Dmitry Goncharov [Sun, 13 Nov 2022 21:20:33 +0000 (16:20 -0500)]
[SV 63333] Be more lenient when failing to create temporary files
If make cannot create a temporary lock file for output sync, continue
without output sync enabled rather than dying.
However, if make cannot store a makefile from stdin to a temporary
file that is still a fatal error.
* misc.c (get_tmppath): Keep running on failure to generate a
temporary file name.
(get_tmpfd): Keep running on failure to get a temporary file.
(get_tmpfile): Keep running on failure to open a temporary file.
Ensure memory is freed if we return an error.
* posixos.c (os_anontmp): Keep running on failure to open an
anonymous temporary file.
* output.c (setup_tmpfile): Print an error on failure to create an
output sync lock file.
* main.c (main): Die on failure to store makefile from stdin to a
temporary file.
* tests/scripts/features/output-sync: Add tests.
* tests/scripts/features/temp_stdin: Ditto.
Paul Smith [Mon, 7 Nov 2022 20:55:21 +0000 (15:55 -0500)]
Keep going if we can't connect to the jobserver
* src/posixos.c (jobserver_parse_auth): Don't invoke fatal() if we
can't connect to an existing jobserver: just keep going without it.
* src/w32/w32os.c (jobserver_parse_auth): Ditto.
* tests/scripts/features/jobserver: Add a test for invalid FIFO
auth files.
Paul Smith [Mon, 7 Nov 2022 18:08:11 +0000 (13:08 -0500)]
* src/dir.c (dir_contents_file_exists_p): Show dir name in error.
If we fail to read a directory show the directory name in the error
message. Pass struct directory instead of directory_contents to
allow that.
(dir_file_exists_p): Change dir_contents_file_exists_p caller.
(open_dirstrem): Ditto.
Paul Smith [Sun, 6 Nov 2022 19:14:28 +0000 (14:14 -0500)]
[SV 63315] tests: Simplify TERM signaling
Tests that try to kill the make process were not behaving as expected
on OpenBSD: the signal was sent from make to its children but the
sleep didn't die. Something odd about the way the shell treats TERM.
To reduce platform dependencies add "term" to the helper tool and run
that instead of kill / sleep.
* tests/thelp.pl: Add a new operation "term" that takes a PID.
* tests/scripts/features/output-sync: Use it.
* tests/scripts/features/temp_stdin: Ditto.
Paul Smith [Sat, 5 Nov 2022 16:56:52 +0000 (12:56 -0400)]
[SV 63315] Allocate function names when defining functions
* src/function.c (define_new_function): Don't keep a pointer to the
user-provided name of a user-defined function: if the .so is unloaded
it will point to garbage. Add the name to the strcache instead.
Paul Smith [Sat, 5 Nov 2022 16:46:29 +0000 (12:46 -0400)]
tests: Don't convert \ to / when checking regex's
When tests compare the output they will try converting backslashes
to slashes to see if that works. When we compare using regex's,
we can't do that because backslashes can escape special characters.
* tests/test_driver.pl (compare_output): Clean up this function.
(compare_answer_vms) [VMS]: Comparing answers on VMS is complex;
move all of it into its own function returning 0/1.
(compare_answer): A new function to compare answers: return 0/1.
Remember the CRLF->LF conversion forever; only check \ -> / when
we compare strings, not regex's.
Eli Zaretskii [Sat, 5 Nov 2022 14:46:58 +0000 (16:46 +0200)]
Fix build with Tiny C
* src/config.h.W32 (strtoll, strtoull) [__TINYC__]: Redirect to
_strtoi64 and _strtoui64, respectively. Reported by Christian
Jullien <eligis@orange.fr>.
Paul Smith [Mon, 31 Oct 2022 05:48:33 +0000 (01:48 -0400)]
Fix issues found by ASAN and Coverity
* tests/test_driver.pl: Preserve the LSAN_OPTIONS variable.
* tests/scripts/targets/ONESHELL: Don't set a local variable.
* tests/scripts/functions/let: Test empty let variable.
