.c - check_hashed_filenames: initialize using CHECKHASH_DEFAULT lib/readline/histexpand.c - history_expand: double quotes can inhibit recognition of the history comment character if history_quotes_inhibit_expansion is non-zero lib/readline/doc/{history.3,hstech.texi} - history_quotes_inhibit_expansion: expand definition to note that it inhibits scanning for the history comment character as well; correct typo to make it clear that it only works on double-quoted strings lib/sh/zgetline.c - add new fourth argument: DELIM, allows delimiter to be something other than newline (if DELIM != '\n', UNBUFFERED_READ should be non-zero) - UNBUFFERED_READ is now fifth argument - check character against DELIM rather than strictly newline externs.h - zgetline: change function prototype for extern declaration builtins/mapfile.def - mapfile: change calling sequence for zgetline calls - mapfile_builtin: new -d option: DELIM, like in read builtin - mapfile_builtin: pass `delim' to mapfile() as new argument; default to '\n' unless -d option supplied - mapfile: take new DELIM argument, pass to zgetline - mapfile: if DELIM != '\n', set unbuffered_read to 1 doc/{bash.1,bashref.texi} - mapfile: document new `-d DELIM' option 7/5 --- lib/readline/histfile.c - history_truncate_file: if there is an error writing the truncated history list back to the history file, use the same strategy as history_do_write: create a backup file, rename the history file to the backup file, and restore the original history file from the backup file name if the write or the close fails. Suggestion from Chen Gang to bug-readline execute_cmd.c - evalnest, evalnest_max: new variables establishing maximum number of recursive `eval' calls; current max is 4096 - execute_builtin: unwind-protect value of evalnest around calls to eval builtin. Suggested by Oliver Morais - {initialize_subshell,execute_subshell_builtin_or_function}: reset evalnest to 0 in a subshell builtins/setattr.def - show_name_attributes: show a variable's attributes even if it's invisible (don't show any value since it has none). This means that declare -p var will display VAR's attributes even when var marked as invisible. Feature request from Peggy Russell - show_var_attributes: don't print assignment if array or assoc attribute is set but variable marked as invisible tests/array.right - special note: changed all declare -a output tests because the shell will no longer print out values for invisible array variables. This is a change, but one for correctness: declare -a foo='()' and declare -a foo are not equivalent 7/22 ---- subst.c - parameter_brace_expand: after calling parameter_brace_expand_indir, turn off the W_ARRAYIND flag in the word it returns, because there is no way for it to return the index that should be used, and the rest of the function assumes that IND is valid if W_ARRAYIND is set. Fixes bug reported by Corentin Peuvrel 8/2 --- parse.y - read_token_word: if we read a character that will end a command substitution, don't skip over quoted newlines when we read an additional character to figure out whether it's a two-character token. This lets the higher layers deal with quoted newlines after the command substitution. Fixes bug reported by EmanueL Czirai 8/11 ---- execute_cmd.c - execute_pipeline: check whether lastpipe_jid corresponds to a valid job before calling append_process, for the same reason as fix from 6/19. Fixes bug reported by 8/12 ---- lib/sh/unicode.c - stub_charset: use strncpy instead of strcpy because we are copying into a local fixed-length buffer. Fixes vulnerability reported by execute_cmd.c - execute_pipeline: if we don't call append_process, call wait_for_single_pid to get the status of `lastpid', since that will check the status of already-reaped processes. Fixes spurious error message about non-existent process from fix of 8/11 8/15 ---- jobs.c - running_in_background: new variable, keeps track of whether or not we are running in the background (not perfect yet) - initialize_job_control: even if we are not turning on job control, get the terminal pgrp so we can use it later - {set_job_control,initialize_job_control}: set running_in_background to 1 if terminal pgrp != shell pgrp - {stop_pipeline,make_child,wait_for}: if we are running in the background, don't mess with the terminal's process group; assume that the parent shell will do that. Fixes bug reported by Greg Wooledge shell.c - shell_reinitialize: reset running_in_background back to 0 8/24 ---- execute_cmd.c - {execute_connection,execute_command_internal}: make sure that asynchronous commands always set $? to 0 and are not affected by the command's exit status being inverted using `!'. Fixes bug reported by Vincent Lefevre lib/readline/display.c - rl_message: call vsnprintf with full msg_bufsiz, since it counts one fewer than the buffer length passed as an argument. Bug report and fix from Dylan Cali 8/26 ---- builtins/evalstring.c - evalstring: if CURRENT_TOKEN == yacc_EOF, reset it to newline. This is instead of calling reset_parser(); that might still be needed. Fixes bug with eval and a subsequent statement ending with EOF reported by pcomplete.c - filter_stringlist: when extglob is on, a leading ! in the filter pattern should be left alone when it introduces a !(pat) pattern; otherwise it messes up the pattern. Fixes bug reported by David Korn 8/27 ---- doc/{bash.1,bashref.texi} - clarify the behavior of bash when given the -c option, since $0 is technically not a positional parameter. Bug reported by Stephane Chazelas 8/28 ---- lib/readline/history.c - add_history: use history_max_entries (if history is stifled) or DEFAULT_HISTORY_INITIAL_SIZE if not (new define, defaults to 502) to size the initial allocation of the history array. Assumption is that this will reduce the number of allocations 8/29 ---- execute_command.c: - sourcenest, sourcenest_max: new variables used to track level of sourced files and (maybe) one day catch infinite source recursion - execute_builtin: if current source level exceeds sourcenest_max, trigger an error and jump back to the top level - {initialize_subshell,execute_subshell_builtin_or_function}: reset sourcenest to 0 in a subshell 9/2 --- variables.c - bind_variable: if a nameref expands to an array reference, make sure that assign_array_element gets called (maybe even recursively) instead of bind_variable_internal, so invalid variable names (like arr[0]) don't get created. Fixes bug reported by 9/3 --- execute_cmd.c - evalnest_max,sourcenest_max: initialize from EVALNEST_MAX and SOURCENEST_MAX, respectively. Feature suggested by config-top.h - define EVALNEST_MAX and SOURCENEST_MAX to 0 9/6 --- bashline.c - find_cmd_start: fix to (crudely) deal with >| token; even though skip_to_delim finds `|' as a delimiter, we call it again and use what the second call finds. Fixes bug reported by Dan Jacobson findcmd.c - find_in_path_element: if in posix mode, do not expand a literal tilde in a $PATH element doc/bashref.texi - add change to tilde expansion in $PATH elements to posix mode description builtins/common.h - ISHELP: new define for builtins that do their own option parsing and don't use internal_getopt(); checks whether argument is --help - CHECK_HELPOPT: convenience define to help builtins that do their own option parsing to check for --help with one line of code - CASE_HELPOPT: convenience define to help builtins that use internal_getopt() check for --help with one line of code builtins/help.def - builtin_help: new function, prints out --help output for current builtin builtins/{kill,let,pushd}.def - add CHECK_HELPOPT to builtins that use ISOPTION; call builtin_help and return EX_USAGE (kill/let/pushd/popd/dirs) builtins/{caller,fg_bg}.def - use CHECK_HELPOPT to recognize --help, since these builtins perform checks that can cause them to return before calling no_options (caller/fg/bg) builtins/{exit,return}.def - use CHECK_HELPOPT to recognize --help before calling get_exitstat() (return/exit/logout) builtins/{break,shift}.def - use CHECK_HELPOPT to recognize --help before any other checks (break/continue/shift) builtins/bashgetopt.h - GETOPT_EOF: convenience define - GETOPT_HELP: new define, to indicate internal_getopt saw --help builtins/bashgetopt.c - internal_getopt: return GETOPT_HELP for --help builtins/common.c - no_options: recognize --help, call builtin_help and return 2 (builtin/eval/source/./times) builtins/command.def - use CASE_HELPOPT() to handle --help after calling internal_getopt() (command) builtins/{colon,echo,test}.def - do not recognize --help (:/true/false/echo/test) 9/8 --- sig.c - termsig_sighandler: if readline is active now, set the bashline event hook. Old code just set it for interactive shells. Part of fix for bug reported by bashline.c - bash_event_hook: call rl_cleanup_after_signal if terminating_signal is non-zero, since check_signals_and_traps() will cause the shell to exit if it is and we want to clean up the readline state first. Rest of fix for bug reported by 9/9 --- jobs.c - waitchld: when running the wait builtin in posix mode, with a trap set on SIGCHLD, use queue_sigchld_trap instead of trap_handler (SIGCHLD), otherwise you will lose SIGCHLDs when children_exited > 1. Fixes bug reported by builtins/read.def - read_builtin: if we are changing the tty settings, call initialize_terminating_signals so we have a chance to catch all terminating signals and potentially clean up the terminal before exiting - read_builtin: tty_modified: new variable, set to 1 if we change the terminal attributes and have to call ttyrestore() to restore them - if one of the `reads' returns -1/EINTR due to a terminating signal, and we have modified the terminal, call ttyrestore before calling CHECK_TERMSIG - ttyrestore: set tty_modified to 0 when called 9/10 ---- builtins/read.def - termsave: now global to file so other functions can use it - read_tty_cleanup: if tty_modified is non-zero, call ttycleanup to restore old terminal settings and turn off tty_modified sig.c - termsig_handler: call read_tty_cleanup if currently executing read builtin; it does the right thing. Final piece of fix for bug reported by Jan Rome 9/11 ---- general.c - printable_filename: general function to return a printable representation of a string (presumed to be a filename) general.h - extern declaration for printable_filename execute_cmd.c - execute_disk_command: use printable_filename builtins/{bind,cd,enable,hash,source}.def - use printable_filename as appropriate when printing error messages. From a suggestion by Vincent Lefevre builtins/bind.def - use CASE_HELPOPT() to handle --help after calling internal_getopt() (bind) 9/12 ---- builtins/common.h - SEVAL_FUNCDEF: new flag for parse_and_execute; it means that we only accept a single function definition command, as when we are importing functions from the environment - SEVAL_ONECMD: new flag for parse_and_execute; for future use builtins/evalstring.c - parse_and_execute: if the SEVAL_FUNCDEF flag is set, disallow anything but a function definition command variables.c - initialize_shell_variables: don't allow functions with invalid names to be imported from the environment, even though we still allow them to be defined - initialize_shell_variables: when importing function definitions from the environment, call parse_and_execute with the SEVAL_FUNCDEF flag to force the command to be just a function definition subst.c - param_expand: when expanding a $name variable expansion, make sure that the variable is visible and set before following the nameref chain - param_expand: when expanding a $name variable expansion and following the nameref chain, make sure the resulting variable is visible and set before using it 9/13 ---- variables.c - initialize_shell_variables: when importing function definitions from environment, use SEVAL_ONECMD flag for parse_and_execute. Part of CVE-2014-6271 builtins/evalstring.c - parse_and_execute: if SEVAL_ONECMD flag set, return immediately after calling execute_command_internal. Final piece for fix for bug reported by Stephane Chazelas . Part of CVE-2014-6271 9/24 ---- parse.y - reset_parser: reset eol_ungetc_lookahead to 0 here, since we don't want shell_getc returning it on the next call. Fixes problem reported by Tavis Ormandy and Michal Zalewski . Potentially part of CVE-2014-6271; fix for CVE-2014-7169 9/25 ---- parse.y - push_heredoc: new function, pushes a here-doc redirection onto redir_stack handling overflow of redir_stack. Exits on overflow. Original fix from Florian Weimer . Fix for CVE-2014-7186 - change straight assignments to redir_stack to call push_redir - add one to size of word_lineno stack to avoid off-by-one error below in read_token_word. Overflow just results in line numbers being wrong. Fix for CVE-2014-7187 9/27 ---- {execute_cmd,trap}.c - changes to make minimal-config builds work again, mostly missing #ifdefs for optional features builtins/common.c - builtin_help: dummy version to be included if HELP_BUILTIN not defined, for minimal-config builds variables.c - initialize_shell_variables: incorporated patches from Florian Weimer to change the strings bash looks for when importing shell functions from the environment. It adds a prefix (BASH_FUNC_) and a suffix (%%) to the name to mark it as having been created by bash as an exported function. Fix for remote attacks part of CVE-2014-6271 and CVE-2014-7169 - mk_env_string: takes new argument, indicating whether we are constructing a function - mk_env_string: encodes function names as described above, so initialize_shell_variables can find them 9/28 ---- copy_cmd.c - copy_redirects: before calling savestring on here_doc_eof, make sure it's not NULL (it could have been the result of a here document delimited by EOF or EOS). Fixes bug reported by Michal Zalewski . Fix for CVE-2014-6277 make_cmd.c - make_redirection: initialize here_doc_eof member to NULL. Rest of fix for CVE-2014-6277 9/29 ---- parse.y - current_input_line_state: return a sh_input_line_state_t containing the current shell_input_line and its index and size variables shell.h - current_input_line_state: extern declaration builtins/evalstring.c - parse_and_execute: call reset_parser() before returning if SEVAL_ONECMD set. Fixes bug reported by Michal Zalewski and designated CVE-2014-6278 - parse_and_execute: if we parse a function definition when SEVAL_FUNCDEF is set, but don't consume the entire passed string, throw an error, reset the parser, and return. Part of fix for CVE-2014-6278 - parse_and_execute: if parsing the shell function definition when SEVAL_FUNCDEF is set transforms the function name (e.g., if it begins with a newline or begins or ends with whitespace), throw an error, reset the parser, and return. Fixes bug reported by Eric Kobrin 10/2 ---- jobs.c - bgp_prune: don't do anything if bgpids.npid == 0 or bgpids.list == NULL. This can happen if something gets run before the job control framework is initialized. Bug report from 10/3 ---- parse.y - xparse_dolparen: don't set token_to_read to newline after calling parse_string() and cleaning up when the shell is not interactive. This makes the parser think it's ok to read new commands even if it's not in a state where that should be possible. Underlying fix for bug reported by Michal Zalewski and designated CVE-6278 - parser_remaining_input: new function, returns the portion of shell_input_line that hasn't yet been read - current_input_line_state: removed shell.h - parser_remaining_input: extern declaration - current_input_line_state: removed builtins/evalstring.c - parse_and_execute: change code that checks whether parse_command has consumed the entire passed string when SEVAL_FUNCDEF is used to use parser_remaining_input instead of messing around with (new) current_input_line_state. Part of fix for CVE-2014-6278 variables.c - initialize_shell_variables: if we don't call parse_and_execute, free the temporary string, since parse_and_execute won't. Report and fix from Eric Kobrin 10/4 ---- print_cmd.c - print_function_def: when in posix mode, print shell function definitions as posix specifies them, without the leading `function' keyword general.c - exportable_function_name: return 1 if the passed string can be added to the environment as an exported function name. Currently prohibits function names containing `/' and `=' from being exported general.h - exportable_function_name: extern declaration builtins/setattr.def - set_or_show_attributes: if exporting a function with export -f, call exportable_function_name to determine whether the function should be exported; don't export function if it returns 0 10/7 ---- builtins/setattr.def - set_or_show_attributes: don't show identifiers that are invisible and imported from the environment, since that combination of attributes means that the imported variable is not a valid shell identifier. Report from Stephane Chazelas 10/8 ---- shell.c - shell_initialize: set new variable should_be_restricted, which says whether or not the shell will be a restricted one based on the shell name; use in calls to initialize_shell_variables (to inhibit importing shell functions) and initialize_shell_options (to inhibit parsing $SHELLOPTS) and initialize_bashopts (to inhibit parsing $BASHOPTS). Report from 10/12 ----- execute_cmd.c - execute_function: unwind-protect loop_level, set loop_level to 0 when entering a function so break and continue in functions don't break loops running outside of the function. Fix picked up from dash via Herbert Xu 10/13 ----- doc/Makefile.in - bashref.pdf: create using texi2dvi --pdf rather than postprocessing the dvi file, so we have PDF bookmarks and links. Fix from Siep Kroonenberg 10/14 ----- subst.h - Q_ARITH: new quoting flag. Semantics are per Posix's spec for arithmetic expansion: act as if string is quoted, but don't treat double quotes specially (in this case, they will be removed by quote removal) - Q_ARRAYSUB: new quoting flag, indicates we are expanding an indexed array subscript subst.c - expand_arith_string: if we are not expanding the string, but we saw a quote with Q_ARITH specified as one of quoting flags, perform quote removal even if Q_DOUBLE_QUOTES is specified - param_expand: change calls to expand_arith_string for $[ and $(( cases to specify Q_ARITH. Now $(( "$x" )) and $(( "x" )) work if x has a value that evaluates to a valid number, as Posix specifies - expand_word_internal: add test for quoted&Q_ARITH to the tilde case, so we continue to perform tilde expansion in arithmetic contexts - expand_word_internal: if quoted&Q_ARITH, continue processing when we see a `"', acting as if the double quote isn't present (already Q_DOUBLE_QUOTED) arrayfunc.c - array_expand_index: pass Q_DOUBLE_QUOTED|Q_ARITH|Q_ARRAYSUB as quoted argument in call to expand_arith_string. This inhibits word splitting (Q_DOUBLE_QUOTED) while discarding double quotes (Q_ARITH), identical to the quote flags passed while expanding $(( )) and $[ ]. Q_ARRAYSUB reserved for future use. Fixes problem reported by Stephane Chazelas 10/16 ----- subst.c - parameter_brace_expand_word: if the PF_ASSIGNRHS flag is set and we are expanding what looks like an array subscripted with @ or *, make sure the variable we're expanding is actually an array before we add Q_DOUBLE_QUOTES to the flags. If we don't, things like scalar[@] will remain quoted. Fixes ubuntu bug 1381567 https://bugs.launchpad.net/ubuntu/+source/bash/+bug/1381567 10/17 ----- {jobs,nojobs}.c - get_original_tty_job_signals: get original signal dispostions for SIGTSTP, SIGTTIN, SIGTTOU before we start manipulating them in make_child - default_tty_job_signals: make sure we set SIGTSTP, SIGTTIN, or SIGTTOU to SIG_IGN if they were ignored at shell startup instead of unconditionally setting them to SIG_DFL. Fixes bug reported by idallen@idallen.ca jobs.h - get_original_tty_job_signals: extern declaration trap.c - initialize_traps: add call to get_original_tty_job_signals 10/22 ----- subst.c - expand_string_for_rhs: when expanding in this context (rhs of a word expansion or pattern removal), we don't perform word splitting, so we don't want to split $* if IFS is empty. Fixes bug reported by Stephane Chazelas 10/23 ----- subst.c - param_expand: when expanding $* in a pattern context where the expansion is quoted (Q_PATQUOTE), don't quote the expansion -- the outer quotes don't make the characters in the expansion of $* special. Posix interp 221. Reported by Stephane Chazelas 10/28 ----- lib/readline/bind.c - enable-bracketed-paste: new bindable variable, enables support for a terminal's `bracketed paste mode'. Code contributed by Daniel Colascione doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - enable-bracketed-paste: add description lib/readline/{readline.c,rlprivate.h} - _rl_enable_bracketed_paste: declarations - #defines for use by bracketed paste implementation lib/readline/rltty.c - rl_prep_terminal: send BRACK_PASTE_INIT string to terminal if we are supposed to enable bracketed-paste mode; change terminal_prepped to indicate we sent that string and need to clean up - rl_deprep_terminal: if terminal_prepped indicates we sent the bracketed-paste mode init string to the terminal, send the cleanup string before restoring the terminal modes lib/readline/kill.c - rl_bracketed_paste_begin: function to read bracketed paste until BRACK_PASTE_SUFF; discard the suffix, and insert the rest of the paste as a single (undoable) object. Bound to BRACK_PASTE_PREF lib/readline/funmap.c - bracketed-paste-begin: new bindable command, executes rl_bracketed_paste_begin lib/readline/readline.c - bind_bracketed_paste_prefix: new function, sets up to recognize the bracketed paste prefix sequence (BRACK_PASTE_PREF) in emacs keymap and vi insertion keymap - readline_initialize_everything: call bind_bracketed_paste_prefix 11/1 ---- builtins/ulimit.def - RLIMIT_POSIXLOCKS: now synonym for RLIMIT_LOCKS - -k: new option: RLIMIT_KQUEUES, max kqueues allocated for this process - -P: new option: RLIMIT_NPTS, max number of pseudoterminals available to this process doc/{bash.1,bashref.texi} - document `ulimit -k' option - document `ulimit -P' option parse.y - `timespec list_terminator' production: if the list terminator is `;' set last_read_token to `;' to allow things like `time ; command' to time null commands and not throw a syntax error. Patch from Piotr Grzybowski - `BANG list_terminator' production: do the same thing variables.c - sv_optind: use find_shell_variable and get_variable_value so we can have the variable's context in the case we need to do something when we are restoring a previous variable context's value builtins/getopt.h - sh_getopt_state_t: struct to save sh_getopt's internal state so we can restore it around function calls in the event that we have a local copy of OPTIND builtins/getopt.