* src/posixos.c (osync_parse_mutex): Free existing osync_tmpfile.
* src/misc.c (get_tmpfd): Set umask() before invoking mkstemp().
* src/ar.c (ar_parse_name): Check invalid name (shouldn't happen).
* src/function.c (define_new_function): Free previous function entry
when replacing it with a new one.
* src/job.c (child_execute_job): Initialize pid for safety.
(construct_command_argv_internal): In oneshell mode ensure that the
returned argv has the right format (0th element is a pointer to the
entire buffer).
Paul Smith [Fri, 28 Oct 2022 17:03:18 +0000 (13:03 -0400)]
Avoid C99 constructs
Although gnulib requires C99, most of the code does compile with a
C90 compiler (perhaps with a lot of warnings). Reinstate our C90
configuration test, and clean up a few C99 things that crept in.
* src/job.c (construct_command_argv_internal): Don't use loop-local
variables or C++ comments.
* src/read.c (eval_makefile): Don't use loop-local variables.
Paul Smith [Fri, 28 Oct 2022 19:21:55 +0000 (15:21 -0400)]
tests: Avoid the use of File::Temp->newdir()
This was added in Perl 5.8 but some systems still only provide older
versions such as Perl 5.6. We don't really need it anyway.
Paul Eggert <eggert@cs.ucla.edu> reported this issue.
* tests/README: Update this to be a bit more modern.
* tests/test_driver.pl: Delete the $TEMPDIR variable.
* tests/scripts/features/temp_stdin: Use $temppath not $TEMPDIR.
Paul Smith [Fri, 28 Oct 2022 19:51:49 +0000 (15:51 -0400)]
[SV 62174] Force locale to be "C" before retrieving error messages
We attempt to do this with POSIX::setlocale() but apparently on some
systems (AIX) this isn't sufficient. So, in addition force the LC
environment variables to use "C".
Reported by Dmitry Goncharov <dgoncharov@users.sf.net>.
* tests/run_make_tests.pl: Move the global setup into set_default().
Force the %ENV locale variables to use the ones we'll use when running
make, then reset them back again after we find error messages.
Paul Smith [Tue, 25 Oct 2022 18:46:58 +0000 (14:46 -0400)]
* tests/scripts/features/exec: Don't test with C shells
Using C shells (csh, tcsh) with make is known to be problematic due
to incorrect ways it handles open file descriptors, at least. If
the user's shell is *csh then don't try it during exec tests.
Paul Smith [Tue, 25 Oct 2022 18:44:26 +0000 (14:44 -0400)]
Increase the test framework timeout from 5s to 60s
It seems that some of the test environments hit the 5s timeout on
some tests. Since it doesn't really matter, as long as we don't
hang forever, increase the timeout to 60s.
* tests/test_driver.pl: Increase $test_timout to 60. We don't need
to handle VMS timeouts specially anymore.
* tests/scripts/features/parallelism: We don't need to override the
default timeout anymore.
* tests/scripts/features/patternrules: Remove confusing comment.
Paul Smith [Mon, 24 Oct 2022 04:48:47 +0000 (00:48 -0400)]
[SV 63260] Don't recurse forever if setup_tmpfile() fails
If we fail during setup_tmpfile() we'll try to write an error, which
will invoke setup_tmpfile() again, etc. Avoid infinite recursion.
Original patch by Dmitry Goncharov <dgoncharov@users.sf.net>
* src/output.c (setup_tmpfile): Remember we're in this function and
return immediately if we enter it during recursion.
(message): Remember the starting location and use that instead of
fmtbuf.buffer.
(error): Ditto.
(fatal): Ditto.
Paul Smith [Sun, 23 Oct 2022 22:36:44 +0000 (18:36 -0400)]
Provide a maintainer-only debug method
Generating debug logs to stdout or stderr makes it impossible to
run tests etc. so create a dumb DBG facility to log to a temp file.
This exists only in maintainer mode and the DBG macro gives a
compile error if it's used in non-maintainer mode.
* src/makeint.h (DBG): Call dbg() in maintainer mode, else error.
(dbg): Define it in maintainer mode.
* src/misc.c (dbg): Open a log file for append, write, then close.
Paul Smith [Sat, 22 Oct 2022 23:42:07 +0000 (19:42 -0400)]
Set PATH_MAX on systems without a default value
Some systems (HURD) use fully-dynamic pathnames, with no limit.
We can't support this without significant effort so for now set
PATH_MAX to a large value.
* src/makeint.h: Set PATH_MAX to 4096 if not set and MAXPATHLEN
is also not set. Remove MAXPATHLEN setting: we won't use it.
* src/misc.c (get_path_max): If we can't get the path max via
pathconf() use the default PATH_MAX.
* src/dir.c (find_directory) [W32]: Use MAX_PATH not MAXPATHLEN.
(local_stat) [W32]: Ditto.
* src/job.c (create_batch_file) [W32]: Ditto.
* src/remake.c (name_mtime) [W32]: Ditto.
* src/w32/w32os.c (os_anontmp) [W32]: Ditto.
Paul Smith [Sat, 22 Oct 2022 23:09:44 +0000 (19:09 -0400)]
[SV 63098] Enhance detection of missing peer also-make targets
The previous attempt to detect missing peer targets for implicit
rules had some holes. Move the detection to notice_finished_file().
* src/remake.c (check_also_make): If we don't have the current mtime
for the file, obtain it.
(update_goal_chain): Don't call check_also_make() here.
(check_dep): Ditto.
(notice_finished_file): If we finished running an implicit rule that
has also_make targets, invoke check_also_make().
Paul Smith [Sat, 22 Oct 2022 21:23:17 +0000 (17:23 -0400)]
Collect failure details when the regression tests fail
* README.in: Add a section on running regression tests.
* Makefile.am (check-regression): Capture the test run output, and
on failure collect configure and test results into a tar file.
Paul Smith [Sat, 22 Oct 2022 19:35:16 +0000 (15:35 -0400)]
Enhance tests to work on different systems
The GNU platform testers reported a number of test errors on
different systems; try to address them.
* tests/thelp.pl: A number of tests timed out with a 4-second
timeout. Increase the default timeout to 10 seconds.
* tests/run_make_tests.pl: Executing directories on cygwin behaves
differently in Perl than make so skip these tests there.
* tests/scripts/options/symlinks: Check for the symlink feature
in make, rather than whether the system supports them.
* tests/scripts/features/implicit_search: On some systems "false"
exits with a different exit code. Use the helper instead.
* tests/scripts/features/loadapi: Ditto.
* tests/scripts/features/output-sync: Sleep before make -f bar in
the first test as well as the second one.
* tests/scripts/features/exec: Skip on cygwin, which seems to
be "UNIX" but where scripts don't run normally.
* tests/scripts/misc/fopen-fail: Skip on cygwin, where make
eventually exits with exit code 0 and no error messages.
Dmitry Goncharov [Fri, 21 Oct 2022 23:35:09 +0000 (19:35 -0400)]
[SV 63243] tests: Avoid SIGTERM racing with make error messages
Original patch from Frank Heckenbach <f.heckenbach@fh-soft.de>.
* tests/scripts/features/output-sync: Introduce a sleep to let make
write its error message. Some systems use different names for
SIGTERM so match with a regex.
* tests/scripts/features/temp_stdin: Ditto.
Dmitry Goncharov [Fri, 21 Oct 2022 22:54:39 +0000 (18:54 -0400)]
[SV 63248] Ignore SIGPIPE
Don't terminate when make's output is redirected to a pipe and the
reader exits early; e.g.:
$ echo 'all:; sleep 2' | make -f- -j2 -O |:
This lets us unlink temporary files, and tell the user that make was
not able to write its output.
Reported by Frank Heckenbach <f.heckenbach@fh-soft.de>.
Paul Smith [Tue, 18 Oct 2022 03:44:31 +0000 (23:44 -0400)]
Rework temp file handling to avoid GNU libc warnings
Original patch provided by Paul Eggert <eggert@cs.ucla.edu>.