[ch] - sh_getopt_{save,restore}_istate: new functions to save and restore getopt's internal state - sh_getopt_{alloc,dispose}_istate: new functions to allocate and deallocate sh_getopt_istate_t objects execute_cmd.c - maybe_restore_getopt_state: restore sh_getopt state after executing function body iff the funtion declared a local copy of OPTIND - execute_function: save sh_getopt state before executing function body - execute_function: note in getopt_state->flags whether or not the function declared a local copy of OPTIND; used by maybe_restore_getopt_state - execute_function: maybe restore sh_getopt state before returning via call to maybe_restore_getopt_state. Fixes bugs with getopts and state between calls reported in 2011 by Bernd Eggink and in 2014 by Oyvind Hvidsten configure.ac - enable-function-import: new option, controls whether function imports are included. Enabled by default. Patch from David Galos config.h.in - FUNCTION_IMPORT: define controlled by enable-function-import above variables.c - initialize_shell_variables: include code to import function definitions from the environment if FUNCTION_IMPORT is defined doc/bashref.texi - --enable_function-import: document new configuration option 11/5 ---- lib/readline/history.c - history_lines_read_from_file: new variable, set by read_history and read_history_range to the actual number of lines read from the history file. The value is valid immediately after a call to one of those functions - history_lines_written_to_file: new variable, set by write_history, history_do_write, and history_truncate_file to the actual number of lines written to the history file. The value is valid immediately after a call to one of those functions variables.c - sv_histsize: set history_lines_in_file after history_truncate_file() only if hmax < history_lines_in_file (lines we've already read); a cosmetic change only bashhist.c - load_history: set history_lines_in_file after read_history() from history_lines_read_from_file, since read_history reads all of the lines from the history file even if it's more than $HISTSIZE - maybe_save_shell_history: after calling write_history(), set history_lines_in_file to history_lines_written_to_file, since we can assume that we read everyhing we just wrote builtins/history.def - history_builtin: after calling read_history (history -r), set the new value of history_lines_in_file, for the same reason as above - history_builtin: after calling read_history_range (history -n), set history_lines_in_file from history_lines_read_from_file 11/6 ---- lib/readline/histfile.c - history_truncate_file: since we move the old file to a backup copy before truncating, make sure the new file is owned by the same uid as the old - history_do_write: use chown in the same way as history_truncate_file 11/12 ----- lib/readline/display.c - expand_prompt: takes a new `flags' argument; only one flag defined so far: PMT_MULTILINE - expand_prompt: changed all callers to add new flags argument - rl_expand_prompt, redraw_prompt: make sure to set PMT_MULTILINE in FLAGS argument to expand_prompt if expanding parts of a prompt containing embedded newlines - expand_prompt: only add mode char to last line of a prompt with embedded newlines, so mode indicator doesn't get lost and gets updated properly. Fixes problem reported by Renlin Li 11/13 ----- lib/readline/display.c - prompt_modestr: changed prompt_modechar to return a string denoting the editing mode; default strings for emacs and both vi modes are #defines in this file. prompt_modestr takes an argument in which it returns the length of the mode string - expand_prompt: if expanding mode strings in the prompt, get the mode string to use and add it at the beginning of the prompt string, before expanding it. This will allow future work allowing the mode string to contain invisible characters 11/15 ----- lib/readline/rlprivate.h - _rl_{emacs,vi_cmd,vi_ins}_mode_str: extern declarations for variables to hold current values of user-settable mode strings; variables to hold lengths lib/readline/rlconf.h - defines for default values of the mode strings for each editing mode and keymap lib/readline/display.c - _rl_{emacs,vi_cmd,vi_ins}_mode_str: new variables to hold values of user-settable mode strings - _rl_{emacs,vi_cmd,vi_ins}_modestr_len: new variables to hold lengths of corresponding mode string variables - prompt_modestr: return appropriate user-settable mode string variables lib/readline/bind.c - {emacs,vi-ins,vi-cmd}-mode-string: new user-settable mode string variables - sv_{emacs,viins,vicmd}_modestr: variable handling functions for user- settable mode string variables. Non-null values are run through rl_translate_keyseq so users can include invisible character sequences in the mode strings; null values restore the default - _rl_get_string_variable_value: handle values for new user-settable mode string variables. Original code contributed by Dylan Cali lib/readline/doc/rluser.texi - {emacs,vi-ins,vi-cmd}-mode-string: document, including the fact that you can use \1 and \2 to bracket sequences of non-printing characters 11/16 ----- lib/readline/history.c - add_history: replace loop that copies history list down one item with call to memmove to take advantage of whatever efficiencies libc can offer. Won't be any slower than current loop lib/readline/display.c - rl_redraw_prompt_last_line: new function, calls redraw_prompt if the prompt contains multiple lines; calls rl_forced_update_display if not lib/readline/readline.h - rl_redraw_prompt_last_line: extern declaration, undocumented in texinfo manual until I get it a little more work bashline.c - bash_execute_unix_command: instead of unconditionally calling rl_forced_update_display, call rl_redraw_prompt_last_line if we cleared the last line before executing the command. This keeps commands that don't display any other output but just manipulate the contents of the line buffer from redisplaying the prompt lines before the last newline multiple times. Fixes bug reported by Jesper Nygards and Rob Foehl . This means that commands that display output will *only* display the final line of the prompt - bash_execute_unix_command: if the command returns 124, we redraw the line unconditionally, including all lines of the prompt 11/18 ----- builtins/mapfile.def - mapfile_builtin: don't allow a valid array reference through to mapfile(), since it will just create a shell variable with that name. Bug and fix from Eduardo A. Bustamante López 11/19 ----- lib/readline/complete.c - _rl_colored_completion_prefix: new variable, if non-zero, and color support is compiled in, and the $LS_COLORS variable exists and contains color definitions, display any common prefix of a set of completions in blue when displaying all the possible completions. Doesn't work with menu-complete, which inserts possible completions inline - colored_prefix_start(), colored_prefix_end(): new functions, used to bracket colored completion prefixes - fnprint: if prefix_bytes is non-zero, and _rl_colored_completion_prefix is > 0, display the first PREFIX_BYTES bytes of the word bracketed by calls to colored_prefix_start and colored_prefix_end - print_filename: if _rl_colored_completion_prefix is > 0, compute the length in bytes of the common prefix and pass that to fnprint lib/readline/readline.c - readline_initialize_everything: initialize the colors from $LS_COLORS if _rl_colored_completion_prefix is non-zero lib/readline/colors.c - _rl_print_prefix_color: new function, changes text color to that defined for the common prefix of a set of possible completions (currently cyan, same as directories); currently uses C_PREFIX, defined in colors.h as C_DIR lib/readline/colors.h - _rl_print_prefix_color: new extern declaration lib/readline/bind.c - colored-completion-prefix: new bindable variable, if set, common prefix of a set of possible completions is displayed in color. Feature requested by several, most recently by Richard Neill (in a slightly different form) and Duy Nguyen 11/20 ----- builtins/printf.def - printf_builtin: allow null (empty) format strings supplied with -v var to set `var' to the empty string. That is, printf -v var "" is now the same as var="". Change suggested by Mike Frysinger pathexp.h - FNMATCH_NOCASEGLOB: macro to decide whether or not to pass FNM_CASEFOLD flag to strmatch depending on whether glob_ignore_case is set; analogout to FNMATCH_IGNCASE pathexp.c - glob_name_is_acceptable: use FNMATCH_NOCASEMATCH to determine flags passed to strmatch; if nocaseglob is used to generate glob matches, it should be used to generate ignored matches 11/21 ----- pcomplete.c - filter_stringlist: the call to strmatch now honors the setting of nocasematch. Feature request from Ville Skytta back in 2010 doc/bash.1,lib/readline/doc/rluser.texi - complete -X: document that filtering the list of possible completions honors the nocasematch option when performing matching lib/glob/gmisc.c - include chartypes.h and strmatch.h for new defines - match_pattern_{wchar,char}: now take new third FLAGS argument, flags have same meanings as strmatch(); intent is to handle case insensitive comparisons under same conditions as strmatch - FOLD: imported case-folding define from sm_loop.c; wide and single- byte character versions - match_pattern_{wchar,char}: use FOLD when comparing characters to honor FNM_CASEFOLD if set in FLAGS argument externs.h - match_pattern_{wchar,char}: updated extern declarations subst.c - match_{upattern,wpattern}: update strmatch/wcsmatch calls to include FNMATCH_IGNCASE in flags argument - match_{upattern,wpattern}: update match_pattern_{char,wchar} calls to include FNMATCH_IGNCASE in flags argument (consistent with calls to strmatch). This makes pattern substitution word expansion honor nocasematch shell option. Feature requested by Davide Baldini - match_wpattern: make sure to fold case if necessary when doing simple matching doc/{bash.1,bashref.texi} - pattern substitution: updated description to include honoring setting of nocasematch when performing matching subst.c - expand_word_internal: optimize handling of "$@" idiom by calling list_rest_of_args() and quote_list() directly at the top of the function instead of going through normal code path - cached_quoted_dollar_at: WORD_LIST of quoted positional parameters, used by same code above that optimizes "$@"; if non-null we just return a copy of the list; if null, we save a copy of the list we create - invalidate_cached_quoted_dollar_at: convenience function to allow other parts of the shell (e.g., remember_args()) to destroy the cached list of quoted positional parameter when the positional parameters change builtins/common.c - remember_args: call invalidate_cached_quoted_dollar_at() builtins/shift.def - shift_builtin: call invalidate_cached_quoted_dollar_at after modifying dollar_vars[] builtins/source.def - maybe_pop_dollar_vars: call invalidate_cached_quoted_dollar_at just to be safe 11/23 ----- builtins/evalfile.c - _evalfile: return -1 if errno == ENOENT and the flags don't include FEVAL_ENOENTOK. If we print an error message we should return an error - force_execute_file: new function, reads and executes commands from a file but returns an error if file doesn't exist builtins/common.h - force_execute_file: new extern declaration shell.c - main: call start_debugger even if dollar_vars[1] == 0 if the shell isn't interactive (interactive_shell == 0) - start_debugger: call force_execute_file instead of maybe_execute_file; turn off debugging mode if it returns value < 0 11/24 ----- hashlib.h - DEFAULT_HASH_BUCKETS: doubled to 128, cost in memory use is small but changes traversal order when not sorting results 11/25 ----- doc/{bash.1,bashref.texi} - make it clearer, by breaking it out into a separate paragraph, that referencing an array without a subscript is equivalent to referencing it with subscript 0 - add text saying that referencing any variable using a valid subscript is OK 11/28 ----- arrayfunc.c - bind_array_variable, bind_assoc_variable: allow binding value to a readonly variable if the ASS_FORCE flag is set in the FLAGS argument subst.h - ASS_FORCE: new assignment flag; means to allow assignment even if variable is marked readonly builtins/declare.def - when assigning a value to an array or assoc variable using something like `declare -r foo=bar' where foo is an existing array variable, pass the ASS_FORCE to assign_array_var_from_string so the assignment is allowed. Fixes debian bug 765759 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=765759 builtins/setattr.def - var_attribute_string: new function to return (as argument) a char array with attribute flag values for a given variable; returns the length of the array lib/sh/shquote.c - sh_quote_reusable: function returning a version of its string argument that is quoted for reuse externs.h - sh_quote_reusable: extern declaration builtins/common.h - MAX_ATTRIBUTES: define used to size arrays for attribute flag characters - var_attribute_string: new extern function declaration subst.c - array_remove_pattern: fixed a bug where `var' instead of `v' was tested for invisible attribute - get_var_and_type: fill in a valid *VARP if returning VT_VARIABLE because callers may need to use it - parameter_brace_transform: family of functions to implement the new mksh-inspired ${param@spec} transformation word expansions. Some of the operators transform the (expanded) value of the parameter, the rest expand to information about the parameter itself (array_transform, parameter_list_transform, list_transform, string_transform, pos_params_assignment, array_var_assignment, string_var_assignment) - parameter_brace_expand: changes to parse the new `@' word expansion operator and call parameter_brace_transform appropriately - parameter_brace_expand: make sure we handle ${#@} as we have before even in the presence of the new `@' operator variables.c - push_temp_var: make sure to call bind_variable_internal with the ASS_FORCE flag so we override readonly variables created with something like `tempvar=foo declare -r foo'. - bind_variable_internal: honor ASS_FORCE flag to allow binding even if a variable is readonly execute_cmd.c - struct func_array_state: new state to save state of BASH_LINENO, BASH_SOURCE, and FUNCNAME during function execution so it can be restored on a jump to top level - restore_funcarray_state: new function to restore func_array_state - execute_function: fill in func_array_state variable, add unwind- protect to restore it on jump to top level, call explicitly at end of function if subshell != 0 (may not be necessary, but safe for now). Fixes bug with local assignments to FUNCNAME reported by Arfrever Frehtes Taifersar Arahesis 11/29 ----- arrayfunc.c - assign_compound_array_list: turn off ASS_APPEND flag when processing each individual assignment inside the parens in var+=(...). The outer += should not affect assignments to existing subscripts; those should be treated like usual assignments unless += supplied inside the parens. Bug report from Maarten Billemont , fix from Eduardo A. Bustamante López config.h.in - HAVE_PSELECT: define if pselect(2) available configure.ac - check for pselect(2), define HAVE_PSELECT if found lib/readline/input.c - rl_getc: use pselect(2) to wait for input ready on readline's input fd or for a signal to arrive, will handle SIGWINCH (which does not interrupt read(2)) and thus allow resize to happen without having to wait to read more input. Only works if pselect available and returns -1/EINTR on a signal even if the signal was installed with SA_RESTART. From a suggestion from Egmont Koblinger 12/3 ---- variables.c - flush_temporary_env: new function, disposes all temp variables in temporary_env hash table - bind_variable: only try to update a temporary variable's value in the temporary env if the value argument is not null. Fixes bug reported by variables.h - flush_temporary_env: new extern declaration subst.c - command_substitute: if running command substitution as part of expanding a redirection (expanding_redir == 1), flush any temporary environment we've inherited as part of this command, since we are not supposed to have access to the temporary environment. Since expanding_redir only controls access to the temporary environment for variable lookup and binding, we can turn it off in the subshell 12/4 ---- builtins/printf.def - printstr: make sure a missing precision specifier after a `.' is treated as 0, as printf(3) specifies. Fixes ubuntu bash bug 1399087 (https://bugs.launchpad.net/ubuntu/+source/bash/+bug/1399087) 12/5 ---- subst.c - skip_to_delim: if scanning past process substitution (skipcmd == 1, noprocsub == 0), use extract_delimited_string instead of extract_process_subst, which was changed a while back (bash-4.3.23) to use xparse_dolparen. xparse_dolparen complains if the command or process substitution is unterminated, since it runs the parser, which is not what we want here. Command substitution does the same thing. Fixes bug reported by Daniel Kahn Gillmor as Debian bash bug 771968 (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=771968) 12/6 ---- subst.c - command_substitute: short-circuit without forking on a command string that consists entirely of s and newlines jobs.c - make_child: changes to allow interrupts through if fork fails and we are sleeping for `forksleep' seconds - waitchld: make things a little more resilient if CHILD ends up NULL 12/12 ----- lib/readline/complete.c - rl_display_match_list: when calculating common prefix to display in color, make sure we correctly handle a common prefix with a trailing `/' as we do when checking whether or not to add an ellipis. printable_part() doesn't return the whole pathname if it ends in a slash, to avoid printing null strings, so we have to make sure we have the entire prefix lib/readline/complete.c - _rl_complete_display_matches_interrupt: new variable, set to 1 by _rl_complete_sigcleanup to let rl_display_match_list know it has freed the match list - display_matches: check for signals during the printing loops with RL_SIG_RECEIVED(), return immediately if there is a pending signal (might not want to do this if it's SIGWINCH -- CHECK) - rl_complete_internal: if _rl_complete_display_matches_interrupt set after calling display_matches, just null out `matches' since it's already been freed and call any application-set signal hook 12/14 ----- parse.y - time_command_acceptable: if the token before a newline is `|', return 0, since it's not really valid to time inside a pipeline. Only handles a single newline but allows things like echo a | time cat to invoke /usr/bin/time, which is probably enough to catch the stray carriage return. Fixes bug reported by Andre Majorel builtins/declare.def - declare_internal: don't try to perform compound assignments unless the WORD_DESC has flags including W_COMPASSIGN (maybe should check W_ASSIGNMENT as well), avoiding unexpected evaluation if a word is of the form (word) and is assigned to an array variable like so: declare -x var=$value. Bug reported by Stephane Chazelas . Will eventually be contingent on compatibility level > 43, but not there yet. TENTATIVE 12/15 ----- lib/sh/Makefile.in - add missing dependencies for shmatch.o. Pointed out by Sergey Mikhailov 12/16 ----- {execute_cmd,subst}.c - W_ASSIGNINT: remove, not used any more execute_cmd.c - fix_assignment_words: don't look for `-i' option and set W_ASSIGNINT flag any more; doing things a different way - shell_expand_word_list: instead of using W_ASSIGNINT flag, since it doesn't take into account all options that can transform values on assignment (-l/-u/-c can also), go through option arguments looking for options that need special handling and add them to the `opts' array for make_internal_declare to use. Fixes bug with constructs like `declare -al foo=(UPONE UPTWO UPTHREE)' not being lowercased on assignment reported by Linda Walsh 12/18 ----- lib/readline/readline.c - rl_internal_char: when we read EOF on a non-empty line, check for signals and invoke any readline signal handling and any application- installed signal hook 12/20 ----- lib/readline/readline.c - rl_internal_char: if we read EOF on a non-empty line, set c to _rl_eof_char the first time through. If we read EOF the next time, return EOF from readline(). If callbacks are defined, this returns EOF immediately, since lastc isn't available. Fix for problem most recently identified by Jiri Kukacka , it has come up in the past 12/21 ----- builtins/declare.def - declare_internal: keep track of whether or not an assignment statement argument to declare is an array subscript assignment; need to differentiate assignments from straight declarations (declare a[4]) which are accepted for backwards compatibility - assignment statements like declare a[2]=foo are now treated as straight subscript assignment statements if a already exists as an array variable - declare foo='(1 2 3)' is treated as an assignment to foo[0] if foo exists and is an array, just as it would be if it were an assignment statement and `declare' was not present. All this from a proposal by Stephane Chazelas 12/22 ----- builtins/read.def - read_tty_modified: function to tell the rest of the shell if the read builtin has modified the tty - read_builtin: make sure to initialize terminating signals before installing a SIGALRM signal handler in case we modify the tty as well as ask for a timeout; the subsequent call to initialize_terminating_signals would overwrite the read-builtin- local SIGALRM handler builtins/common.h - read_tty_modified: new extern declaration shell.c - exit_shell: if read_tty_modified() returns true, call read_tty_cleanup to undo the terminal modifications. Extension of previous fixes; fixes bug with read -s reported by Richard W. Marsden 12/23 ----- builtins/setattr.def - show_var_attributes: call print_array_assignment and print_assoc_assignment with a `not quoted' flag so the assignment statements are not surrounded by single quotes. Caused changes to a lot of test output 12/29 ----- lib/readline/complete.c - stat_char: Windows doesn't handle X_OK flag to access(2); use file extensions to determine whether or not a file is executable. Bug report and fix from Eli Zaretskii configure.ac - changed version to bash-4.4-devel lib/readline/readline.h - changed version to readline-6.4 12/30 ----- readline.h - struct readline_state: fix types of `ignorefunc' and `attemptfunc' members lib/sh/shquote.c,externs.h - sh_contains_shell_metas, sh_contains_quotes: now take `const char *' as parameter stringlib.c,externs.h - strcreplace: `text' argument now `const char *' pathexp.