GNU libc will generate a link-time warning if we use mktemp() even
though we are using it safely (we only use it with mkfifo()).
Avoid this and clean up some handling.
First, check all calls related to temporary files and exit with a
fatal error and a useful message if we can't obtain them. In some
situations it might be possible to continue with reduced capability
but it's not worth the effort.
On POSIX systems we can create anonymous temp files using O_TMPFILE
if it's supported, else if we're using the default location and we
have dup(2), we can use standard tmpfile() and get an FD from it.
If we need a named temp file and FILE* and we have mkstemp() we can
use that, else if we have fdopen() we can get a temp FD and open it.
If none of those are available all we can do is generate a temp name
then open it with fopen() which is not secure.
* src/makeint.h (get_tmpdir): Declare it for use elsewhere.
* src/misc.c (get_tmpdir): Make it public not static.
(get_tmptemplate): Simplify the implementation.
(get_tmppath): Only define this if we have to have it to avoid
warnings from GNU libc.
(get_tmpfd): Generate fatal errors on error.
(get_tmpfile): Ditto. Open files in "wb+" mode to match tmpfile().
Require a filename pointer (all callers want it).
* src/os.h (os_anontmp): Implement for posixos.c as well.
* src/posix.c (jobserver_setup): Don't use mktemp to avoid GNU libc
errors. Instead construct the FIFO name based on the PID.
(osync_setup): get_tmpfd() can't fail so don't check it.
(os_anontmp): If the system supports O_TMPFILE use it. If not, and
we want to create the temporary file in the default directory, we
can use tmpfile() then use dup() to copy the file descriptor.
* src/main.c (main): get_tmpfile() can't fail.
* src/vmsjobs.c (child_execute_job): get_tmpfile() can't fail.
Paul Smith [Tue, 18 Oct 2022 03:36:36 +0000 (23:36 -0400)]
Add ISDIRSEP() helper macro and use it
Create a ISDIRSEP() macro to check for directory separator chars
using the stopchar_map, and replace inline checks and explicit
STOP_SET calls with this macro.
Paul Smith [Mon, 17 Oct 2022 21:56:18 +0000 (17:56 -0400)]
Add preprocessor macros for different platforms
* configure.ac: MK_CONFIGURE shows config.h was generated by configure.
* src/config.ami: Define MK_AMIGAOS.
* src/config.h-vms: Define MK_VMS.
* src/configh.dos: Define MK_DJGPP.
* src/config.h.W32: Define MK_W32 and WINDOWS32.
* src/build_w32.bat: Let WINDOWS32 be defined by config.h. Remove
unused setting of WIN32.
* src/job.c: Clean up use of WIN32.
* src/main.c: Ditto.
* tests/scripts/features/default_names: Ditto.
Paul Smith [Mon, 10 Oct 2022 00:17:18 +0000 (17:17 -0700)]
Check for recipe line count overflow before it overflows
awk 'BEGIN {
print "x:"
for (i = 0; i < 65536; i++)
printf "\techo %d\n", i}
' | make -f -
Outputs only "make: 'x' is up to date." Larger values run only the
lines above 65536. Reported by Paul Eggert <eggert@cs.ucla.edu>.
* src/commands.c (chop_commands): Check the line count before it has
a chance to overflow. Use size_t for max count so it can't overflow.
Remove stray 'd' in diagnostic.
Paul Smith [Sun, 16 Oct 2022 19:13:41 +0000 (15:13 -0400)]
Remove template files to simplify distribution creation
The README templates were not useful since the replacement step
didn't have anything to replace: rename them.
Rather than creating template files for the config variants, create
mkconfig.h.in containg PACKAGE_* variables to be replaced, and have
config variant header files include it. Note on POSIX we don't use
this, and continue to generate a single config.h.in file.
Use config.status to convert the README.in and mkconfig.h.in files
during distribution creation.
Modify all users of VERSION to use PACKAGE_VERSION instead.
* configure.ac: Use GNU Make not GNU make as the package name.
* README.in: Use GNU Make not GNU make.
* README.git: Remove references to README.W32.template.
* .gitignore: Update for new behavior.
* Basic.mk.template: Remove unused posix_SOURCES and VERSION, and
references to unshipped mk/Posix.mk
* Makefile.am: Add src/mkconfig.h as an extra dist file.
* bootstrap.bat: Rewrite mkconfig.h.in to mkconfig.h
* maintMakefile: Remove obsolete template files; add mkconfig.h.in.
* prepare_vms.com: Rewrite mkconfig.h.in to mkconfig.h
* mk/VMS.mk: Fix incorrect header file prerequisite.
* src/mkconfig.h.in: New file containing PACKAGE variables.
* src/config.ami: Include mkconfig.h.
* src/config.h.W32: Ditto.
* src/configh.dos: Ditto.
* src/config.h-vms: Ditto.
* src/version.c: Use PACKAGE_VERSION not VERSION.
Juan M. Guerrero [Sat, 15 Oct 2022 23:14:49 +0000 (19:14 -0400)]
Support building with DJGPP 2.05
* builddos.bat: Use env var settings for paths. Fix a typo in
expand.o. Add the missing load.o compilation and link. Enable
Guile support. Copy Basic.mk from the correct location.
* src/configh.dos.templage: DJGPP supports strtoll() and ssize_t
given new enough versions. Set preprocessor variables to 1 not 0.
Paul Smith [Sat, 15 Oct 2022 22:11:21 +0000 (18:11 -0400)]
[SV 63215] Remember the random seed for shuffle mode
Basic fix provided by James Hilliard <james.hilliard1@gmail.com>.
Ensure we remember and propagate the random seed we generate during
shuffle mode. Also add a debug statement displaying the seed.
* src/shuffle.c (shuffle_set_mode): Init and save the randoms seed.
* src/misc.c (make_rand): Code cleanups.
* src/main.c (main): Show a debug message containing the seed.
Paul Smith [Sat, 15 Oct 2022 20:34:54 +0000 (16:34 -0400)]
[SV 63157] Ensure temporary files are removed when signaled
Original patch from Dmitry Goncharov <dgoncharov@users.sf.net>.
When handling a fatal signal ensure the temporary files for
stdin and the jobserver fifo (if in use) are deleted.
* src/makeint.h (temp_stdin_unlink): Declare a new method.
* src/main.c (temp_stdin_unlink): Delete the stdin temporary file
if it exists. If the unlink fails and we're not handling a signal
then show an error.
(main): Call temp_stdin_unlink() instead of unlinking by hand.
* src/commands.c (fatal_error_signal): Invoke cleanup methods if
we're handling a fatal signal.
* tests/scripts/features/output-sync: Test signal handling during
output sync and jobserver with FIFO.
* tests/scripts/features/temp_stdin: Test signal handling when
makefiles are read from stdin.
(jobserver_setup): Set job_root to true.
(jobserver_clear): if we're the root instance and we're using a
FIFO, unlink it. If we're not in a signal handler, free memory.
(jobserver_acquire_all): Call jobserver_clear().
(sync_root): Rename from sync_parent for consistency.
Paul Smith [Sun, 9 Oct 2022 17:41:21 +0000 (13:41 -0400)]
Set the proper type for variables set in signal handlers
* bootstrap.conf: Get gnulib's sig_atomic_t type checking M4 macro.
* configure.ac: Invoke it.
* src/makeint.h (handling_fatal_signal): Set the type correctly.
* src/commands.c (handling_fatal_signal): Ditto.
Paul Smith [Mon, 3 Oct 2022 19:06:12 +0000 (15:06 -0400)]
Clean up some warnings on Windows builds
* src/arscan.c (parse_int): Avoid conversion from int to char and
check for overflow given a max value.
(ar_scan): Check intmax sizes then cast to the proper type.
(ar_member_touch): Get proper return type from ar_scan and cast it
to off_t.
* src/function.c (a_word_hash_cmp): Don't cast from size_t to int.
(func_filter_filterout): Count using unsigned long to avoid casts.