[ch] - quote_globbing_chars: `string' argument now `const char *', accompanying changes to function local variables pcomplete.c - preproc_filterpat: `text' argument now `const char *' - filter_stringlist: `text' argument now `const char *' 12/31 ----- builtins/evalstring.c - should_suppress_fork: new function, broke code that decides whether to turn on CMD_NO_FORK flag out of parse_and_execute into a separate function execute_cmd.c - execute_command_internal: AND_AND, OR_OR: call should_suppress_fork for the RHS of && and ||, make `make' invocations marginally more efficient flags.c - get_current_flags: returns a bitmap of all single-letter shell options with a non-zero value meaning option is enabled - set_current_flags: set current single-letter options from a passed bitmap, which has presumably been initialized with get_current_flags flags.h - {get,set}_current_flags: new extern declarations builtins/declare.def - declare_internal: if invoked as `local', interpret name argument of `-' to mean to save (and restore at shell return) single-letter shell options variables.c - push_func_var: if we encounter `-' as the name of a local variable, take the value and call set_current_flags on it to restore old flag settings builtins/set.def - get_current_options: returns a bitmap of all shell options settable with `set', with a non-zero value meaning option is enabled - set_current_options: set current shell options from a passed bitmap, which has presumably been initialized with get_current_options 1/1/2015 -------- lib/readline/parens.c - _rl_blink_matching_paren: initialize to 0 whether or not select(2) is available. Inconsistency reported by Mark Karpov builtins/declare.def - declare_internal: use get_current_options instead of get_current_flags, so `local -' can be generalized variables.c - push_func_var: use set_current_options instead of set_current_flags so `local -' can be generalized doc/{bash.1,bashref.texi} - document new `local -' feature; originally inspired by Robert Elz describing feature in NetBSD sh 1/3 --- lib/readline/{complete,funmap,kill,histfile,util} lib/readline/readline.h lib/tilde/tilde.c - fixes for filename- and evironment-related issues on Windows. Fixes from Eli Zaretskii - Windows can paste to the console from the clipboard like Cygwin - Windows uses $APPDATA as a pseudo-$HOME lib/readline/input.c - rl_getc: use _getch on Windows to avoid being overridden by a getch ncurses function - win32_isatty: replace Windows isatty(3) with a function that does additional checks, sinces Windows isatty returns non-zero for every character device. From Eli Zaretskii lib/readline/display.c - delete_chars: if compiled with ncurses on Windows, this code can work, so build it in if NCURSES_VERSION defined - open_some_spaces: ditto 1/6 --- doc/{bash.1,bashref.texi} - minor grammar and typo corrections from TonyMc 1/8 --- sig.c - throw_to_top_level: only attempt to run the interrupt trap if SIGINT is trapped -- minor optimization - throw_to_top_level: if we are running a trap, call run_trap_cleanup to turn off the sigmodes flags trap.c - _run_trap_internal: temporarily suppress pending SIGINTs while running one of the traps the shell handles internally (e.g., ERR). Fixes bug reported by Keith Thompson 1/10 ---- bashhist.c - bash_history_inhibit_expansion: use skip_to_delim with the history expansion character to see whether or not the instance of the history expansion character should be skipped because, for instance, it is in a command substitution. Fixes issue reported by Zigmund.Ozean@zig-home.localdomain 1/14 ---- lib/readline/isearch.c - _rl_isearch_dispatch: if we are in a multibyte locale, make sure to use _rl_find_prev_mbchar when trying to delete characters from the search string, instead of just chopping off the previous byte. Fixes bug reported by Kyrylo Shpytsya lib/readline/complete.c - rl_filename_completion_function: if we tilde-expand `dirname', make sure we use the expanded result instead of throwing it away for the quote-removed `users_dirname', which still has any leading tilde. Fixes bug reported by Dave Rutherford execute_cmd.c - select_query: if the read builtin doesn't return anything in REPLY, return NULL (failure). Fixes bug reported by Etherial Raine builtins/read.def - read_builtin: if REPLY has the readonly or noassign attribute, return failure 1/16 ---- redir.c - redir_open: if open() returns -1/EINTR, check traps as well as using QUIT to handle signals 1/18 ---- jobs.c - wait_for: make sure queue_sigchld is set to 1 before unblocking SIGCHLD if MUST_UNBLOCK_CHLD is defined 1/19 ---- pcomplete.h - new progcomp option: COPT_NOSORT, means to not sort list of possible completions pcomplete.c - pcomp_set_readline_variables: if COPT_NOSORT set in compspec flags, set rl_sort_completion_duplicates to 0 builtins/complete.def - _compopt: new option value: -o nosort, sets COPT_NOSORT in compspec option flags. Provides new -o nosort option for complete and compgen bashline.c - attempt_shell_completion: make sure rl_sort_completion_matches is set to 1 before completion is attempted; allow a compspec to turn it off lib/readline/kill.c - rl_vi_yank_pop: new function, vi-mode version of yank-pop. Original code from Ian Kelling lib/readline/funmap.c - `vi-yank-pop': bindable name mapped to rl_vi_yank_pop lib/readline/readline.h - rl_vi_yank_pop: extern declaration 1/21 ---- lib/sh/shquote.c - sh_backslash_quote: change to understand and handle multibyte characters, using is_basic, COPY_CHAR_P. See https://bugzilla.redhat.com/show_bug.cgi?id=1184320 for original report 1/23 ---- include/posixjmp.h - use setjmp_sigs instead of redefined setjmp, it fails on Cygwin. Report from Eric Blake builtins/wait.def,shell.c - use setjmp_sigs instead of setjmp call, since setjmp no longer redefined to something known include/chartypes.h, lib/readline/chardefs.h - make sure all ctype.h macros are called with unsigned char args, casting to make sure lib/sh/casemod.c - sh_modcase: don't assume that the upper and lower case versions of a character are the same width, so don't try to do the conversion in place: convert and copy the converted characters one or more at a time to the returned string - sh_modcase: since upper and lower case versions of same character may have different widths in some locales, don't shortcut and assume that we can do single-byte case modification (toupper) with a single-byte input character (e.g., `i'). Fix for problem reported by Stephane Chazelas 1/31 ---- lib/readline/histfile.c - history_truncate_file, history_do_write: if the first rename(2) of the history file to the backup file fails, set the backup file name to NULL to prevent any later attempts to restore the original. Report from Jonathan Hankins - history_do_write: don't attempt to back up non-regular files. Report from Jonathan Hankins 2/2 --- trap.c - run_interrupt_trap: now takes an argument indicating whether or not it was called just before a longjmp to top_level; change callers - run_interrupt_trap: if running a trap and we are being called before a longjmp to top_level, call run_trap_cleanup on running trap so a QUIT call while running a DEBUG trap, for instance, doesn't leave sigmodes[DEBUG_TRAP] including SIG_INPROGRESS. Fixes bug reported by Chih-Chieh Chou sig.c - throw_to_top_level: change call to run_interrupt_trap to pass arg 1 - termsig_handler: change call to run_interrupt_trap trap.h - run_interrupt_trap: update extern prototype declaration 2/4 --- lib/readline/histfile.c - histfile_backup, histfile_restore: new functions to back up and restore history file to `filename-', following (one level of) symlinks through so we don't rename the symlink itself. From a suggestion by Jonathan Hankins assoc.c - assoc_to_assign: if an index happens to be `@' or `*', make sure it's double-quoted. Bug report from Greg Wooledge 2/5 --- lib/readline/bind.c - rl_bind_key: if we are rebinding a key that is not bound to a function or a macro, construct a key sequence and pass it to rl_bind_keyseq. Make sure that '\' when it is used as a key is changed to "\\" so it can survive rl_translate_keyseq. Fixes https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=776743 2/8 --- include/posixjmp.h - _rl_longjmp, sh_longjmp: new defines, expand to siglongjmp or longjmp depending on autoconf results, used to avoid redefining longjmp, remove _OPENNT conditional code. Report from Eric Blake {bashjmp,quit}.h,{execute_cmd,expr,jobs,sig,trap}.c,builtins/{evalstring.c,return.def} - change calls to longjmp to use sh_longjmp lib/readline/util.c - change calls to longjmp to use _rl_longjmp 2/18 ---- variables.c - bind_int_variable: make sure `v' is non-null before making it visible 2/19 ---- arrayfunc.c - assign_array_var_from_word_list: after assignment, mark variable as no longer invisible - assign_array_var_from_string: after assignment, mark variable as no longer invisible builtins/declare.def - declare_internal: add warning if an attempt is made to use a quoted compound assignment as an argument to declare (declare -a foo='( 1 2 )'); backwards compatible with bash-4.3. Still a tentative change 2/20 ---- lib/glob/smatch.c - is_wcclass: check malloc() return value, return -1 if it fails Report from Tobias Stoeckmann lib/sh/shmatch.c - sh_regmatch: check malloc() return value, handle NULL value if it fails. Report from Tobias Stoeckmann 2/22 ---- lib/readline/doc/rltech.texi - rl_callback_handler_install: note that the handler function should free the line it receives, as with readline. Suggested by Ulf Magnusson - Readline Signal Handling: note that application needs to clean up readline's terminal state if it wants to handle a signal before the line handler restores it. Suggested by Ulf Magnusson 2/24 ---- lib/readline/histfile.c - history_do_write,history_truncate_file: if the write fails, make sure to set history_lines_written_to_file to 0 so we don't lose those history entries 2/27 ---- execute_cmd.c - execute_function: unwind-protect the value of line_number_for_err_trap, so a function that returns a non-zero value and triggers an ERR trap has the right value for LINENO. Any simple command in the function body can change it. Bug report from Paul Donohue 3/5 --- lib/sh/shquote.c - sh_backslash_quote: cast `c' to `unsigned char' before using as index into backslash_table. Report and fix from alexeyv.90@gmail.com arrayfunc.[ch] - valid_array_reference: added second `flags' argument for future use; currently unused parse.y,{redir,subst,variables,test}.c builtins/{declare,printf,read,set}.def - valid_array_reference: changed callers 3/7 --- lib/readline/signals.c - _rl_handle_signal: if readline gets a SIGTSTP or SIGTTIN, block SIGTTOU so a backgrounded readline can reset the terminal settings appropriately, otherwise the terminal state and readline's idea of it can get messed up. Original report from Dave Wysochanski and Dave Anderson trap.c - run_pending_traps: don't allow recursive SIGWINCH trap invocations; they're wasteful and trying to handle a stream of signals while a user resizes the terminal can lead to malloc errors trying to save and restore the_pipeline multiple times. Problem reported most recently by Martin M. S. Pederson (Fedora bash bug 1197768 https://bugzilla.redhat.com/show_bug.cgi?id=1197768) - reset_or_restore_signal_handlers: set pending_traps to 0 for all signals when forking a child. TENTATIVE fix jobs.c - {save,restore}_pipeline: block SIGCHLD while manipulating the_pipeline like cleanup_the_pipeline does. XXX - may not be necessary lib/readline/display.c - rl_redisplay: fixed case where multibyte character in the prompt (e.g., \u03bb) in the last screen column causes problems with line wrapping because the newline marker is put in the middle of the multibyte char. Most recently reported on Ubuntu by pathexp.c - quote_string_for_globbing: if a bracket expression doesn't have a closing right bracket, rescan it from just after the opening bracket without treating it as a bracket expression. Bug report from Stephane Chazelas 3/9 --- subst.c - process_substitute: keep the last pipeline created to run a process substitution in LAST_PROCSUB_CHILD; discard it when another one is created jobs.[ch] - discard_pipeline: now global, so process_substitute can call it jobs.c - last_procsub_child: new variable, set and modified only by the process substitution code - find_pipeline: if the requested pid is in LAST_PROCSUB_CHILD, return it; analogous to the way this code handles the_pipeline - waitchld: if we find a child via find_pipeline that is a part of the last process substitution child, mark that child appropriately. If the child is dead, add it to the bgp list. This and the previous change allow waiting for the most recent process substitution (the one corresponding to $!). Original report from Stephane Chazelas 3/10 ---- arrayfunc.c - unbind_array_element: make sure skipsubscript() doesn't try to parse embedded quotes in the subscript, since this has already been through one round of word expansion before calling the unset builtin. (This function is only called by the unset builtin). Fixes bug reported by 3/17 ---- jobs.c - waitchld: move code that adds the last process substitution pid to the bgpids array from waitchld (where it can call malloc in a signal handler context) to cleanup_dead_jobs 3/19 ---- subst.c - parameter_brace_expand: if expanding an array in a context where word splitting will not take place (W_ASSIGNRHS only for now), make sure to pass the double-quoted flag to chk_atstar so we don't get the unquoted $* behavior, which will cause unwanted word splitting. Fixes bug reported by isabella parakiss arrayfunc.c - unbind_array_element: don't assume that caller will check and pass only array variables; don't allow non-array variables to be unset using the `@' or `*' subscripts and don't allow any subscripts other than 0; return -2 in those cases - unbind_array_element: unset a non-array variable if passed a subscript of `0' (e.g., `unbind "scalar[0]"'). This is new functionality that parallels the ${scalar[0]} expansion builtins/set.def - unset_builtin: don't check whether a variable is an array if it's referenced using array syntax (scalar[0]); pass it to unbind_array_element anyway and check the result. This means that `unset scalar' and `unset scalar[0]' are equivalent 3/20 ---- builtins/set.def - set_builtin: make `-i' an invalid option, unconditionally for now. From a bug-bash discussion started by Peng Yu 3/21 ---- lib/readline/colors.c - _rl_print_color_indicator: handle dangling symlinks (C_ORPHAN) and the `target' specifier for links (dereference S_ISLNK if the target exists). Bug report and fix from andreas@stapelspeicher.org {arrayfunc,subst}.c - add code to save and restore this_command_name in places that set this_command_name to NULL to prevent error messages from arithmetic evaluation. Fixes bug reported by brian.carpenter@gmail.com builtins/mapfile.def - do_chop: remove a new DELIM argument from the line, instead of unconditionally removing newline - mapfile: if -d delim is supplied and we are breaking fields at DELIM, remove DELIM if the -t option is supplied by passing DELIM to do_chop(). Suggested by Geir Hauge doc/{bash.1,bashref.texi} - mapfile: document new -t option behavior, defaults are the same builtins/alias.def - print_alias: if the first character of an alias name is `-', add a `-- ' after `alias' to avoid option errors when trying to reuse the output. Bug report and fix on savannah from Pasha Bolokhov 3/22 ---- builtins/declare.def - declare_internal: if creating a variable does not return a valid variable in VAR, check for null variable before dereferencing it. This happens with nameref variables referencing variables set to the empty string. Fixes bug reported by Arthur200000 3/23 ---- doc/{bash.1,bashref.texi} - INVOCATION: add language to make it clear that non-interactive login shells execute ~/.bash_logout if they run the `exit' builtin. From a question from Christoph Anton Mitterer 4/6 --- subst.c - process_substitute: increment and decrement subshell_level around calls to parse_and_execute, as in command substitution, so $BASH_SUBSHELL gets new values in the subshell. Fixes inconsistency pointed out by Peng Yu doc/{bash.1,bashref.texi} - read: add clarifying text to note that the -N option does not split on $IFS, but does retain backslash as an escape character subject to -r. Report from isabella parakiss builtins/declare.def - declare_internal: unless it's the last statement in a {...} block, NEXT_VARIABLE must be protected by braces. Bug introduced in patch from 3/22, reported by Eduardo A. Bustamante López 4/7 --- parse.y - shell_getc: if bash is reading input from a string that ends with an unquoted backslash, add another backslash instead of a newline, since the backslash and newline will disappear in normal processing. Fixes bug with `bash -c 'eval \\; echo y' ' skipping the eval command and setting incorrect exit status, and `bash -ic 'eval \\; echo y' ' seeing EOF on empty line and exiting before the echo. Keep track of backslash state with last_was_backslash; set in char reading loop. Fixes bug reported by Eduardo A. Bustamante López 4/8 --- shell.c - main: check for $INSIDE_EMACS as well as $EMACS when deciding whether or not bash is being run in a GNU emacs shell window. Don't bother checking whether $TERM is unset because variable initialization sets it to `dumb' if unset. Fixes from Paul Eggert parse.y - push_token: new function, rotates current_token through last_read_token, saving old values like yylex() does, sets current_token to argument evalfile.c - _evalfile: if current_token is yacc_EOF, push a newline as the value of current_token instead, using push_token(). yacc_EOF can result if sourced file doesn't end with a newline but does end with an unescaped backslash. Fixes bug reported by Pedro Gimeno lib/readline/histfile.c - use PARAMS(x) instead of __P(x). Bug reported by Evan Gates 4/9 --- execute_cmd.c - execute_simple_command: if we haven't already forked, and we have created FIFOs, turn off the no-fork optimization in the simple command's flags. Originally reported to savannah by Kuang-che Wu as https://savannah.gnu.org/support/index.php?108732 4/10 ---- trap.c - trap_if_untrapped: new function, sets a trap on a signal if that signal is not already trapped - maybe_set_debug_trap: new function, sets DEBUG trap to command string argument only if it's not already trapped trap.h - maybe_set_debug_trap: new extern declaration builtins/source.def - source_builtin: if the -T option isn't set, use maybe_set_debug_trap in the unwind-protect instead of unconditionally restoring the old trap string. That means if the sourced file changes the DEBUG trap, that new trap will persist instead of being undone when the old value is restored. Fixes bug reported by Scott Bronson builtins/evalstring.c - parse_string: before we jump to top_level, if we are at the `bottom' of the parse_and_execute stack (parse_and_execute_level == 0), call top_level_cleanup() before jumping so the right cleanup happens (e.g., cleaning up redirections and other state). Fixes bug reported by Sam Liddicott 4/15 ---- jobs.c - delete_job: only add J_ASYNC (and not J_FOREGROUND) jobs to the bgpids array, since Posix says shells only need to save the status of async jobs. Report and fix from John Fremlin 4/17 ---- pcomplib.c - COMPLETE_HASH_BUCKETS: increase to 256 since bash-completion is over 300 completions by now variables.c - VARIABLES_HASH_BUCKETS: constant used to size shell variables hash table instead of relying on hash library default. Used in calls to hash_create in create_variable_tables(), push_temp_var(), and push_func_var(). Currently set to 1024 because why not? From a discussion starting at https://lists.gnu.org/archive/html/bug-bash/2015-04/msg00089.html - initialize_shell_variables: if FUNCTION_IMPORT is not defined, the nested conditional compilation causes syntax errors. Bug and fix from Eric Blake configure.ac - function-import: fix typo, assign to correct variable. Found as the result of a report by Eric Blake shell.c - open_shell_script: check whether or not file argument is a directory before opening; Posix says that implementations may open a directory with O_RDONLY. Bug report from Eduardo A. Bustamante López 4/19 ---- ulimit.def - fix typo in limits[] declaration of `k' 4/20 ---- configure.ac - add template definitions set by AC_USE_SYSTEM_EXTENSIONS from a report from Andreas Schwab variables.c - FUNCTIONS_HASH_BUCKETS: constant used to size the hash tables used for shell functions and the text of shell function definitions. Initially set to 512, even though it's a really rare case that someone has more than around 150 functions, even with bash-completion 4/21 ---- shell.c - main: malloc_set_register: change arg to 0 to disable keeping track of allocations and frees in table; change to 1 for malloc debugging on demand. Table is too small to keep effective track of allocations on a modern system, need to use something like valgrind 4/22 ---- shell.c - main: move cmd_init() (initialize word desc and word list object caches) after startup files are read so the cache doesn't get filled up with semi-permanent storage for shell functions and variables execute_cmd.c - execute_function: use maybe_set_debug_trap in the same way as source builtin change from 4/10 to allow functions to reset a DEBUG trap when set -T not in effect. From a report by Peng Yu 4/23 ---- execute_cmd.c - execute_disk_command: don't overwrite value of subshell_environment by assigning SUBSHELL_FORK; just OR in SUBSHELL_FORK so something like SUBSHELL_ASYNC or SUBSHELL_PIPE is preserved jobs.c - wait_for: don't call give_terminal_to if running in a subshell environment, not just if running_in_background is set; don't want shell functions waiting for a process to complete in a pipeline to give the terminal to the wrong process group. Part of fix for bug reported by Valentin Bajrami - make_child: child process should not try to give the terminal to the pipeline_pgrp if running as part of a pipeline (SUBSHELL_PIPE in subshell_environment); functions in execute_cmd.c don't set that flag until after make_child returns and the child process is running. Rest of fix for bug reported by Valentin Bajrami 4/24 ---- redir.c - redirection_error: if the redirection involves {varname}, only use `varname' in the error message if the error code is < 0 (one of the internal redirection errors, like ambiguous or bad variable name); otherwise the issue is probably with the filename and should be treated as usual. Fixes problem reported by Dr. David Alan Gilbert configure.ac,config.h.in - sys/ioctl.h: add check in AC_CHECK_HEADERS macro, define HAVE_SYS_IOCTL_H lib/readline/{rltty,terminal}.c - sys/ioctl.h: include unconditionally for ioctl declaration, avoid issues with `implicit declaration' warnings. Issue originally raised on gdb list by Chen Gang 4/27 ---- lib/malloc/table.c - mem_table: now a circular buffer showing the state of the last REG_TABLE_SIZE allocations rather than a hash table that quickly fills up builtins/declare.def - typeset_builtin: add -n to list of supported options. Omission reported by Valentin Bajrami unwind_prot.c - use object caches instead of malloc/free to allocate and deallocate unwind-protect elements - uwp_init: initialize unwind-protect element object cache unwind_prot.h - uwp_init: extern declaration shell.c - main: call uwp_init right after calling cmd_init -- initialize all the object caches at the same place lib/malloc/table.