(construct_include_path): Explicitly cast to void*.
* src/shuffle.c (reverse_shuffle_array): Use size_t index.
* src/w32/subproc/sub_proc.c (process_wait_for_multiple_objects):
Initialize the return value in case the count is 0.
Pacify 12.2.1 20220819 (Red Hat 12.2.1-2) -Wsign-compare by checking
the readbuf() return explicitly for errors then casting from ssize_t.
Initial patch provided by Paul Eggert <eggert@cs.ucla.edu>.
Paul Eggert [Sun, 2 Oct 2022 21:24:53 +0000 (17:24 -0400)]
Support systems with 32-bit long and 64-bit time_t
Don't assume that time_t fits in long, as some hosts (e.g.,
glibc x86 -D_TIME_BITS=64) have 32-bit long and 64-bit time_t.
* bootstrap.conf (gnulib_modules): Add largefile, to support files
with timestamps after Y2038 on hosts with 32-bit long.
* configure.ac: Do not call AC_SYS_LARGEFILE, as the largefile module
does that for us.
* src/makeint.h (PRIdMAX, PRIuMAX, SCNdMAX): Define if not already
defined (taken from gnulib).
* src/ar.c: Include intprops.h, for TYPE_MAXIMUM, as
INTEGER_TYPE_MAXIMUM does not work on time_t without issuing a bunch
of warnings.
(ar_member_date): Check that result is in time_t range.
(ar_member_date_1): Use intmax_t to hold the date.
(ar_glob_match): Ditto.
* src/arscan.c (VMS_function, VMS_function_ret, ar_scan)
(parse_int, ar_scan, ar_member_pos, ar_member_touch)
(describe_member): Convert long int to intmax_t.
* src/file.c (file_timestamp_sprintf): Use intmax_t/uintmax_t instead
of long/unsigned long for values that might be time_t.
* src/arscan.c (ar_member_touch): Fix buffer overrun if the timestamp
is too large.
Paul Smith [Sun, 2 Oct 2022 14:18:21 +0000 (10:18 -0400)]
[SV 63098] Temporarily revert the change to pattern rule behavior
The fix for SV 12078 caused a backward-compatibility issue with some
makefiles. In order to allow users to resolve this issue, revert
that change for this release cycle: it will be reinstated in the
next release cycle. Introduce a warning if we detect that the recipe
of a multi-target pattern rule doesn't create all the targets.
* NEWS: Announce the future backward-incompatibility.
* doc/make.texi (Pattern Intro): Describe the behavior and that it
will change in the future.
* src/remake.c (check_also_make): Check for also_make targets that
were not created and generate a warning.
(update_goal_chain): Call the new function.
(check_dep): Ditto.
(update_file_1): Defer implicit rule detection until after we check
all the also_make files (as it used to be).
* tests/scripts/features/patternrules: Add tests of the new warning.
Skip the tests for SV 12078.
Andreas Schwab [Sat, 1 Oct 2022 16:47:42 +0000 (12:47 -0400)]
[SV 63111] Ensure output-sync is cleaned on re-exec and recursion
Now that output-sync is using a separate file as a mutex rather than
stdout, ensure the new file descriptor is closed before re-exec and
not inherited by children.
Paul Smith [Mon, 26 Sep 2022 18:38:27 +0000 (14:38 -0400)]
Implement a simple xorshift 32bit random number generator
Avoid relying on the system random number generator for our random
shuffle, so that the same seed gives the same results on all systems.
This generator doesn't need to be amazing, just pretty good, so don't
bother with xorshift* or xorshift+, etc.
* src/makeint.h: Declare make_seed() and make_rand().
* src/misc.c (make_seed): Set the seed value for the RNG.
(make_rand): Return the next random number. If the seed was not set
initialize it first.
* src/shuffle.c (shuffle_set_mode): If we don't get a seed from the
user just leave it unset (0).
(shuffle_deps_recursive): Use make_seed() not srand().
(random_shuffle_array): Use make_rand() not rand().