[ch] - mlocation_table: new table to keep track of allocation locations by file and line, functions to initialize table, record an allocation, and dump the table to stderr - mregister_alloc: call mlocation_register_alloc to record the location (source file/line) for each allocation to pinpoint malloc hot spots by number of calls. More detail than gprof parse.y - set_line_mbstate: replace free/xmalloc pair with xrealloc call - read_token_word: call alloc_word_desc instead of xmalloc so we can take advantage of the WORD_DESC cache 4/28 ---- execute_cmd.c - execute_simple_command: if command execution fails because the command is a directory, and the `autocd' option is set, add a `--' argument to the constructed `cd' command to protect against command names with the same name as options to `cd'. Report and fix from isabella parakiss 4/30 ---- jobs.c - printable_job_status, j_strsignal: change calls to strcpy and sprintf that write to retcode_name_buffer to use strncpy and snprintf to avoid buffer overflows caused by malicious translations. Bug and fix from Trammell Hudson 5/1 --- strtrans.c - ansicstr: make sure the buffer is at least 12 bytes to ensure enough space for any eventual call to u32cesc for one multibyte char 5/4 --- jobs.c - wait_for: if an interactive shell is running a loop and waiting for a non-builtin command to exit, and the command exits due to SIGINT, act as if the shell received the SIGINT as well and break the loop. This matches the behavior when the shell is running a builtin command in a loop, and when running a non-builtin command outside a loop, and seems more broadly useful than running the trap handler over and over again. Report originally from Kaz Kylheku builtins/set.def - unset_builtin: use different variables for keeping the state of the -f and -v options than the loop uses to decide whether or not to treat a name as a function or a variable. Fixes problem with unset_function setting `sticking' after you unset a function when invoked with no options. Bug report from Dreamcat4 shell.c - open_shell_script: set running_shell_script to 1, set to 0 in every other case (new variable) - main: when checking whether or not to call start_debugger, test running_shell_script instead of dollar_vars[1]. The goal is to not invoke the debugger for interactive shells but allow it to run for things like `bash --debugger -i /tmp/script'. Problem reported by Rocky Bernstein lib/readline/histexpand.c - history_event_delimiter_chars: new (as yet undocumented) variable containing by default characters that can delimit a history event specifier without requiring a `:': "^$*%-" as the documentation has always said. Fixes bug reported by Anders Granlund 5/10 ---- lib/glob/gmisc.c - match_pattern_char, match_pattern_wchar: if passed an empty string, return a match if the first character of the pattern is `*' subst.c - pat_subst: change to allow empty strings to be replaced as long as pattern matches empty string. Report and fix from isabella parakiss 5/12 ---- lib/sh/strtrans.c - ansic_wshouldquote: make behavior consistent in the event of an invalid multibyte sequence: return 1 to the caller so the string will be $'...' quoted builtins/cd.def - cd_builtin: print the right error message even if printable_filename changes errno (e.g., EILSEQ) lib/sh/shquote.c - sh_backslash_quote: in multibyte locales, even if is_basic does not return true, backslash-quote an ASCII character that backslash_table says should be quoted. Part of fix for bug reported by Igor Bogomazov bashline.c - directory_exists: new function, dequotes the string argument, removes any trailing slash, and returns true if the result is a valid existing filename - bash_filename_stat_hook,bash_directory_completion_hook: use directory_exists instead of calling lstat/stat on the quoted directory name. Rest of fix for bug reported by Igor Bogomazov 5/15 ---- aclocal.m4 - BASH_CHECK_MULTIBYTE: when checking for locale_charset, add $LIBINTL to $LIBS. If we're using the included lib/intl/libintl.a, it will include a version of locale_charset 5/17 ---- lib/readline/bind.c - sv_isrchterm: make sure we check for v[end] == 0 while in the loop looking for whitespace. Bug report and fix from Sergio Durigan Junior lib/sh/shmbchar.c - mbstrlen,mbsmbchar: move calculation of MB_CUR_MAX out of loop; calculate it once at beginning lib/sh/mbscmp.c - mbscmp: move calculation of MB_CUR_MAX out of loop; calculate it once at beginning lib/sh/mbscasecmp.c - mbscasecmp: move calculation of MB_CUR_MAX out of loop; calculate it once at beginning lib/sh/shquote.c - sh_backslash_quote: move calculation of MB_CUR_MAX out of loop; calculate it once at beginning lib/sh/casemod.c - sh_modcase: move calculation of MB_CUR_MAX out of loop; calculate it once at beginning subst.c - expand_word_internal: move calculation of MB_CUR_MAX out of loops; calculate it once at beginning of function (XXX should use locale_mb_cur_max) 5/22 ---- lib/readline/rlmbutil.h - _rl_wcwidth: new function, short-circuits wcwidth calls for ASCII printable characters, returns 1 for those - WCWIDTH: call _rl_wcwidth instead of wcwidth lib/readline/display.c - rl_redisplay: move calculation of MB_CUR_MAX out of loop; calculate it once at beginning. Report and patch from Ole Laursen 5/24 ---- lib/readline/text.c - rl_insert: change to attempt to batch-insert pending typeahead (not pushed input or input from a macro) that maps to rl_insert. An attempt to suppress redisplay until readline reads all typeahead -- pasted input, for instance. Inspired by report from Ole Laursen . XXX - need to make this a bindable variable; already controlled by _rl_optimize_typeahead lib/readline/rlprivate.h - _rl_optimize_typeahead: extern declaration lib/readline/doc/{hstech.texi,history.3} - next_history: clarify under what circumstances the history offset is incremented; suggestion from Glenn Golden print_cmd.c - print_arith_command: should be compiled in if either DPAREN_ARITHMETIC or ARITH_FOR_COMMAND is defined. Report from Flavio Medeiros flags.c - change_flag: new variable verbose_flag, set when `v' flag modified; sets value of echo_input_at_read correspondingly flags.h - verbose_flag: new extern declaration shell.c - long_options: --verbose now sets verbose_flag - main: set echo_input_at_read from verbose_flag after parsing any long options builtins/fc.def - fc_builtin: don't unwind-protect echo_input_at_read directly; set it to 1 before calling fc_execute_file as before, and reset it to value of global verbose_flag using set_verbose_flag called from an unwind_protect. Report from isabella parakiss builtins/shopt.def - shopt_set_debug_mode: new function, called when the extdebug shell option changes. Right now, it sets function_trace_mode when extdebug is enabled, and turns off when extdebug is disabled. The documentation has always said that shopt does this. Report from Peng Yu 5/27 ---- findcmd.[ch] - add support for EXECIGNORE shell variable, list of extglob patterns that prevent matching filenames from being identified as executable files variables.c - EXECIGNORE: arrange for findcmd.c:setup_exec_ignore to be called when $EXECIGNORE changes variables.h - sv_execignore: extern declaration doc/{bash.1,bashref.texi} - document EXECIGNORE variable 5/28 ---- lib/readline/callback.c - rl_callback_sigcleanup: new function, cleans up and unsets any state rl_callback_read_char uses. Intended to be used after a signal lib/readline/signals.c - _rl_handle_signal: call rl_callback_sigcleanup on SIGINT. Fixes bug reported to python group by Martin Panter lib/readline/isearch.c - _rl_isearch_cleanup: now a public function so rl_callback_cleanup can call it lib/readline/search.c - _rl_nsearch_cleanup: now a public function so rl_callback_cleanup can call it lib/readline/rlprivate.h - _rl_[in]search_cleanup: extern declarations lib/readline/readline.h - rl_callback_sigcleanup: new extern declaration lib/readline/doc/rltech.texi - rl_callback_sigcleanup: documented lib/readline/readline.h - bump readline version to 7.0 due to addition of rl_callback_sigcleanup 5/29 ---- builtins/declare.def - declare_internal: if we are trying to change attributes or value for a nameref variable whose value points to an unset variable, make sure we create a new variable whose name is the value of the nameref variable. That is, declare -n foo=bar unset foo # unsets bar declare -i foo should create a (invisible) variable named `bar' with the integer attribute. Fixes problem reported by Greg Wooledge builtins/set.def - unset_builtin: if we find a nameref variable when we look it up with find_variable (the returned variable has a name different from what we looked up), make sure we use that new name for the rest of the function rather than rely on unbind_variable to do the same thing as find_variable. Fixes problem reported by Greg Wooledge - unset_builtin: if we try to unset a nameref variable whose value is not a set variable, make sure we don't try to unset the nameref itself, but rather the variable it points to. This ensures that the following always works as it should declare -n foo=bar unset foo # unsets bar and leaves foo unchanged lib/readline/funmap.c - vi-insertion-mode: make sure it maps to rl_vi_insert_mode, which is actually what the `i' keybinding in vi_movement_keymap maps to. Cosmetic fix from isabella parakiss 6/2 --- general.c - conf_standard_path: moved get_standard_path here from command.def; made public function general.h - conf_standard_path: extern declaration 6/3 --- parse.y - parse_matched_pair: make sure that command substitution constructs within ${...} expansions are parsed as such using parse_comsub, so the P_FIRSTCLOSE flag doesn't stop parsing if a right brace appears within the comsub. Report from Christos Zoulas findcmd.c - search_for_command: take two new flags for FLAGS arg: CMDSRCH_STDPATH, which means use the command -p path, and CMDSRCH_TEMPENV, currently unused. Existing flag value of 1 now CMDSRCH_HASH - search_for_command: don't set `path' to NULL if it wasn't found in the temporary environment; we can use it later - search_for_command: look for the hashed file if temp_path == 0, not if we set path back to 0 (or path lookup failed) - search_for_command: just call find_user_command_in_path with a pathlist argument that's set depending on the flags passed - find_in_path: new function, just calls find_user_command_in_path at this point - search_for_command: don't put non-existent files into the hash table findcmd.h - CMDSRCH_*: new flags for search_for_command - find_in_path: extern declaration execute_cmd.c - execute_disk_command: check for CMD_STDPATH and call search_for_command with CMDSRCH_STDPATH if it's set builtins/command.def - command_builtin: if -p given without -v or -V, set the CMD_STDPATH flag in the simple command we construct from the rest of the argument list. This fixes the command -p cmd issue with changing $PATH for the duration of `cmd's execution reported by builtins/common.h - CDESC_STDPATH: new flag for describe_command; means to use standard path for command -pv or -pV builtins/command.def - command_builtin: if -p given with -v or -V, pass the CDESC_STDPATH flag to describe_command builtins/type.def - describe_command: if given the CDESC_STDPATH flag, use find_in_path with the conf_standard_path() result. Since `all' will never be on with this flag, at least not yet, we can allocate and free the path list. Fixes the bug of a temporary environment value for PATH preceding a `command -pv' or `command -pV' persisting when the command finishes and means we use fewer unwind-protects 6/4 --- Makefile.in (plus other Makefile.ins) - VPATH: remove `.' from VPATH value. It's not needed any more and it trips a GNU make 4.1 bug on Cygwin. Report and fix from Eric Blake lib/readline/histfile.c - history_tempfile: new function, takes history file name as template and creates a temp file name including the PID - history_truncate_file, history_do_write: instead of creating backup and restoring it on failure, use history_tempfile to create a temp file name, write to it, then rename to original history filename (handling symlinks using history_restore()). Original report and patch from Russell Stuart doc/{bash.1,bashref.texi} - ulimit: clarify that -c and -f are in increments of 512 bytes when in posix mode. Fix from Robin Johnson via 6/9 --- execute_cmd.c - execute_in_subshell: don't call restore_default_signal for the exit trap, reset_signal_handlers already does the right thing and keeps the trap string around; no need to kill trap string. Bug report from Miroslav Koskar 6/11 ---- nojobs.c - find_proc_slot: now takes pid as an argument to avoid finding old procs when pids wrap around - add_pid: pass pid to find_proc_slot to avoid multiple instances of the same pid in the list when pids wrap around. Fixes bug reported by Roy Keene execute_cmd.c - REAP: test for job_control == 0 also to determine whether or not to call reap_dead_jobs, since shells without job control enabled don't report on background process status doc/bash.1,lib/readline/doc/hsuser.texi - history: clarify documentation of -a option to note that it will not append the same line to the history file more than once. Fixes problem reported by Reuben Thomas 6/12 ---- execute_cmd.c - execute_in_subshell: don't bother decrementing subshell_level before this returns; the caller will just exit. This means that $BASH_SUBSHELL will have consistent values in the subshell and any subsequent exit trap. Fixes bug reported by Miroslav Koskar - shell_execve: before longjmp to subshell_top_level, call reset_parser to free up any input line and stack of pushed strings parse.y - mk_alexpansion: if the last character of the alias is a shell metacharacter, don't add a space to the string following the alias value, since that will change the meaning of the command. THIS IS NOT BACKWARDS COMPATIBLE AND MAY REQUIRE REVERTING. Inspired by an email message from Jilles Tjoelker to austin group 6/13 ---- subst.c - dequote_string: don't turn strings consisting of a single CTLESC into empty strings; return them unmodified. The idea is that there is nothing to quote. This means that something like c=$'\001' x=$c results in x containing '\001' when IFS=$'\001'. See if this will cause problems by adding a debugging statement in the code 6/15 ---- parse.y - CHECK_FOR_RESERVED_WORD: don't return ESAC when you read `esac' after a `|' while in a case pattern list (parser_state & PST_CASEPAT). This is Posix grammar rule 4 (http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#tag_18_10_02) 6/16 ---- jobs.c - wait_for: make sure we queue SIGCHLD signals if a SIGCHLD trap handler is running (signal_in_progress or IMPOSSIBLE_TRAP_HANDLER), since we allow recursive trap invocations now - wait_for: make sure we treat SIGCHLD as trapped if the trap handler is set to IMPOSSIBLE_TRAP_HANDLER so we don't miss signals during calls to run_sigchld_trap() trap.c - run_pending_traps: since we allow recursive trap handler invocations, make sure we set pending_traps[SIGCHLD] to 0 while we mark the signal handler in progress (SIG_INPROGRESS) and don't reset it to 0 after run_sigchld_trap returns in case more SIGCHLDs have been queued while it's running 6/22 ---- lib/readline/readline.c - _rl_dispatch_subseq: if we shadowed any function, not just self- insert, bind that function to the key we shadowed and dispatch on it. This preserves and dispatches off the original key, not ANYOTHERKEY, and makes sure we are dispatching from a keymap with the right key binding. Fixes bug reported by Carlos Pita 6/23 ---- bashline.c - attempt_shell_completion: don't read past the end of rl_line_buffer while skipping assignment statements before command names. Fix for bug reported by Hanno Böck lib/readline/parens.c - _rl_enable_paren_matching: make paren matching work in vi insert mode. Really, this time. Bug report from Carlos Pita 6/29 ---- parse.y - parse_comsub: when starting to parse a word, make sure that we set lex_rwlen to 0 along with lex_wlen if we are in a position where we can read a reserved word. Fixes bug reported by Nathan Neulinger 6/30 ---- sig.h - include unconditionally and rely on guards in sig.h to keep from including it more than once. Irix 6.5 defines SIG_DFL in other files and defeats the test of SIG_DFL. Fixes bug reported by Klaus Ziegler alias.c - ALIAS_HASH_BUCKETS: table size increased to 64 7/7 --- {jobs,nojobs}.c - wait_sigint_cleanup: new function, to be called by longjmp targets for wait_intr_buf. Performs any necessary cleanup of internal waitchld state before going back to top level jobs.h - wait_sigint_cleanup: new extern declaration builtins/wait.def - wait_builtin: call wait_sigint_cleanup after longjmp back to wait_intr_buf on interrupt [bash-4.4-alpha frozen] 7/7 --- bashline.c - attempt_shell_completion: when skipping assignment statements, don't skip past rl_point to find the word on which to perform programmable completion. Fixes bug reported by Hanno Böck 7/10 ---- pcomplib.c - compspec_copy: set the refcount on the copy to 1, since it's a freshly allocated copy and there are no other pointers to it. Fixes memory leak from gen_progcomp_completions lib/readline/complete.c - printable_part: if the pathname passed is just `/', just return it instead of potentially reading before the `/'. Fixes bug reported by Hanno Böck 7/11 ---- support/man2html.c - change calls to malloc to use xmalloc; provide standard xmalloc definition. Report from Bill Parker builtins/wait.def - wait_intr_flag: set to denote validity of wait_intr_buf as target for longjmp; reset by WAIT_RETURN macro; should always be 0 when wait_builtin returns quit.h - CHECK_WAIT_INTR: now also checks wait_intr_flag to ensure valid state of wait_intr_buf before sh_longjmp {jobs,nojobs}.c - wait_intr_flag: extern declaration for use by CHECK_WAIT_INTR sig.c - wait_intr_flag: set back to 0 every place the shell resets state after a jump to top_level - sigint_sighandler: only set wait_signal_received if wait_intr_flag indicates that the wait builtin has correctly set wait_intr_buf. Final fix for bug reported by Ondrej Oprala subst.c - parameter_brace_expand: fix so that ${!vvv@} is not interpreted as introducing a `@' operator, but instead retains its meaning as all variables whose name begins with vvv. Fixes bug reported by konsolebox 7/14 ---- execute_cmd.c - execute_connection: remove the fork optimization from and AND_AND and OR_OR cases; it has trouble with `a && b && c', since the `&&' is left-associative. Putting that code here can result in unwanted fork optimizations since the `b' will end up as `second' on a recursive call to execute_command. Bug reported by Mike Frysinger builtins/evalstring.c - optimize_fork: moved code from execute_connection to here; checks whether the RHS of a && or || passes the suppress_fork() tests and turns on the NO_FORK bit if so - parse_and_execute: if suppress_fork() tests fail, check whether the command to be executed is a connection and attempt to optimize the fork for the right side using optimize_fork(). The key is dealing with the entire command passed to `bash -c' or `eval' builtins/common.h - optimize_fork: new extern declaration execute_cmd.c - execute_command_internal: enable code put in to undo redirections before exiting the shell due to `set -e' in the case of a shell function that gets command-not-found or a syntax error. Reported back in May, 2014 by Mark Ferrell bashhist.c - bash_history_inhibit_expansion: enable change from 1/10 that skips over history expansion character in command and process substitution 7/15 ---- lib/readline/{history,undo}.c - replace_history_data now _hs_replace_history_data 7/16 ---- lib/readline/readline.c - make sure _rl_to_lower doesn't ever get characters that are outside of its valid range by casting arg to unsigned char lib/readline/colors.h - change completion prefix color (C_PREFIX) to be the same as C_SOCK, which is less likely to collide with file type colorings (magenta) lib/readline/complete.c - fnprint: now takes an additional argument: the full, expanded pathname corresponding to the `toprint' argument - fnprint: don't zero out prefix_bytes if we're not printing the ellipisis but we still want to use prefix_bytes for displaying the common prefix in a color - print_filename: move colored-stats display responsibility into fnprint - fnprint: change so colored-prefix and colored-stats can cooperate and display filenames with multiple colors: one for the prefix and one for the file type. The prefix is displayed in the prefix color and the remainder of the pathname, if any, is displayed in the color corresponding to the file type. Report and suggestion from Mike Frysinger 7/17 ---- lib/readline/readline.[ch] - rl_readline_state: now unsigned long, prepare for more than 32 states 7/18 ---- lib/readline/history.h - history_offset: now declared here, was always a global variable - copy_history_entry: now declared here, part of the public interface - alloc_history_entry: now declared here, part of the public interface 7/20 ---- lib/readline/readline.h - RL_STATE_CHARSEARCH: new state, for vi mode character search. Was previously handled by simply setting _rl_callback_func lib/readline/vi_mode.c - rl_domove_motion_callback: split into two functions, remaining code after _rl_dispatch call goes into new function _rl_vi_domove_motion_cleanup - rl_vi_char_search: set RL_STATE_CHARSEARCH - rl_vi_char_search: sets `i2' member of callback data to the key used to invoke the char search - _rl_vi_callback_char_search: unset RL_STATE_CHARSEARCH after reading the character to find - _rl_vi_domove_motion_cleanup: make sure RL_STATE_VIMOTION is unset in all cases where this function returns - _rl_vi_domove_motion_cleanup: use `end' member of vi motion context as value of rl_end to restore after adding blank at end of line in rl_domove_motion_callback; avoids need to keep `old_end' and somehow pass it between the two functions lib/readline/rlprivate.h - _rl_vi_domove_motion_cleanup: new extern declaration lib/readline/callback.c - rl_callback_read_char: handle RL_STATE_CHARSEARCH, including case where char search is vi motion command for d/c/y/m and RL_STATE_VIMOTION is set when this is called. Last of set of fixes for bug reported by Carlos Pita with vi-mode `dty' command in callback mode 7/21 ---- subst.c - command_substitute, process_substitute: add QUIT call in the child after restoring original signal handlers to catch post-fork SIGINT or terminating signal jobs.c,nojobs.c - make_child: after forking, clear interrupt_state in child before restoring signal mask so children get their own set of interrupts 7/22 ---- lib/sh/shquote.c - sh_double_quote: since all uses of the return value from this function are for display and not subject to subsequent expansion, we don't need to protect CTLESC and CTLNUL with CTLESC. Bug report and fix from isabella parakiss parse.y - decode_prompt_string: need to protect literal instances of CTLESC and CTLNUL with CTLESC, even if they are not coming out of one of the backslash-expansions subst.c - make_internal_declare: fixed memory leak; make_internal_declare should free the memory it allocates to construct the word list. From Red Hat bug 1245233; https://bugzilla.redhat.com/show_bug.cgi?id=1245233 lib/readline/readline.c - _rl_dispatch_subseq: when checking whether an ESC in vi insert mode should be treated as part of a key sequence or as a single char, make sure we are getting input from the keyboard before we check _rl_input_queued(). If we are getting input from a macro, this causes a needless delay. Report from Carlos Pita 7/23 ---- input.c - save_bash_input: before checking to see whether or not a new fd has a stale associated buffer, make sure that the new fd is within bounds. Bug report and fix from Alexey Makhalov 7/24 ---- doc/{bash.1,bashref.texi} - note that the `return' builtin accepts arguments beginning with `-' without requiring `--' 7/27 ---- examples/loadables/Makefile.in - add variables and targets to support `make install'; installs supported builtins in $(libdir)/bash 7/28 ---- lib/readline/{readline,rlprivate}.h, lib/readline/display.c - rearrange some structure members for better alignment and to reduce padding. Fix from tests/run-* - use BASH_TSTOUT instead of hardcoded /tmp/xx as output file trap.[ch] - maybe_set_error_trap: set a trap for the ERR trap if it is not already trapped; similar to maybe_set_debug_trap back from 4/10 lib/malloc/malloc.c - internal_free: make sure `ubytes' is initialized if MALLOC_WATCH is defined and we'll be passing it to _malloc_ckwatch. Report from Flavio Medeiros execute_cmd.c - execute_function: use maybe_set_error_trap in the same way as maybe_set_debug_trap. Fixes bug reported by David Waddell 7/29 ---- jobs.c - wait_for: no longer need to call job_signal_status when checking whether a comsub shell reaped a child that died due to SIGINT; code now uses child_caught_sigint 7/30 ---- builtins/exec.def - exec_builtin: after call to restore_original_signals, need to call default_tty_job_signals to undo changes done by initialize_job_signals. Fix for problem reported by Andreas Schwab doc/{bash.1,bashref.texi} - clarify definition of metacharacter to explicitly include newline; prompted by report from George Gallo 8/3 --- lib/readline/search.c - _rl_free_history_entry: should be void, not int. Report from Dilyan Palauzov general.h - sh_load_func_t, sh_unload_func_t: new function pointer types, used by enable for load and unload hook functions builtins/enable.def - dyn_load_builtin: attempt to execute a function named BUILTIN_builtin_load, where BUILTIN is the name of the command being loaded. If that function returns 0, the load fails - dyn_unload_builtin: attempt to execute a function named BUILTIN_builtin_unload, where BUILTIN is the name of the command being unloaded. It offers a dynamic builtin a way to clean up after itself. Inspired by suggestion from Piotr Grzybowski in response to a bug report from isabella parakiss 8/6 --- lib/readline/colors.[ch] - _rl_print_color_indicator: now takes `const char *' argument lib/readline/complete.c - colored_stat_start: now takes `const char *' argument lib/malloc/table.h - ma_table_t: `file' member is now `const char *' lib/malloc/table.c - mlocation_register_alloc: make sure variable assigned to `file' in ma_table_t is of type `const char *'. Fixes from Dilyan Palauzov lib/termcap/termcap.[ch] - tputs: should return int instead of void lib/readline/readline.h - RL_STATE_DONE: correct value, avoid collision lib/readline/vi_mode.c - _rl_vi_redoing: now global, added _rl_ prefix to `vi_redoing' lib/readline/readline.c - _rl_subseq_result: call _rl_dispatch_subseq instead of rl_dispatch to avoid changing _rl_dispatching_keymap - _rl_subseq_result: in the -2 case, set _rl_dispatching_keymap to the map passed as an argument. Without this, vi mode doesn't call rl_vi_set_last when it should - _rl_dispatch_subseq: if redoing a vi mode command with `.' (_rl_vi_redoing != 0) in callback mode and we're indirecting into another keymap that has shadowed the key corresponding to the command we are redoing, just call _rl_subseq_result immediately: vi redo doesn't need to read any additional input, and rl_vi_redo assumes that a single call to rl_dispatch is sufficient. Fixes bug reported by Carlos Pita 8/7 --- subst.c - parameter_brace_expand: if a substitution (parameter expansion) error occurs, and shell_compatibility_level is <= 43, return expansion error as in all previous versions. If shell_compatibility_level is > 43, a posix-mode non-interactive shell will consider this a fatal error. Problem reported by Christian Neukirchen doc/bashref.texi - documented that word expansion errors cause fatal errors in posix mode non-interactive shells 8/8 --- subst.h - SD_COMPLETE: skip_to_delim being called as part of word completion - SX_COMPLETE: one of the string_extract functions being called as part of word completion bashline.c - find_cmd_start,find_cmd_end,find_cmd_name: call skip_to_delim with SD_COMPLETE flag subst.c - skip_double_quoted: takes new flags argument; changed callers - skip_double_quoted: if flags argument includes SX_COMPLETE, pass it to extract_command_subst - extract_command_subst: if flags&SX_COMPLETE, call extract_delimited_string instead of xparse_dolparen, since completion may call this for unterminated command substitutions. Fixes (imperfectly) bug reported by Ingo Ruhnke - skip_to_delim: if passed the SD_COMPLETE flag, pass the SX_COMPLETE flag to skip_double_quoted - char_is_quoted: pass SX_COMPLETE flag to skip_double_quoted - unclosed_pair: pass SX_COMPLETE flag to skip_double_quoted 8/9 --- execute_cmd.c - time_command: catch longjmps to top_level and print command timing statistics even after a jump to top_level. Fixes issue reported by Sam Watkins 8/10 ---- config-top.h - OPENLOG_OPTS: if SYSLOG_HISTORY is defined, define to LOG_PID, so each message is tagged with the pid bashhist.h - bash_syslog_history: the first time it's called, call openlog with OPENLOG_OPTS and SYSLOG_FACILITY 8/11 ---- doc/{bash.1,bashref.texi} - GROUPS,FUNCNAME: change description to note that assignments are silently ignored, but do not return an error (which would constitute an assignment error and cause posix mode shells to abort). Problem pointed out by Grzegorz Bajson 8/12 ---- subst.c - parameter_brace_expand_indir: if the value of the indirectly expanded variable isn't something that the shell would expand if it were inside ${}, error out right away before calling parameter_brace_expand_word. Fixes problem reported by isabella parakiss - parameter_brace_expand: handle returning &expand_wdesc_error or &expand_wdesc_fatal from parameter_brace_expand_indir in case it does that someday 8/13 ---- jobs.c - {save,restore}_pipeline: saved_pipeline now a linked list of pipelines, new ones allocated in save_pipeline and freed in restore_pipeline. This allow multiple nested calls to save_pipeline (e.g., in traps and then in process substitution). Fix for bug reported by isabella parakiss print_cmd.c - named_function_string: if printing a function with the same name as a reserved word, add the `function ' keyword before the name to avoid parsing problems when trying to reuse it as input. Fix for bug reported by isabella parakiss 8/14 ---- lib/readline/text.c - rl_insert: when optimizing typeahead, make sure we set rl_last_func ourselves if we set pending input, since the mainline code path won't set rl_last_func if input is pending. Fixes bug reported by Hiroo Hayashi 8/15 ---- builtins/type.def - describe_command: if using the short description (CDESC_SHORTDESC) in posix mode, describe posix special builtins as such. Requested by Stephane Chazelas builtins/enable.def - BASH_LOADABLES_PATH: a colon-separated list of directories where bash looks for loadable builtins specified as arguments to `enable -f' doc/{bash.1,bashref.texi} - BASH_LOADABLES_PATH: document new shell variable configure.ac,Makefile.in,builtins/Makefile.in - loadablesdir: set in configure, substitute into Makefiles. Reserved for future use 8/18 ---- subst.c - shell_expand_word_list: if make_internal_declare fails, make sure to propagate that error return back and make the assignment statement fail. Fixes seg fault reported by Sergey Tselikh 8/20 ---- builtins/declare.def - declare_internal: made a slight tweak to the warning message about quoted compound assignments by printing it only if the array (indexed or assoc) does not already exist 8/21 ---- braces.c - mkseq: call strvec_dispose before throw_to_top_level if we saw SIGINT to avoid a memory leak trap.c - maybe_set_return_trap: set the RETURN trap to string if it's not already trapped, in the same way as the debug and error traps execute_cmd.c - execute_function: use maybe_set_return_trap to allow functions to set return traps that persist across calls even if function tracing is enabled lib/readline/input.c - rl_gather_tyi: make sure errno reset to 0 after select call, for ioctl and read - rl_gather_tyi: if read returns -1/EIO, return -1 - rl_read_key: if rl_gather_tyi returns -1/EIO, return READERR if in RL_STATE_READCMD (reading command in readline_internal_char), EOF if not (like rl_getc). Continue to return '\n' on other errors; rl_done = 1 in any case. Fix for issue reported by Lubomir Rintel lib/readline/{misc,text,vi_mode}.c - fix return values from rl_read_key to handle < 0 8/22 ---- parse.y - parsing_redirection: macro that expands to true if the last read token (always passed as an argument) is a redirection token that will leave us reading a word that's the target of the redirection - command_token_position: make sure that even if the parser state indicates we are reading a simple command (PST_REDIRLIST), we do not perform alias expansion on a WORD that is part of a redirection, using parsing_redirection() to do so. Fixes but reported to Red Hat by Robert Alm Nilsson https://bugzilla.redhat.com/show_bug.cgi?id=795795 builtins/declare.def - declare_internal: only print the warning now if the variable is not already an array or if we are not creating an array with -a or -A, cuts down the spurious warnings - declare_internal: even if the rhs of the assignment looks like a compound array, if we're not assigning to an existing array or we're not creating an array, make it an array subscript assignment. This means things like declare a[1]='(foo)' and a[1]='(foo)' behave identically builtins/help.def - wdispcolumn: change use of displen and add new dispchars variable to deal with locales where each wide character does not take up one column position. Fix for bug reported by Mingye (Arthur) Wang 8/23 ---- sig.c - sigint_sighandler: if we get a SIGINT while this_shell_builtin == wait_builtin, perform the special handling only if wait_intr_flag is non-zero, don't just use it to set wait_sigint_received. This makes sure we run bashline_set_event_hook. Fixes bug reported by isabella parakiss 8/25 ---- subst.h - SD_HISTEXP: new flag for skip_to_delim, indicates we are looking for the history expansion character (special double quote handling) - skip_to_delim: handle SD_HISTEXP flag, changes double quote handling to be similar to history library: don't call skip_double_quote, make sure single quotes aren't special in double quotes, but continue to skip over $(), ${}, and other similar constructs 8/27 ---- support/bash.pc.in - new file, pkgconfig template for bash, installed to support nascent loadable builtins dev environment examples/loadables/Makefile.inc - new file, example Makefile with correct definitions for loadable builtin development configure.ac - headersdir: directory where headers get installed to support loadable builtin development - examples/loadables/Makefile.inc, support/bash.pc: create Makefile.in - INSTALLED_HEADERS, INSTALLED_BUILTINS_HEADERS, INSTALLED_INCFILES, CREATED_HEADERS: new variables holding names of include files to install into $(headersdir) for loadable builtin development - install-headers,install-headers-dirs: new targets to support installing headers into $(headersdir) and subdirectories for loadable builtin development - uninstall-headers: target to uninistall headers in $(headersdir) - uninstall: call uninstall-headers whether or not install-headers was done; add uninstall in examples/loadables - install: add install in examples/loadables - clean: add clean in examples/loadables examples/loadables/Makefile.in - install-supported, install-unsupported: new targets, make install chooses one based on $(SHOBJ_STATUS) - install-supported: now cds to the top level build directory and installs the header files - install-supported: now installs Makefile.inc to $(loadablesdir) as an example - uninstall-supported,uninistall-unsupported: uninstall targets, make uninstall chooses one based on $(SHOBJ_STATUS) 8/31 ---- lib/glob/Makefile.in - CPPFLAGS: move after ${INCLUDES} so library-specific include paths take precedence. This is what the other parts of the shell do. Report from Poor Yorick builtins/read.def - reset_alarm: cancel alarm before restoring signal handler to avoid possible race condition. Report and fix from Oleg Popov subst.c - string_extract_verbatim: if SEPARATORS is "'", don't short-circuit to string_extract_single_quoted unless the SX_NOCTLESC flag is set. Callers expect this function to honor CTLESC, even to quote a single quote. Fixes bug reported by isabella parakiss and several others 9/2 --- include/chartypes.h - TOCTRL: handle '?' by special-casing it to 0x7f. Fixes a problem with $'\c?' reported by Helmut Karlowski redir.c - write_here_string: don't word-split the here string document. The bash documentation has always said this doesn't happen, even though bash has done so for years, and other shells that implement here- strings don't perform any word splitting. The practical effect is that sequences of IFS characters are collapsed to spaces. Fixes bug reported by Clint Hepner 9/3 --- doc/{bash.1,bashref.texi} - add \? to the list of backslash escapes expanded by $'...' ANSI-C quoting. It was expanded but not documented lib/readline/util.c - _rl_audit_tty: updated from Red Hat patch lib/readline/readline.c - HAVE_DECL_AUDIT_USER_TTY: correct #define check 9/13 ---- shell.c - read_from_stdin: make sure value set to 1 appropriately if the shell is reading from standard input, as Posix seems to specify (?) lib/readline/kill.c - _rl_copy_to_kill_ring: don't assume that rl_kill_ring has been allocated if _rl_last_command_was_kill is non-zero; make sure it's non-null before indexing into it 9/15 ---- variables.c - initialize_shell_variables: cope with a NULL env pointer. Fixes bug reported by ziyunfei <446240525@qq.com> 9/16 ---- builtins/exec.def - exec_builtin: if -c is supplied, pass an empty array instead of a NULL pointer to shell_execve variables.c - set_pwd: only believe $PWD if it begins with a slash; try to canonicalize it and set it to physical path if canonicalization fails. Reported by ziyunfei <446240525@qq.com> 9/17 ---- subst.c - do_compound_assignment: make sure to dispose of word list generated by expand_compound_array_assignment; fixes memory leak reported in https://bugzilla.redhat.com/show_bug.cgi?id=1264101 variables.c - adjust_shell_level: clamp the value of shell_level at 1000 and reset there, instead of > 1000, since the itos replacement code doesn't handle the value 1000. Fixes bug reported by ziyunfei <446240525@qq.com> shell.h,parse.y - save_parser_state,restore_parser_state: now save and restore redir_stack, short-circuiting if need_here_doc == 0. If we save a non-zero value for need_here_doc, we have to make sure there is something valid for gather_here_documents to work on. Fixes bug reported by Brian Carpenter 9/18 ---- array.c - array_to_assign: use ansic_shouldquote to check whether each element value contains non-printable characters and use ansic_quote to generate the value instead of using sh_double_quote unconditionally assoc.c - assoc_to_assign: if either the key or the value of an associative array element contains non-printable characters (ansic_shouldquote returns true), use ansic_quote to quote them instead of using double quotes unconditionally 9/19 ---- subst.c - pat_subst: handle REP being NULL. Fixes bug reported by Brian Carpenter builtins/evalstring.c - parse_string: if we get a longjmp to top_level with DISCARD as the code (in which case we are going to go on), return -DISCARD to our caller (always xparse_dolparen) after doing our own cleanup instead of calling jump_to_top_level parse.y - xparse_dolparen: if parse_string returns < 0, do the appropriate cleanup and then jump_to_top_level with the negative of the return value. This allows us to do the appropriate parser cleanup in case we're not going to exit the shell. Fixes bug reported by Brian Carpenter subst.c - extract_delimited_string: if a recursive call to one of the extract_ functions or a call to ADVANCE_CHAR leaves i past the end of the string, cut the loop off at the end of the string. Fixes bug reported by Brian Carpenter 9/20 ---- subst.c - get_var_and_type: return appropriate values if variable indirection results in a NULL variable. Fixes bug reported by Brian Carpenter lib/readline/history.c - history_get_time: handle strtol overflows caused by malicious modifications to timestamps in the history file. Fixes issue reported by rens@endoria.net 9/21 ---- lib/readline/rlconf.h - ENABLE_AUDIT_TTY_SUPPORT: now undefined (off) by default 9/24 ---- jobs.c - waitchld: if we get a SIGINT while waiting for a child to exit, but the kernel doesn't interrupt the waitpid(2) call, assume the child caught SIGINT if it exited for some reason other than SIGINT. Fix suggested by Stephane Chazelas input.c - make_buffered_stream: use B_TEXT in buffered stream flags instead of (typo) O_TEXT. Report and fix from Eric Blake 9/27 ---- execute_cmd.c - shell_execve: call reset_parser before calling initialize_subshell, which calls delete_all_aliases. reset_parser wants to free the pushed string list, which has pointers back into the alias table (use after free) - execute_simple_command: if we fork for an async command, make sure the child process increments shell_level before performing any word expansions, so $BASH_SUBSHELL is incremented. Fixes issue reported by ziyunfei <446240525@qq.com> 10/1 ---- builtins/kill.def - kill_builtin: make -L equivalent to -l for compatibility with other (Linux) versions of kill doc/{bash.1,bashref.texi} - kill: document new `-L' option 10/2 ---- [bash-4.4-beta frozen] 10/5 ---- doc/{bash.1,bashref.texi} - LC_TIME: document that the shell recognizes it and calls setlocale when it changes; and that the shell uses it for time formatting 10/6 ---- subst.c - skip_single_quoted: now takes an additional `flags' argument, like skip_double_quoted; changed callers - skip_single_quoted: now pays attention to SX_COMPLETE flag, allows backslash to escape single quote, like $'...' allows - skip_to_delim: added case where we are completing and we are skipping over a $'...' string to call skip_single_quoted with SX_COMPLETE flag to allow backslash to quote single quote. Fixes bug reported by Daniel Colascione - char_is_quoted: make analogous changes to handle $'...' so we can accommodate $'ab \' cd' 10/11 ----- subst.c,bashjmp.h - no_longjmp_on_fatal_error: now global so other parts of the shell can use it arrayfunc.c - expand_array_index: if no_longjmp_on_fatal_error is non-zero, don't longjmp to top_level, just return 0. Fixes bug reported by isabella parakiss subst.c - CQ_RETURN: restore old value of no_longjmp_on_fatal_error (oldjmp); make sure all callers (skip_matched_pair, skip_to_delim, char_is_quoted) save no_longjmp_on_fatal_error to oldjmp before setting it to 1 expr.c - expr_streval: if longjmping after an unbound variable error, jump back to expr toplevel (evalbuf) if no_longjmp_on_fatal_error is set in an interactive shell variables.c - push_func_var: if pushing an array variable, such as when a variable in the temporary environment is promoted to an array by mapfile when the shell is in posix mode (so variable assignments preceding special builtins or shell functions persist in the shell environment after the builtin returns), make sure to copy the array or hash table so the variable is duplicated correctly. Fixes seg fault reported by Linda Walsh 10/13 ----- variables.c - initialize_shell_variables: add call to sv_shcompat, so BASH_COMPAT can be set in the initial environment execute_cmd.c - execute_function: set loop_level to 0 only if shell_compatibility_level is greater than bash-4.3; this was kind of an incompatible change. Report from Carlos Pita COMPAT,doc/{bash.1,bashref.texi} - compat43: added loop_level changes to description 10/15 ----- lib/sh/casemod.c,lib/readline/display.c - make sure mb_cur_max variable is defined outside of HANDLE_MULTIBYTE if it's used outside HANDLE_MULTIBYTE. Fixes from Greg Wooledge support/Makefile.in - bash.pc: remove as part of distclean, not clean. Report from Andreas Schwab 10/17 ----- subst.c - array_var_assignment: make sure to call quote_string or quote_escapes on the assignment string, so spaces in the assignment survive word splitting. Fixes bug reported by isabella parakiss 10/20 ----- doc/{bash.1,bashref.texi} - word splitting: make sure that newline is listed as one of the IFS whitespace characters. Fixes omission reported by ziyunfei <446240525@qq.com> lib/readline/histfile.c - history_do_write: make sure that we only create and use the tempfile if the history file exists and is a regular file. Reported several times, most recent check the result of a report from 10/22 ----- jobs.c - delete_all_jobs: if running_only == 0, we are eventually going to clear the bgpids list, so don't bother to add pids to it in delete_job (call with DEL_NOBGPID flag if running_only == 0) 10/24 ----- jobs.