[SV 63100] Set the floc of every created goal dep struct
* src/read.c (eval): Initialize the goaldep floc pointer.
* tests/scripts/features/loadapi: Verify that the floc is set after
unloading and reloading dynamic objects.
Paul Smith [Sun, 25 Sep 2022 20:12:21 +0000 (16:12 -0400)]
Provide new functions to convert long long to string
The previous attempt to use PRI* macros to avoid compiler-specific
printf format specifiers didn't work because we are using raw
long long type, not the uintX_t types. On systems where long and
long long are the same size, uint64_t might be type "long" and PRId64
is just "ld".
Instead write new functions that convert [unsigned] long long to a
string and call those instead.
* src/makeint.h: Declare make_lltoa() and make_ulltoa().
* src/misc.c (make_lltoa): New function that writes a long long value
into a provided buffer. Return the buffer for ease-of-use.
(make_ulltoa): Ditto, for unsigned long long.
* src/function.c (func_wordlist): Call these new methods. Also
rework the error strings so we share the translated string.
* src/dir.c (print_dir_data_base): Call the new methods instead of
using MSVC macros.
Paul Smith [Sun, 25 Sep 2022 17:29:59 +0000 (13:29 -0400)]
Always restore global environ if we use vfork
We may change the global environ variable in the child; when using
vfork() this also sets it in the parent. Preserve the parent's
environ in child_execute_job() so it takes effect for all callers.
Reported by Denis Excoffier <bug-tar@Denis-Excoffier.org>
Root cause found by Martin Dorey <Martin.Dorey@hitachivantara.com>
* src/job.c (start_job_command): Remove save/restore of the parent
environment.
(child_execute_job): Add save/restore of the parent environment,
if we use vfork().
* tests/scripts/functions/shell: Add a test the crashes if we don't
reset environ after we run $(shell ...).
Eli Zaretskii [Sat, 24 Sep 2022 18:15:41 +0000 (21:15 +0300)]
Avoid compilation warnings
* src/implicit.c: (pattern_search):
* src/main.c: (main): Avoid compilation warnings for variables
only used when compiling a debug version with assertions.
[SV 56301] Fail if an included makefile can't be built
Fail if a mandatory include file fails to be built even if it's
built as part of a grouped target where the other include file
is optional.
* src/main.c (main): If a makefile doesn't build set any_failed.
* tests/scripts/features/include: Add tests.
* tests/scripts/options/dash-k: Stop after include build failure.
Paul Smith [Tue, 20 Sep 2022 06:10:14 +0000 (02:10 -0400)]
[SV 12078, SV 62809] Rebuild grouped targets if any is missing
If any of a set of grouped targets is missing or out of date, even
if make is not trying to build that target, rebuild them all.
Ensure this is true for explicit grouped targets as well as pattern
rule grouped targets.
Original patch by Jonathan Gravel <jo@stashed.dev>
* src/remake.c (update_file_1): After matching any pattern rules,
go through the also_make targets and set noexist as needed. Also
compute the oldest this_mtime.
* tests/scripts/features/grouped_targets: Add regression tests.
* tests/scripts/features/patternrules: Ditto.
* tests/features/vpath: Rewrite to use modern run_make_test().
Add a test that we check for VPATH before implicit rule search.
Move the tests in vpath2 and vpath3 into this suite.
* tests/features/vpathplus: Rewrite to use modern run_make_test().
Paul Smith [Mon, 12 Sep 2022 22:18:49 +0000 (18:18 -0400)]
[SV 13862] Implement the .WAIT special target
The next version of the POSIX standard defines parallel execution
and requires the .WAIT special target as is implemented in some other
versions of make.
This implementation behaves similarly to others in that it does not
create a relationship between targets in the dependency graph, so
that the same two targets may be run in parallel if they appear as
prerequisites elsewhere without .WAIT between them.
Now that we support .WAIT it's trivial to also support prerequisites
of the .NOTPARALLEL special target, which forces the prerequisites of
those targets to be run serially (as if .WAIT was specified between
each one).
* NEWS: Announce the new .WAIT and .NOTPARALLEL support.