[ch] - bgpids: new implementation from a patch from John Fremlin , uses an array for the list of the last CHILD_MAX terminated background pids, and a separate hash table to search it. The storage can be freed as a unit, and the size of the hash table (currently 4096) is independent of the size of the bgpids table subst.c - inherit_errexit: new variable to control whether or not command substitution inherits the -e (errexit) option. Disabled by default general.c - posix_initialize: set inherit_errexit = 1 when Posix mode is enabled builtins/shopt.def - inherit_errexit: new shell option, tracks value of inherit_errexit, allows command substitution to inherit the setting of errexit without posix mode. From a request and patch submitted by Christoph Gysin {version,version2}.c - use #if HAVE_SNPRINTF instead of #if defined in case configure decides to #define it to 0. Fixes problem reported by Klaus Ziegler configure.ac - when checking for sys/resource.h, make sure to include for the benefit of both old systems that require it and new versions of autoconf that require a header file to compile to report its presence. Reported by Klaus Ziegler 10/26 ----- subst.h - SD_ARITHEXP: new flag value for skip_to_delim, supports parsing arithmetic expressions in parameter expansions subst.c - skip_to_delim: handle SD_ARITHEXP flag by skipping parentheses for subexpressions and allowing ?: expression to not terminate an arithmetic expression delimited by `:' - skiparith: just call skip_to_delim with the SD_ARITHEXP option and the right delimiter string and return the right result. Fixes bug reported by include/shmbchar.h - strip out everything except what is needed to support is_basic and similar functions, since the mbchar_t typedef apparently conflicts with some AIX-specific type definition. Problem reported by Michael Felt 10/27 ----- builtins/{set,ulimit}.def - {set,ulimit}_builtin: make sure that --help is treated the same as -? and prints a message and returns. Fixes bug reported by ziyunfei <446240525@qq.com> builtins/*.def - make sure to consistently use builtin_help() instead of mix of that function and builtin_usage() 10/29 ----- doc/{bash.1,bashref.texi} - BASH_CMDS, BASH_ALIASES: note that removing elements from these array variables is not currently reflected in the command hash table and alias list, respectively. Reported by isabella parakiss 10/30 ----- eval.c - reader_loop: if PS0 is set in an interactive shell, expand and display it after reading a (complete) command but before executing it. This differs from the DEBUG trap because the DEBUG trap is executed once for each simple command (and some others, like each time through a for loop). From a patch submitted by Dan Stromberg parse.y - prompt_again: set ps0_prompt from $PS0 in an interactive shell doc/{bash.1,bashref.texi} - PS0: document new prompt string 11/3 ---- subst.c - cond_expand_node: if special > 0, remove quoted nulls from the word list, since no word splitting takes place. Fixes bug reported by Corentin Peuvrel 11/4 ---- doc/{bash.1,builtins.texi} - inherit_errexit: add to shopt description; modify Posix mode description 11/9 ---- lib/readline/util.c - rl_tilde_expand: fix out-of-bounds read caused when this function is invoked with rl_point == 0. Bug reported by Hanno Böck 11/11 ----- {nojobs,jobs}.c - get_original_tty_job_signals: force the original signal disposition for SIGTSTP/SIGTTIN/SIGTTOU to SIG_DFL in interactive shells. This will force child processes to get SIG_DFL even if the shell is started with those signals ignored. From a bug report from Keith Thompson 11/13 ----- make_cmd.c - make_function_def: if a function is being defined without a source file, and the shell has not been initialized, assume that the function is being imported from the environment and set the source_file struct member to "environment". Set to "main" if the shell has been initialized. Fixes bug reported on savannah by Arno-Can Uestuensoez https://savannah.gnu.org/support/index.php?108903 subst.c - skip_to_delim: if the shell is in posix mode (posixly_correct != 0) and performing history expansion ((flags & SD_HISTEXP) != 0), allow double quotes to quote the history expansion character. TENTATIVE CHANGE, inspired by https://savannah.gnu.org/support/index.php?108491 11/14 ----- jobs.c - wait_for: modify change from 4/23 to only give the terminal to shell_pgrp if not running in the background and subshell_environment indicates we're not in a pipeline and not an async command. Fixes bug reported by konsolebox 11/16 ----- doc/{bash.1,bashref.texi} - document new posix-mode behavior of not giving ! special meaning within double quotes, even if histexpand is enabled 11/17 ----- builtins/evalfile.c - fc_execute_file: pass FEVAL_BUILTIN flag to _evalfile so it returns the value of parse_and_execute after executing the commands in the file. Fixes bug reported by Chris Marusich 11/20 ----- subst.c - process_substitute: don't unconditionally set pipeline_pgrp to shell_pgrp; do that only if pipeline_pgrp == 0 (we haven't forked yet) or if we know we haven't forked for a command, a pipeline, or an async command. Fixes bug reported by as https://savannah.gnu.org/support/index.php?108593 11/24 ----- variables.c - set_pwd: inherit the value of OLDPWD from the initial environment if it names a directory. From a report from John Wiersba builtins/set.def - -T: fix help text to note that the DEBUG and RETURN traps are both inherited if this is set. Report from Grisha Levin doc/{bash.1,bashref.texi} - HISTIGNORE,GLOBIGNORE: note that the pattern matching is subject to the setting of the `extglob' variable. Clarification raised by Aharon Robbins builtins/jobs.def,doc/{bash.1,bashref.texi} - disown: add text to synopsis making it clear that disown takes a pid argument. From a report from Matthew Stanfield findcmd.c,doc/{bash.1,bashref.texi} - EXECIGNORE: change pattern matching to use FNMATCH_EXTFLAG like other IGNORE variables. From a report from Aharon Robbins trap.c - _run_trap_internal: make sure to catch and use return values supplied as arguments to `return'; instead of just catching return, make sure we use return_catch_value as well. Fixes bug reported by Grisha Levit , affects RETURN, DEBUG, ERROR traps 11/30 ----- lib/readline/display.c - rl_redisplay: fix code that deals with prompts that contain multibyte characters whose physical length is longer than the screen width (so the prompt line wraps) to deal with invisible characters in the prompt and set the inv_lbreaks offsets correctly and set lpos appropriately - _rl_move_cursor_relative: always called with either visible line or invisible line. Decide whether we are being called with the invisible line (or a portion thereof) as the DATA argument; make sure to account for invisible characters in the last line of the prompt by offsetting the desired buffer offset (NEW) by where DATA starts in invisible line and checking whether that is past the last invisible character in the prompt - _rl_move_cursor_relative: do the same for the visible line by using the visible line lbreaks array instead of _rl_screenwidth, since that fails in the presence of multibyte characters (buffer offset vs. physical screen position). These fix bug with prompts with multibyte characters and invisible characters that are longer than the screen width reported by Ryo Furue 12/1 ---- parse.y,make_cmd.c - here_doc_first_line: new variable, set to non-zero to indicate the first line of possibly multiple here documents associated with the current command. Set in parse.y:gather_here_documents() before reading any here documents, set to 0 after reading the first line of any here document in make_cmd.c:make_here_document(). parse.y - history_delimiting_chars: if parser_state indicates we are reading a here document, don't use current_command_line_count to see whether or not we're on the first line of a here document -- it can be fooled by multi-line commands (or even backslash-escaped newlines). Use here_doc_first_line to check that directly. Fixes bug reported by Geoffrey Allott shell.h,parse.y - save and restore here_doc_first_line as part of the shell's parser state lib/sh/tmpfile.c - use ANSI-C rand() interface if random() not available. Fixes problem with linking on (ancient) Solaris 2.4 reported by Klaus Ziegler 12/2 ---- aclocal.m4 - RL_LIB_READLINE_VERSION: if cross-compiling, assume current version of readline if cross-compiling bash while using an external readline library. Fix from Barry Davis execute_cmd.c - execute_builtin: save and restore value of builtin_ignoring_errexit instead of assuming we can set it back to 0 -- the ignore-errexit cases might be nested - execute_builtin: after a builtin executes and eval_unwind is set, set exit_immediately_on_error to 0 if builtin_ignoring_errexit is set, and back to the value of errexit otherwise, just like the code that sets -e does. Fixes short-circuiting bug reported by Marcin Swigon 12/5 ---- builtins/setattr.def - set_or_show_attributes: instead of using a fixed string for the options argument to `declare', which results in export adding the readonly attribute, build the option string dynamically like in subst.c:shell_expand_word_list() - set_or_show_attributes: don't add the -r option to declare if the attribute passed as an argument doesn't include att_readonly; add -x if the attribute includes att_exported - set_or_show_attributes: add the -g option to declare so readonly and export in functions that go through this code path don't create local variables. Part of fix for bug reported by ziyunfei <446240525@qq.com> execute_cmd.c - make_internal_declare: if we have an assignment builtin that isn't supposed to create local variables (export/readonly), make sure to add the W_ASSNGLOBAL flag to each assignment statement word. Part of fix for bug reported by ziyunfei <446240525@qq.com> where readonly can create local variables when invoked with a compound array assignment as an argument within a shell function subst.c - make_internal_declare: added third argument: command name. Not used in any useful way yet - do_compound_assignment: handle attempted assignments to readonly variables and print error messages, like bind_variable does and the callers expect. This means that assignments to readonly array variables can't go through a code path that allows them to fail silently doc/Makefile.in - uninstall: run install-info --delete after removing the installed info file to update the directory file execute_cmd.c - execute_disk_command: since we still (and have always) perform redirections in child processes, we need to mark that state so we know that we're in a child process. subshell_performing_redirections is the new variable; non-zero when executing do_redirections() subst.c - process_substitute: if we are executing process substitution in a redirection expansion context (expanding_redir == 1), the child process should not have access to any temporary environment the parent has. Call flush_tempenv() in the child process to get rid of it. Fix for bug reported by Clint Hepner 12/9 ---- jobs.c - waitchld: if in posix mode and attempting to let SIGCHLD interrupt the wait builtin, make sure wait_intr_flag is non-zero before trying to longjmp to wait_intr_buf. Fixes bug reported by Aharon Robbins 12/10 ----- lib/readline/display.c - update_line: if a prompt containing invisible characters changes, and the change is within the prompt string, we need to redraw the entire prompt. After that redraw, if the cursor position is after the first index where the old and new lines differ, we can short- circuit the update if the line length stays the same - update_line: if we redraw the entire prompt string because the prompt contains invisible characters and the first difference is in the middle of the prompt, we possibly need to adjust our idea of the first differing character in the old and new lines. We punt and do a dumb update in this case. Fixes bug with dynamic vi-mode mode strings containing invisible characters reported by Dylan Cali 12/11 ----- lib/readline/display.c - update_line: if we move up to a new screen line, we need to recompute the number of invisible characters on the new current line (visible_wrap_offset). Fixes bug reported to bug-readline by Per Bothner - update_line: update code that attempts to compute where we are in the new line buffer. Fixes `vt100' bug reported to bug-readline by Per Bothner 12/18 ----- execute_cmd.c - execute_function: unwind-protect and restore BASH_ARGC and BASH_ARGV - restore_funcarray_state: now global execute_cmd.h - make func_array_state type global, add extern function declaration for restore_funcarray_state builtins/evalfile.c - _evalfile: use restore_funcarray_state and pop_args as unwind-protects to restore BASH_{SOURCE,LINENO,ARGC,ARGV} and FUNCNAME on interrupts. Fixes bug reported back on 11/10 by Grisha Levit 12/20 ----- lib/readline/isearch.c - _rl_isearch_dispatch: after removing the only character from the search string with DEL, leaving the search string empty, don't match the previous line if we didn't have a match before 12/22 ----- bashhist.c - enable_history_list,remember_on_history: initialize to 0 instead of 1 shell.c - init_interactive,init_interactive_script: set enable_history_list and remember_on_history to 1 (defaults) 12/23 ----- variables.c - initialize_shell_variables: don't inherit PS4 from the environment if the shell is running with euid == 0; just reset it to `+ '. This is a known potential vulnerability, since PS4 expansion performs command substitution in the root shell's context before displaying it. Discussion started by up201407890@alunos.dcc.fc.up.pt bashhist.c - bash_history_reinit: initialize remember_on_history to the value of enable_history_list (set -o history). Fixes bug reported by Stephane Chazelas with command-line option `-o history' not having any effect 12/28 ----- lib/readline/{history.h,histfile.c} - history_file_version: new int variable, reserved for future use - history_multiline_entries: new int variable, can be set by the calling application to enable reading multi-line history entries from the history file (currently undocumented) lib/readline/history.c - _hs_append_history_line: new function, append a line passed as an argument to a specified history entry, used to create multi-line history entries lib/readline/histfile.c - read_history_range: implement a heuristic that temporarily sets the history comment character if the first line read from the history file looks like it has is a timestamp (#[:digit:]) so we can read timestamps from the history file properly. Originally reported back in March 2015 by Christoph Anton Mitterer - read_history_range: make sure history_multiline_entries is non-zero if the history file looks like it has timestamps - read_history_range: if we think we have a history file with timestamps and we read more than one non-timestamp consecutive history lines, assume they are part of a single multi-line history entry and paste them together using _hs_append_history_line. Feature most recently requested by james harvey , also suggested by Christoph Anton Mitterer examples/loadables/setpgid.c - setpgid: new loadable builtin, originally contributed by Jason Vas Dias 12/29 ----- bashhist.c - bash_history_inhibit_expansion: fix cases where a history expansion should be skipped because it's in a command or process substitution but there is another history expansion preceding the substitution on the command line. Don't let the previous history expansion fool the function into saying the command substitution history expansion should be performed builtins/evalstring.c - parse_prologue: always unwind-protect history_expansion_inhibited, since history expansion can be enabled in non-interactive shells, and calling eval once in a non-interactive shell inhibits history expansion forever even if `set -o histexpand' was run before the eval builtins/common.h - SEVAL_NOHISTEXP: new flag for parse_and_execute/parse_string; means to not perform history expansion (decouple from SEVAL_NOHIST, which now means to not remember commands on history); changed all callers that had SEVAL_NOHIST to have SEVAL_NOHIST|SEVAL_NOHISTEXP builtins/evalstring.c - parse_prologue: instead of calling bash_history_disable, set remember_on_history to 0 if SEVAL_NOHIST and history_expansion_inhibited to 1 if SEVAL_NOHISTEXP 12/30 ----- subst.c - skip_to_histexp: new function, a stripped-down version of skip_to_delim. Used to skip to the next unquoted instance of the history expansion character, handles peculiar quoting and command/process substitution requirements. Better fix for bug reported by Zigmund.Ozean@zig-home.localdomain back in January, prompted by report from Keith Thompson bashhist.c - bash_history_inhibit_expansion: use skip_to_histexp instead of skip_to_delim subst.c - parameter_brace_expand_rhs: if the rhs of an expansion is "$@" and IFS is null, we need to separate the (quoted) positional parameters in the returned word with a space, and mark the word as needing to be split on spaces (W_SPLITSPACE). Fix for issues reported back in October 2014 as the result of an austin-group discussion, and just re-reported by Martijn Dekker 1/4/2016 -------- execute_cmd.c - execute_simple_command: if autocd is set, invoke a function named `cd' if one exists, instead of the shell builtin. Feature requested by transl8czech@gmail.com builtins/mkbuiltins.c - if a command's short description is the same as its name (e.g., `true'), don't mark the short doc to be translated. Report and fix from Benno Schulenberg 1/6 --- subst.c - command_substitute,process_substitute: before replacing the file descriptor underlying stdout (fd 1), make sure to purge any pending stdio output that hasn't been written successfully, even after a call to fflush(). Fixes bug reported by cks@cs.toronto.edu 1/7 --- builtins/{echo,printf}.def - echo_builtin,printf_builtin: don't use terminate_immediately; use calls to QUIT in the body of the print loop after writes and flushes. Fixes problem with running the signal handler and exit trap in a signal context and other bug reported by cks@cs.toronto.edu builtins/common.c - sh_chkwrite: put in calls to QUIT to catch signals that interrupt writes shell.c - get_current_user_info: protect endpwent() with #ifdef HAVE_GETPWENT. Fixes bug reported by pb 1/8 --- lib/readline/bind.c - _rl_init_file_error: now a varargs function so it can take format strings and arguments and pass them to vfprintf - rl_parse_and_bind: print a warning if we encounter a key binding string with one or more hyphens but we don't find a valid modifier (`control', `meta', etc.). Prompted by a report from Andrew Kurn - rl_parse_and_bind: improve several existing error messages now that _rl_init_file_error takes a variable number of arguments - rl_variable_bind: print error message upon encountering unknown variable 1/10 ---- lib/readline/bind.c - rl_parse_and_bind: if a `bare' keybinding is supplied without any terminating `:' or whitespace separating it from the command to be bound, signal an error 1/11 ---- subst.c - process_substitute: when compiled without job control (JOB_CONTROL not defined), make sure we call stop_making_children like we do in the job control code branch, so already_making_children is reset to 0. This is what command substitution does. Fixes bug reported by Alastair Hughes 1/12 ---- execute_cmd.c - execute_in_subshell: if a user subshell (`(command)') or other shell compound command has an input pipe, note for later by setting stdin_redir. This will inhibit the implicit redirection of standard input from /dev/null for async commands executed as part of this subshell. Fixes bug reported by Martin D Kealey - execute_simple_command: if the shell forks to execute this command because it has an input pipe, set stdin_redir to inhibit later redirection from /dev/null for async commands 1/13 ---- execute_cmd.c - HASH_BANG_BUFSIZE: use #define for size of buffer used in READ_SAMPLE_BUF instead of straight constant 80 - HASH_BANG_BUFSIZ: increase to 128 because Linux allows 128 chars in exec header. Reported by Ludovic Courtès configure.ac - remove support for purify and purecoverage - change release level to `rc1' 1/15 ---- builtins/enable.def - dyn_load_builtin: fix missing argument to builtin_error if load function returns an error. Fix from Aharon Robbins 1/18 ---- builtins/history.def - histtime: display a better error message for invalid timestamps; use existing `??' string for empty/missing timestamps or invalid history entries. Suggestion from Reuben Thomas 1/21 ---- lib/readline/text.c - rl_insert: when trying to optimize typeahead, make sure to reset the key sequence and key sequence length when we read a character that is not bound to self-insert. Bug report from Geir Hauge 1/25 ---- doc/{bash.1,bashref.texi} - word splitting: add example showing how -d'' is equivalent to -d after word splitting and null argument removal 1/27 ---- doc/{bash.1,bashref.texi} - read: modify description to make it more clear that the line is split using the same rules as word splitting, and use `delimiters' instead of `separators', which people read the wrong way 2/2 --- variables.c - initialize_shell_variables: allow FUNCNEST to set funcnest_max when imported from the environment. Inspired by https://bugzilla.redhat.com/show_bug.cgi?id=1274553 2/11 ---- bashhist.c - bash_history_inhibit_expansion: function should be compiled in only if BANG_HISTORY is defined. Report from isabella parakiss [bash-4.4-rc1 frozen] 2/15 ---- lib/readline/text.c - rl_refresh_line: call rl_redraw_prompt_last_line instead of rl_forced_update_display to avoid redrawing all lines of a multiline prompt (overwriting the last line of the multiline prompt in the process). Report from Hugh Davenport 2/18 ---- subst.c - parameter_brace_expand: when processing ${!name[@]}, make sure to free `name' before returning the list of keys to avoid a memory leak. Fixes bug reported by Emilio PastorMira 2/19 ---- trap.c - free_trap_strings: when freeing the `special' traps (NSIG to BASH_NSIG), check whether or not the `signal' is trapped, as it would be if the subshell inherited it (errtrace) and don't free the trap string in that case. Fixes bug reported by Jan Klötzke 2/21 ---- lib/sh/netconn.c - isnetconn: return false if getpeername fails with errno == EBADF. Bug and fix from Andrew Gregory builtins/shopt.def - parse_bashopts: when reading BASHOPTS from the environment, make sure to call any set functions associated with a variable, instead of just setting the value to 1. Report and fix from Vehlow, Jörg 2/22 ---- lib/readline/complete.c - rl_display_match_list: don't bother calling strrchr to check the return value from printable_part() if rl_filename_completion_desired is non-zero, since in that case printable_part just returns its argument. Fixes git completion issue reported by Dilyan Palauzov - _rl_colored_completion_prefix: default is now 0, testing is complete 2/23 ---- variables.c - assign_hashcmd: if the shell is restricted, reject attempts to add pathnames containing slashes to the hash table, as the hash builtin does. Fixes bug reported to savannah by Sylvain Beucler as https://savannah.gnu.org/support/?108969 2/24 ---- subst.c - NEED_FPURGE_DECL: need it before externs.h included, since fpurge is now used in this file. Fixes warning reported by Mike Frysinger - skip_double_quoted: add missing argument to declaration parse.y - shell_getc: in call to internal_warning, use %lu to print an argument cast explicitly to unsigned long. Fixes warning reported by Mike Frysinger 2/27 ---- parse.y - decode_prompt_string: quote result of \s expansion in case shell has been invoked with a pathname like '$(id)'. Inspired by report from Travis Garrell - decode_prompt_string: quote result of \h and \H expansions in the (unlikely) case that the hostname has been changed to something malicious including a command substitution. From report from Travis Garrell builtins/getopts.def - getopts_unbind_variable: new static function, just calls unbind_variable for now. Placeholder for any future work to do something special if OPTARG is readonly. Based on discussion started by Robert Elz lib/readline/vi_mode.c - rl_vi_unix_word_rubout: new function, implements ^W as Posix specifies it for vi mode, using whitespace and punctuation characters as word boundaries. Prompted by question from Carlos Pita lib/readline/vi_keymap.c - bind rl_vi_unix_word_rubout to ^W by default in both insert and command mode lib/readline/rltty.c - _rl_bind_tty_special_chars: if in vi mode, bind the WERASE character to rl_vi_unix_word_rubout lib/readline/funmap.c - vi-unix-word-rubout is new bindable name for rl_vi_unix_word_rubout lib/readline/readline.h - rl_vi_unix_word_rubout: new extern declaration 2/28 ---- doc/bash.1 - document readline blink-matching-paren variable - document readline completion-display-width variable - document readline completion-map-case variable - note that the default value of history-size depends on the value of the HISTSIZE variable. All from a message from Branden Robinson doc/bash.1,lib/readline/doc/{readline.3,rluser.texi} - convert-meta: modify the description to note that readline will turn the default to off if the locale contains eight-bit characters - input-meta: modify the description to note that readline will turn the default to on if the locale contains eight-bit characters - output-meta: modify the description to note that readline will turn the default to on if the locale contains eight-bit characters. From a message from Branden Robinson 3/6 --- examples/loadables/Makefile.in - install-dev: make installdirs a prerequisite; don't rely on the install-supported target to create it first. Bug with parallel install (`make -j 4 install') reported by Chris Staub 3/12 ---- lib/readline/display.c: - rl_clear_visible_line: clear all screen lines occupied by the current visible readline line. Inspired by report from Lauri Ranta lib/readline/readline.h - rl_clear_visible_line: extern declaration lib/readline/doc/rltech.texi - rl_clear_visible_line: add documentation 3/19 ---- findcmd.c - file_status: add inadvertently-dropped test for EXECIGNORE to the `other' bits check. The check was in the original patch - search_for_command: if the file should not be considered executable because it's in the EXECIGNORE blacklist, don't set file_to_lose_on, since the attempt to execute the last-ditch path will circumvent the EXECIGNORE setting. Fixes bug reported by Dennis Williamson 3/28 ---- lib/readline/display.c - rl_redisplay: when computing where the cursor should go (and the physical cursor position) after displaying the prompt, don't adjust the cursor position by wadjust if there are no multibyte chars in the prompt. lpos is already a physical cursor position; _rl_screenwidth is the physical screen width, and we're not calling _rl_col_width and don't need to take invisible characters into account again. Fixes bug reported by Andreas Schwab 4/1 --- parse.y - parse_comsub: when we read `esac' followed by a break character, turn on the LEX_RESWDOK flag because we can now read another case statement or other compound command. Fixes bug reported by Christian Franke , original version of patch from Piotr Grzybowski - parse_comsub: leave the reserved-word-ok flag set for four-char reserved words (then/else/elif/done/time) that can be followed by reserved words; set lex_rwlen to 0 so we know to start reading a new one 4/7 --- builtins/{alias,complete,jobs}.def - {alias,compopt,disown}: add missing `CASE_HELPOPT;' so --help option prints help longdoc instead of builtin_usage() 4/8 --- arrayfunc.c - assign_array_var_from_word_list: use bind_array_var_internal instead of inline code, so value conversions happen on assignment (case modification, arithmetic evaluation, etc). Fixes bug reported by Geir Hauge 4/9 --- shell.c - main: change the logic that determines whether or not bash is running under an emacs terminal emulator so it works with future versions of emacs, which will change to use only $INSIDE_EMACS. Report and fix from Paul Eggert lib/glob/sm_loop.c - EXT: if we see ?(pat), we can match 0 or 1 times, so if we don't match, we should consume the extglob pattern and go on to attempt to match the rest of the pattern against the string. Fixes bug reported by Isabella Parakiss - GMATCH: in case `*', don't short-circuit and match anything if the * is the last character in the pattern until after we check the special cases for matching a `.'. Fixes bugs reported back on 3/23 by Isabella Parakiss 4/13 ---- bashhist.c - hc_erasedups: fix memory leak by freeing history entry returned by remove_history. Bug and fix from Seiichi Ishitsuka 4/19 ---- config-top.h - CD_COMPLAINS: now settable in config-top.h; enables (fatal) error messages if cd is given multiple directory arguments 4/20 ---- execute_cmd.c - coproc_setvars: refuse to create variables with names that are not valid identifiers. Suggestion from Piotr Grzybowski rltty.c - rl_tty_set_echoing: interface to set _rl_echoing_p, returning old value. Inspired by gdb use of _rl_echoing_p when switching between terminal and curses modes described by Doug Evans bashline.c - bash_execute_unix_command: use rl_clear_visible_line instead of erasing current line on the screen, inspired by same report from 3/12. Still tentative change configure.ac - update release status to rc2, may skip this and go directly to release 4/21 ---- arrayfunc.c - convert_var_to_array: make sure that the newly-converted array variable is not marked as being an associative array. Fixes bug reported by Grisha Levit - convert_var_to_assoc: make sure that the newly-converted associative array variable is not marked as being an indexed array 4/22 ---- builtins/evalstring.c - should_suppress_fork: don't suppress the fork if there are any traps set, since that requires that we hang around to react to a signal or collect the command's exit status and run something. Fixes bug reported by Brian Vandenberg histexpand.c - history_tokenize_word: handle >| as a single token. Fix from Piotr Grzybowski from a report from idallen@idallen-fibe.dyndns.org 4/24 ---- execute_cmd. - coproc_setvars: don't overwrite readonly variables used as coproc names. From a report from Grisha Levit histexpand.c - history_tokenize_word: handle strings of digits before redirections beginning with `<' or `>' as part of the redirection word; handle strings of digits following `<&' or `>&' as part of the redirection word. Inspired by patch from Piotr Grzybowski from a report from idallen@idallen-fibe.dyndns.org lib/readline/complete.c - rl_display_match_list: if the common prefix is longer than any of the possible matches, set the length of the common prefix to 0 so the entire match gets printed for each match - rl_display_match_list: make sure to output at least one space between each displayed match, even if the displayed length is longer than our computed max - fnprint: if the length of the prefix (prefix_bytes) is greater than or equal to the length of the string to be printed (print_len), make sure to set the prefix length to 0 so the entire string is printed. From a report from Grisha Levit 4/25 ---- subst.c - command_substitute: update the conditions under which we give the terminal to pipeline_pgrp with give_terminal_to to the same ones where wait_for uses to decide whether to give the terminal back to shell_pgrp. This code exists to undo the work wait_for does; it has to give the terminal back to pipeline_pgrp only under those conditions when wait_for gives it back to the shell pgrp. Fix for bug reported by Paulo Bardes 4/26 ---- bashline.c - bash_filename_stat_hook: temporarily disable the `nounset' shell option around calls to expand_prompt_string so we don't get error messages during completion. Fixes issue reported by Eric Pruitt 4/27 ---- doc/{bash.1,bashref.texi} - extdebug: clarify that having this option enabled at shell startup acts identically to --debugger. From a report from Grisha Levit jobs.[ch] - wait_for_single_pid: now takes additional `int flags' argument {jobs,execute_cmd}.c,builtins/wait.def - wait_for_single_pid: changed callers to add extra argument jobs.c - wait_for_single_pid: if (flags & 1) == 0, don't print the error message if PID isn't found; changed execute_pipeline call when lastpipe is set 4/28 ---- general.c - bash_tilde_expand: try not setting interrupt_immediately or terminate_immediately; see what happens with networked password databases 4/29 ---- subst.c - parameter_brace_expand, parameter_brace_expand_rhs: now take an additional `pflags' argument from its caller so we can pass state - parameter_brace_expand_rhs: if expand_string_for_rhs returns a quoted null, but l_hasdollat is set to 1, meaning we saw a quoted "$@" of some form, we need to turn off special handling of "$@" so something like "${@-${@-$@}}" expands to an empty string like Posix says it should. Fixes bug reported by Grisha Levit 5/1 --- variables.c - bind_variable_internal: if we have a nameref variable with a valid array reference that is invalid for assignment (e.g., a[*]), and assign_array_element returns NULL, short-circuit and return NULL. Fixes bug reported by Grisha Levit general.[ch] - valid_nameref_value: new function, return 1 if passed argument is a valid variable name argument for a nameref variable: a valid identifier, a valid array reference, or a valid positional parameter. Second argument indicates whether the value is to be used for an assignment; in this case, return an error if the name consists of all digits builtins/declare.def - declare_internal: disallow values for nameref variables that don't pass the tests in valid_nameref_value. Part of fix for bug reported by Grisha Levit variables.c - bind_variable_internal: if trying to assign a value to a nameref variable, throw an error if valid_nameref_value fails (with a second argument of 1). More fixes for bug reported by Grisha Levit 5/2 --- builtins/declare.def - declare_internal: don't allow nameref attribute to be applied to an existing readonly variable; it can be used to circumvent readonly value. Fix for bug reported by Grisha Levit variables.[ch] - find_variable_last_nameref, find_global_variable_last_nameref: takes a new flags argument. If flags is non-zero, and we find a nameref variable that is unset (has no value), return that variable instead of NULL. {variables,execute_cmd,arrayfunc,subst}.c,builtins/{declare,set}.def - find_variable_last_nameref, find_global_variable_last_nameref: change callers to add flags argument (initially 0) 5/3 --- variables.c - bind_variable_internal: make sure we run the value of a namref variable through valid_nameref_value before doing the assignment. This can go away if we stop allowing namerefs to refer to positional parameters execute_cmd.c - execute_select_command: if bind_variable returns NULL, handle it as an error, as if attempting assignment to a readonly variable - execute_for_command: use 1 as second argument to find_variable_last_nameref, so it will return an invisible nameref to assign for each variable in the list - execute_for_command: if the current word in the list is an invalid identifier and the loop variable is a nameref, handle it as an error, as if attempting assignment to a readonly variable 5/4 --- variables.c - bind_int_variable: add warning if assigning an int to a nameref variable - assign_in_env: if assigning a variable in the temporary environment that is a namref, throw an error if trying to assign an invalid value - bind_variable_value: use valid_nameref_value instead of calls to legal_identifier and valid_array_reference execute_cmd.c - coproc_setvars: if we have a nameref variable as the coproc name, but it can't be assigned to for some reason, remove the nameref attribute and issue a warning that we are doing so - coproc_setvars: don't allow assignment to a nameref variable with an invalid variable name, using same code as find_or_make_array_variable general.c - valid_nameref_value: disallow values consisting solely of digits, like the positional parameters builtins/declare.def - declare_internal: don't call valid_nameref_value on the existing value of a variable that's being converted to a nameref -- that value is going to be thrown away and we've already checked the new value - declare_internal: call bind_variable_value with the ASS_NAMEREF flag if we're assigning to an existing nameref variable (for instance, `declare -n x ; declare x=42') 5/5 --- variables.c - make_local_variable: if we are being asked to create a local copy of a nameref variable from a prior scope (checked by looking up using find_variable_noref), use the nameref variable, not the variable it references (which we find using find_variable). Bug report from Grisha Levit ; fix by Piotr Grzybowski builtins/declare.def - declare_internal: if asked to do the same thing as above (difference is declare -n r=foo in previous case; declare -n r in this one), don't follow reference var from a previous context. Create a new local variable (using make_local_variable) and inherit attributes from previous scope's nameref as above 5/6 --- builtins/declare.def - declare_internal: if we are assigning a value to an existing invisible nameref variable, validate value early so we don't end up removing the variable after the assignment fails builtins/shopt.def - toggle_shopts: don't call set_bashopts here unless $BASHOPTS exists, since we don't want to set it up before it gets read from the environment. Keeping BASHOPTS read-only should keep users from unsetting it. Fix from Grisha Levit - shopt_set_debug_mode: make sure error_trace_mode reflects the setting of extdebug. This one is tentative. Fix from Grisha Levit - shopt_set_debug_mode: call set_shellopts after setting error_trace_mode or function_trace_mode. Fix from Grisha Levit shell.c - start_debugger: make sure that function_trace_mode and error_trace_mode reflect the value of debugging_mode, so they are both disabled if we disable debugging mode because we can't find the debugger start file. Fix from Grisha Levit - start_debugger: set $SHELLOPTS and $BASHOPTS to the option values are reflected if we change or set them here. Fix from Grisha Levit builtins/set.def - reset_shell_options: add reset values for: posix, nolog, pipefail. Fix from Grisha Levit builtins/shopt.def - reset_shopt_options: add reset values for: autocd, checkjobs, complete_fullquote, dirspell, failglob, globstar, gnu_errfmt, huponexit, inherit_errexit, interactive_comments, lastpipe, no_empty_cmd_completion, progcomp. Fix from Grisha Levit - reset_shopt_options: make sure the following options have the right defaults based on #defines: checkjobs, checkwinsize, direxpand, extglob, globasciiranges. Fix from Grisha Levit flags.c - reset_shell_flags: add missing flags, re-initialize history_expansion based on STRICT_POSIX 5/8 --- lib/readline/histexpand.c - history_expand: if any expansion returns 1 from history_expand_internal, indicating that :p was supplied, make it apply to the entire line like tcsh and csh. Fixes bug reported by Dean Stanton variables.[ch] - ifsname: move define to variables.h execute_cmd.c - execute_for_command: check whether or not the word in the for command is IFS; call setifs every time the variable is assigned a value. Fixes bug report from Grisha Levit 5/16 ---- general.c - valid_nameref_value: explicitly handle name == 0 and *name == 0 right away to save calls to legal_identifier builtins/declare.def - declare_internal: if trying to set nameref attribute on a variable with a null value, flag as invalid assignment, just as if running `declare -n r=""'. Fixes bug report from Grisha Levit - declare_internal: when creating a variable with a temporary value (""), use ASS_FORCE in the assignment flags to bypass name validity checks on nameref variable names variables.c - bind_variable_internal: don't call valid_nameref_value if we're forcing assignment with (flags & ASS_FORCE) builtins/read.def - read_builtin: handle bind_variable returning NULL when setting REPLY. Report and fix from Grisha Levit builtins/setattr.def - set_var_attribute: handle bind_variable returning NULL lib/sh/tmpfile.c - sh_seedrand(): call srandom() to seed the random number generator - sh_mktmpname, sh_mktmpfd: if we have random(), assume we have srandom() and seed the random number generator. Still looking for better value to see random number generator with variables.c - check_unbind_variable: use internal_error instead of builtin_error {execute_cmd,variables}.c - fix some places where bind_variable could return NULL and it was not checked. Fix from Piotr Grzybowski 5/17 ---- arrayfunc.c - convert_var_to_{array,assoc}: make sure to turn off nameref attribute, since namerefs can't be arrays execute_cmd.c - coproc_setvars: if the coproc name names a nameref, resolve the nameref and use it as the name of the coproc. Suggested by Grisha Levit subst.c - command_substitute: don't bother calling QUIT after calling reset_signal_handlers in the child, kill the child with SIGINT in case we just reset the signal handler to SIG_DFL and we should just exit instead of throwing to top level. Fixes bug reported by Grisha Levit 5/18 ---- variables.c - find_variable_nameref_for_create: find a nameref variable whose value doesn't resolve to an existing variable and see whether or not that value is appropriate for a new variable to be created - find_variable_nameref_for_assignment: find a nameref variable whose value doesn't resolve to an existing variable and see whether or not that value is appropriate for a new variable to be created. Difference between _assignment and _create is that _assignment allows the nameref value to be a subscripted array reference builtins/setattr.def - set_var_attribute: if variable lookup doesn't return anything, check for a nameref and make sure that any reference value is something we should be working on here, using find_variable_nameref_for_create(). If it fails, error out and return, otherwise it's probably a reference to a variable that hasn't been set yet, so let bind_variable take care of that. Report from Grisha Levit arrayfunc.c - bind_array_variable: if find_shell_variable returns NULL, check for a nameref using find_variable_nameref_for_create and create a new array variable with the value if it returns a valid nameref variable. Makes `unset var; declare -n ref=var ; ref[0]=foo' work right. Report from Grisha Levit 5/19 ---- variables.