* doc/make.texi (Parallel Disable): A new section to discuss ways in
which parallel execution can be controlled. Modify cross-refs to
refer to this section.
* src/dep.h (struct dep): Add a new wait_here boolean.
(parse_file_seq): Add PARSEFS_WAIT to check for .WAIT dependencies.
* src/file.c (split_prereqs): Use PARSEFS_WAIT.
(snap_deps): If .NOTPARALLEL has prerequisites, set .WAIT between
each of _their_ prerequisites.
(print_prereqs): Add back in .WAIT when printing prerequisites.
* src/implicit.c (struct patdeps): Preserve wait_here.
(pattern_search): Ditto. Use PARSEFS_WAIT when parsing prereqs for
pattern rule expansion.
* src/read.c (check_specials): Don't give up early: remembering to
update these options is not worth the rare speedup.
(check_special_file): If .WAIT is given as a target show an error--
once--if it has prereqs or commands.
(record_files): Call check_special_file on each target.
(parse_file_seq): If PARSEFS_WAIT is given, look for .WAIT prereqs.
If we see one assume that we are building a struct dep chain and set
the wait_here option while not putting it into the list.
* src/remake.c (update_file_1): If wait_here is set and we are still
running, then stop trying to build this target's prerequisites.
* src/rule.c (get_rule_defn): Add .WAIT to the prerequisite list.
* src/shuffle.c (shuffle_deps): Don't shuffle the prerequisite list
if .WAIT appears anywhere in it.
* tests/scripts/targets/WAIT: Add a test suite for this feature.
[SV 63045] Reload each intact unloaded shared object
If makefile rules do not update an unloaded shared object, load it
again. Avoid double loading of the same object if the setup function
returns -1.
* src/filedef.h (struct file): Add "unloaded" flag.
* src/makeint.h (load_file): Take struct file *.
(unload_file): Return int.
* src/main.c (main): Reload unloaded shared objects if they weren't
updated.
* src/commands.c (execute_file_commands): Set "unloaded" and reset
"loaded" when a shared object is unloaded.
* src/read.c (eval): Set "loaded" and reset "unloaded" when a shared
object is loaded. Add successfully loaded files to the db.
* src/load.c (load_file): Check "loaded" to avoid double loading the
same object. Fix a memory leak of string loaded. Return -1, rather
than 1, if the object is already loaded. This fixes double loading of
the same object when the setup routine returns -1.
(load_object): Add a log message.
(unload_file): Return an error on dlclose failure. Log a message.
* tests/scripts/features/loadapi: Add new tests.
[SV 63047] Fix shuffle of SECONDEXPANSION prerequisites
Commit 07eea3aa4 `make --shuffle` prevented shuffling prerequisites
that use .SECONDEXPANSION, since shuffle happens before expansion.
This has two problems:
1. No shuffling happens for such prerequisites.
2. Use-after-free when outdated '->shuf' links are used.
Add a reshuffle into expansion phase right after dependency changes.
* src/file.c (expand_deps): Add reshuffle if dependencies change.
* src/shuffle.c (identity_shuffle_array): Fix comment typo.
* tests/scripts/options/shuffle: Add new SECONDEXPANSION test.
Paul Smith [Sun, 11 Sep 2022 14:56:08 +0000 (10:56 -0400)]
Support the MAKE_TMPDIR environment variable
Allow build systems to choose an alternative location for make to
store its temporary files.
* NEWS: Announce the new environment variable.
* doc/make.texi (Temporary Files): Provide documentation.
* src/misc.c (get_tmpdir): Split into a new function. Compute the
temporary directory and store it in a static location.
* tests/scripts/features/jobserver: Add a test of MAKE_TMPDIR.
[SV 63044] load: Update .LOADED if the setup function returns -1
* src/load.c (load_file): Update .LOADED if setup returns non-0.
* tests/scripts/features/load: Change the return value based on an
environment variable. Ensure that returning -1 still adds to
.LOADED. Also add a test that verifies that make doesn't try to
rebuild the loaded file if -1 is returned.