[ch] - unbind_variable_noref: unset a variable NAME without following any nameref chain. If the first instance of the variable with that name is a nameref, just unset that nameref variable. builtins/getopts.def - getopts_unbind_variable: if OPTARG is going to be unbound, use unbind_variable_noref to unbind that name even if it is a nameref variable. Issue raised by Grisha Levit execute_cmd.c - coproc_unsetvars: use unbind_variable_noref in case someone sets the coproc _PID variable as a nameref pointing to something read-only or strange. Issue raised by Grisha Levit builtins/set.def - set_ignoreeof: use unbind_variable_noref to unset "IGNOREEOF" and "ignoreeof" - set_posix_mode: use unbind_variable_noref to unset "POSIXLY_CORRECT" variables.c - make_vers_array: use unbind_variable_noref to unset "BASH_VERSINFO" lib/sh/shmatch.c - sh_regmatch: use unbind_variable_noref to unset "BASH_REMATCH" bashline.c - bash_execute_unix_command: use check_unbind_variable to unset READLINE_LINE and READLINE_POINT pcomplete.c - unbind_compfunc_variables: use unbind_variable_noref to unset COMP_LINE, COMP_POINT, COMP_WORDS, COMP_CWORD, COMP_TYPE, COMP_KEY - gen_shell_function_matches: use unbind_variable_noref to unset COMPREPLY config-top.h - USE_MKTEMP/USE_MKSTEMP: define by default to use libc version of mktemp and mkstemp in lib/sh/tmpfile.c. Recommended by Mike Frysinger to fix a FreeBSD problem configure.ac,config.h.in - mkstemp: check for mkstemp, define HAVE_MKSTEMP if available config-bot.h - USE_MKSTEMP: #undef if HAVE_MKSTEMP not defined 5/22 ---- variables.c - assign_in_env: if appending to a variable's value, make sure we call make_variable_value with the empty string if expand_assignment_string_to_string returns NULL, as do_assignment_internal does. Fixes bug with `str=''; val=foo ; val+=str printenv val' reported by Grisha Levit - assign_in_env: if assigning to a nameref variable in the temporary environment, and the nameref has a valid value for assignment (even if the target variable is not set), resolve the nameref and create a variable in the temporary environment named by the nameref's value. If the nameref variable is invisible or isn't set to a valid value for assignment, just create a regular temporary variable with the nameref's name. This provides a degree of ksh93 compatibility. Suggested by Grisha Levit - find_variable_nameref_context,find_variable_last_nameref_context: instead of returning NULL when detecting a loop, return a distinguished value: &nameref_maxloop_value - bind_variable: catch nameref_maxloop_value so we can do different things based on whether or not we get it. Right now we don't do anything different, but we could 5/24 ---- lib/readline/display.c - update_line,_rl_move_cursor_relative: save value of MB_CUR_MAX in a variable and use it throughout the functions instead of recomputing it every time (which possibly costs a function call) 5/25 ---- builtins/declare.def - declare_internal: only check whether the value being assigned to a nameref variable is a valid identifier if we're not appending to an existing value. Suggested by Grisha Levit - declare_internal: add a couple of more checks on nameref names: o make sure `declare -n x[3]=y' is an error o nameref variable self-references now produce a warning when in function scope o nameref variable self-references using subscripted arrays (a=a[0]) now produce warnings at function scope and errors at global scope 5/29 ---- builtins/declare.def - declare_internal: use name of readonly variable when calling sh_readonly instead of name passed as argument to declare in case we've followed a nameref chain - declare_internal: when turning off attributes on a nameref variable after following the nameref chain and turning them off on the target variable, don't allow the readonly attribute to be removed. This is primarily intended to turn off the nameref attribute. Issue pointed out by Grisha Levit - declare_internal: if we are attempting to modify an existing global nameref variable, and the -n option is supplied, operate on the nameref variable itself, not the variable it references doc/{bash.1,bashref.texi} - change the descriptions of namerefs and declare -n slightly to note that using the -n option to declare will operate on the nameref variable itself rather than the variable it references variables.c - bind_variable_internal: if trying to assign to a read-only variable, print the error message using the variable's name instead of the name passed to bind_variable in case we followed a nameref builtins/declare.def - declare_internal: when we're creating a (possibly invisible) variable as part of another operation, assigning NULL, make sure to use ASS_FORCE to avoid warning messages from any existing nameref variables we follow - declare_internal: allow the nameref attribute to be removed from a readonly nameref variable without a value but do not allow it to be removed from a readonly nameref variable that has a value, even if it doesn't reference an existing variable. This distinction is for ksh93 compatibility. Pointed out by Grisha Levit 5/31 ---- builtins/declare.def - declare_internal: if the call to bind_variable_value fails for some reason, make sure to restore the nameref attribute to flags_on and flags_off before calling NEXT_VARIABLE subst.c - make_internal_declare: handle += append op - shell_expand_word_list: when transforming assignment statement arguments to `declare', make sure to handle += append op to avoid passing invalid identifiers to declare. Report by Grisha Levit 6/1 --- builtins/declare.def - declare_internal: if a nameref assignment fails, only call delete_var to delete the variable if we created it in declare_internal in the first place general.c - check_selfref: new function, checks a NAME against a VALUE for nameref self-reference general.h - check_selfref: extern declaration builtins/declare.def - declare_internal: call check_selfref to determine whether a given NAME and VALUE constitute an invalid nameref variable self-reference variables.c - bind_variable_internal: call check_selfref to determine whether a given NAME and VALUE constitute an invalid nameref variable self- reference 6/2 --- parse.y - clear_shell_input_line: new function, clears contents of shell_input_line and sets index to 0, but doesn't free it externs.h - clear_shell_input_line: extern declaration builtins/evalstring.c - parse_and_execute: call clear_shell_input_line after setting input to string to be executed. Fixes problem with command substitution and multi-line aliases reported by Grisha Levit eval.c - parse_command: only execute PROMPT_COMMAND if the shell is not currently expanding an alias; use the same tests as parse.y:SHOULD_PROMPT and parse.y:prompt_again() use to decide whether or not to print a prompt. Fixes problems with PROMPT_COMMAND and multi-line aliases reported by Grisha Levit builtins/set.def - unset_builtin: changes to fix three problems reported by Grisha Levit : o if -n is supplied, we should not try to unset a function if a variable isn't found o unsetting namerefs whose values are array references does not work o unset -n n[0], where n is a nameref, would unset the referenced variable instead of `n' redir.c - redir_varvalue: handle case where nameref var points to subscripted array reference. Reported by Grisha Levit variables.c - bind_variable_value: make sure to call check_selfref only if aflags includes ASS_NAMEREF and not ASS_FORCE. Reported by Grisha Levit general.c - valid_nameref_value: now understands a FLAGS value of 2 to mean that the name will be used to create a variable, so only legal_identifier matters arrayfunc.c - find_or_make_array_variable: call valid_nameref_value with FLAGS value of 2 to indicate we will be creating a variable. Fixes mapfile issue reported by Grisha Levit 6/5 --- builtins/declare.def - declare_internal: only pass ASS_FORCE as part of assignment flags to assignments concerning arrays - declare_internal: when at the global scope, if we resolve a nameref and commit to using the new name, go back to to the beginning of the loop and use the new name in the checks and variable references. Make sure we construct the new name as a straight substitution of the nameref value into the old name, including array subscripts and rebuilding the correct values for `offset' and `value', since they are relative to the original value of name. Fixes several issues with checking use of subscripted array variables as nameref values - declare_internal: when calling assign_array_element, make sure to pass ASS_APPEND if aflags includes it, so things like declare -a var; var[1]=1; declare var[1]+=4 append to the value appropriately and var[1] ends up being `14' arrayfunc.c - valid_array_reference: make sure the array reference is properly terminated after the first subscript; return invalid if there is anything following the closing `]' 6/12 ---- variables.c - bind_variable_internal: if asked to perform an array subscript assignment on a nameref variable, display a warning and remove the nameref attribute (as bash-4.3 did). Reported by Dan Douglas 6/14 ---- test.c - posixtest: handle four-argument case where argv[1] == '(' and argv[4] == ')' by performing two-argument test on $2 and $3. Reported by Joerg Schilling 6/15 ---- variables.c - sv_ignoreeof: make sure ignoreeof is set only if IGNOREEOF is set and has a value (no `local IGNOREEOF'). Reported by Grisha Levit - sv_strict_posix: make sure posixly_correct is set only if POSIXLY_CORRECT is set and has a value. Reported by Grisha Levit 6/16 ---- builtins/declare.def - declare_internal: analogous to 6/12 change; when performing an array subscript assignment on a nameref, display a warning, remove the nameref attribute, and go on 6/21 ---- execute_cmd.c - shell_execve: make sure errno is (re)set appropriately before returning error value on exec format error 6/23 ---- subst.c - param_expand: when expanding $* in a context where splitting is not desired (pflags & PF_ASSIGNRHS), make sure to quote the word and the included separators. Reported by Dan Douglas 6/30 ---- configure.ac - release version is now `beta2' execute_cmd.c - execute_simple_command: even if already_making_children is non-zero, don't call wait_for if last_made_pid == NO_PID. Indirectly inspired by report from Nicolas Vivet 7/2 --- subst.c - expand_word_internal: when adding an IFS character to the accumulated string, don't quote it if word expansion is not going to be performed ((word->flags & (W_NOSPLIT|W_NOSPLIT2)) == 0) in the default case. This was already the case everwhere else, when we did `goto add_ifs_character;'. Fixes bug reported by Ibrahim M. Ghazal subst.[ch],arrayfunc.c - string_list_dollar_at: now takes a new third argument, currently reserved for future use; changed callers and declaration 7/7 --- [bash-4.4-beta2 frozen] 7/9 --- bashline.c - check_redir: make sure that index is > 0 before checking previous char; fixes out of bounds read reported by Hanno Böck 7/16 ---- subst.c - parameter_brace_expand: if ${array[@]} or ${array[*]} is supplied with an unset (or empty) array, and -u is enabled, don't make it an unbound variable error, like $@ or $* when there are no positional parameters. Based on report from Ian Allen lib/readline/signals.c - rl_pending_signal(): return the number of any signal readline has caught but is waiting to handle lib/readline/callback.c - rl_persistent_signal_handlers: new variable, restores readline-6.2 signal handler semantics: installed at rl_callback_handler_install time and removed with rl_clear_signals when a complete line has been read - _rl_callback_newline: call rl_set_signals if rl_persistent_signal_handlers is non-zero - rl_callback_read_char: install signal handlers with rl_set_signals if rl_persistent_signal_handlers is 0 (the default) - CALLBACK_READ_RETURN: uninstall signal handlers with rl_clear_signals only if rl_persistent_signal_handlers is 0 (the default) lib/readline/readline.h - rl_pending_signal: new extern declaration - rl_persistent_signal_handlers: new extern declaration 7/19 ---- execute_cmd.c - execute_while_or_until: make sure we decrement `continuing' the way we decrement `breaking' if the loop test indicates we will be breaking out of the loop. That keeps `until continue' from keeping `continuing' set to 1. Reported by Dan Douglas 7/20 ---- trap.c - run_pending_traps,_run_trap_internal: save and restore temporary_env, set temporary_env to 0 before trap runs so traps don't inherit a command's temporary env. Fixes bug with IFS assignment in tempenv and trap reported by Andriy Martynets jobs.c - run_sigchld_trap: save and restore temporary_env, set temporary_env to 0 before trap runs so traps don't inherit a command's temporary env (see above) 7/21 ---- execute_cmd.c - execute_function: unwind-protect function_line_number, in case we are executing a function inside another function. Reported by Grisha Levit in context of evaled functions and DEBUG traps - execute_arith_for_command,execute_arith_command,execute_simple_command, execute_cond_command: make sure line_number doesn't go < 0 when adjusting it by function_line_number 7/23 ---- subst.c,command.h - move the PF_ constants to command.h so other files can see them, now that PF_ASSIGNRHS has meaning to string_list_dollar_at arrayfunc.h - AV_ASSIGNRHS: new flag, indicate that ${a[@]} is being expanded on the RHS of an assignment statement arrayfunc.c - array_value_internal: if AV_ASSIGNRHS passed for ${a[@]}, pass PF_ASSIGNRHS to string_list_dollar_at to support args separated by spaces even if $IFS does not have the default value. Reported by Dan Douglas subst.c - string_list_dollar_at: if FLAGS argument includes PF_ASSIGNRHS, obey rules for $@ on the RHS of an assignment statement: expand to postitional parameters separated by spaces no matter what the first character of $IFS is. Reported by Dan Douglas - parameter_brace_expand_word: pass AV_ASSIGNRHS as flags value to array_value if PF_ASSIGNRHS is set and we are expanding an array variable subscripted by @ or * - param_expand: pass pflags to string_list_dollar_at in case it contains PF_ASSIGNRHS - expand_word_internal: if we have an assignment statement argument to a declaration builtin (W_ASSIGNARG), turn on W_ASSIGNRHS when we see the `=' to enable special $@ behavior - expand_word_internal: if W_ASSIGNARG enabled in word flags, pass that flag and W_ASSIGNRHS to recursive call to expand_word_internal when expanding double-quoted string; handles "$@" when IFS is not the default value and word splitting will not be performed subst.c - expand_word_internal: change case that handles '' (single-quoted empty string) to only discard it if we will not be performing word splitting (W_NOSPLIT|W_NOSPLIT2), since we need to add a quoted null argument if the subsequent characters will cause word splitting. This is how "" (double-quoted empty string) is handled after a bug fix back in August 2010. Reported by Grisha Levit , fix for Posix interp 888 7/27 ---- subst.c - param_expand: change fix from 6/23 (expanding $* in a context where word splitting is not performed) to make sure that $* expands to something before trying to quote the string. Bug and fix from Andreas Schwab lib/readline/bind.c - _rl_get_string_variable_value: fix a cut-and-paste error that caused the emacs mode string to be displayed for both vi insert and command mode strings. Report and fix from Steve Jones 7/28 ---- lib/readline/display.c - update_line: we can't use PROMPT_ENDING_INDEX unless we're testing against _rl_last_c_pos; if we are testing buffer indices, we need to use prompt_last_invisible directly. Fixes mode string redisplay issue with short prompt strings reported by Steve Jones 8/4 --- nojobs.c - without_job_control: new function, resets child state that's eventually used to decide whether or not to wait_for children execute_cmd.c - initialize_subshell: call without_job_control in both the JOB_CONTROL and no job control paths so we can reset the state in subshells that will not cause them to wait_for children inappropriately. Fixes bug reported by Dan Cross 8/5 --- configure.ac - release version is now `rc2' bashline.c - command_word_completion_function: if direxpand is set, make sure to run hint_text through bash_directory_completion_hook, since readline will do that and we want to be able to match the results readline returns. Fixes bug reported by Pat Somaru command.h - W_COMPLETE: new flag word value - PF_COMPLETE: new flags value for param_expand subst.h - SX_STRIPDQ: new flag, used for string_extract_double_quoted bashline.c - bash_filename_stat_hook,bash_directory_completion_hook: pass W_COMPLETE flag to expand_prompt_string subst.c - expand_word_internal: if W_COMPLETE in word->flags, pass PF_COMPLETE to param_expand - param_expand: if PF_COMPLETE in pflags, pass SX_COMPLETE to extract_command_subst. This allows `"$(echo ~/' to perform completion successfully - string_extract_double_quoted: turn third argument into a flags arg, old STRIPDQ is now set if flags&SX_STRIPDQ - string_extract_double_quoted: changed callers, if third argument was 1, pass SX_STRIPDQ instead - expand_word_internal: if W_COMPLETE in word->flags, pass SX_COMPLETE in flags to string_extract_double_quoted for case '"' - string_extract_double_quoted: if SX_COMPLETE in flags, pass it to extract_command_subst - expand_word_internal: if W_COMPLETE in word->flags, make sure W_COMPLETE is in temp word passed to recursive call to expand_word_internal to expand double-quoted string. Fixes bug reported by John Passaro 8/9 --- [bash-4.4-rc2 frozen] 8/10 ---- subst.c - sub_append_string: use size_t variable to store string length to avoid integer overflow. Report from Siteshwar Vashisht - sub_append_string: `size' argument is now pointer to size_t; changed istring_index in expand_word_internal to accommodate change lib/sh/ufuncs.c - include "posixselect.h" if HAVE_SELECT is defined, make sure that fd_set and the rest are defined 8/11 ---- lib/readline/histexpand.c - history_tokenize_word: use ISDIGIT instead of isdigit for more argument checking - "chardefs.h": include for ISDIGIT define and . Fix from Mike Frysinger lib/sh/eaccess.c,externs.h - sh_eaccess: first argument now const char * - sh_stataccess,sh_euidaccess: first argument now const char *, since sh_stat and sh_eaccess call them. Fix from Mike Frysinger lib/glob/glob.c - sh_eaccess: change local prototype general.[ch] - all_digits,legal_identifier,valid_nameref_value,importable_function_name, exportable_function_name,check_binary_file,file_exists,file_isdir, file_iswdir,same_file,make_absolute,valid_nameref_value: first argument (always identifier or filename) is now a `const char *'; some second filename arguments now `const char *'. Fix from Mike Frysinger configure.ac,config.h.in - uintptr_t: make sure it's defined to an integer type wide enough to hold a pointer lib/sh/tmpfile.c - sh_seedrand: use `uintptr_t' to cast pointer value to int value, instead of allowing truncation to take place, since srandom takes an `unsigned int'. Fix from Mike Frysinger arrayfunc.[ch] - valid_array_reference,array_variable_name,array_variable_part, array_value_internal,array_value,get_array_value: first argument (always variable name) now a `const char *' Fix from Mike Frysinger 8/19 ---- print_cmd.c - xtrace_print_word_list: if (xtflags & 2), don't try to quote the words in the LIST argument. Intended to be used in places where the list has not yet been expanded - xtrace_print_for_command_head: call xtrace_print_word_list with flags argument of 2. Fixes issue reported by Aharon Robbins - xtrace_print_select_command_head: call xtrace_print_word_list with flags argument of 2 8/21 ---- lib/sh/netopen.c - netopen: fix memory leak by freeing allocated path before error return. Fix from Dmitry Goncharov doc/{bash.1,bashref.texi} - change conditional commands description to make it clearer that bash doesn't emulate /dev/fd if the OS provides it 8/22 ---- braces.c - expand_amble: free partial and tem on failure to allocate memory for merged partial+result strvec. Fix from Dmitry Goncharov 8/23 ---- jobs.c - wait_for: don't call get_tty_state when a job finishes if we're running it as part of programmable completion. We test rl_readline_state. Fixes bug reported by Jaro Punta 8/24 ---- subst.c - pat_subst: fix memory leak (mstr). Fix from Dmitry Goncharov - split_at_delims: fix memory leak (d2). Fix from Dmitry Goncharov 8/26 ---- subst.c - match_upattern,match_wpattern: make sure that a trailing `*' is not escaped (preceded by an odd number of backslashes) before we append a `*' when checking whether a pattern matches anywhere in the string lib/readline/{input,signals}.c - protect instances of SIGQUIT with #ifdef SIGQUIT; apparently MinGW doesn't define it - protect instances of SIGHUP with #ifdef SIGHUP; apparently MinGW doesn't define it. From a report from Simon Sobisch xmalloc.o - allocerr: only try to find the break and report it if we're using the bash malloc; there's no guarantee that any other malloc is using sbrk() doc/{bash.1,bashref.texi} - quote removal: add paragraph mentioning quote removal to beginning of expansions section, referencing later paragraph. Report from Michel Lanne 8/30 ---- lib/readline/input.c - protect instances of SIGALRM and SIGWINCH with #ifdef; apparently MinGW doesn't define them. From a report from Simon Sobisch subst.c - SPECIAL_VAR: make sure name is not the empty string before going on, empty strings can result in unitialized reads - CHECK_STRING_OVERRUN: semi-general macro to use in *extract* and similar functions to to check whether a function called to parse an embedded construct (like a backquoted command substitution) doesn't consume the entire string because it's unterminated - skip_double_quoted: make sure other extraction functions don't consume the entire string (which they can if the construct is unterminated) before incrementing the index and going on. Uses CHECK_STRING_OVERRUN - extract_dollar_brace_string: check for unterminated embedded constructs consuming the entire string with CHECK_STRING_OVERRUN ('`') - skip_matched_pair: use CHECK_STRING_OVERRUN to make sure unterminated embedded "`", $(, ${ don't cause out-of-bounds reads [bash-4.4 frozen]