Eric Blake [Wed, 25 Apr 2012 02:08:06 +0000 (20:08 -0600)]
doc: fix bad @xref uses
Upstream gnulib maint.mk improvements caught a few issues we
should fix, as well as a few issues in files we copy from
other sources that we will just ignore here.
* doc/autoconf.texi (Generic Programs, Special Shell Variables)
(Limitations of Builtins): Use references correctly.
* cfg.mk
(exclude_file_name_regexp--sc_prohibit_undesirable_word_seq)
(exclude_file_name_regexp--sc_useless_cpp_parens): Add exemptions.
Bruno Haible [Sun, 22 Apr 2012 17:42:24 +0000 (19:42 +0200)]
AC_INIT: remove a transitional warning
On bi-arch systems (such as x86 / x86_64) it is often necessary to pass
the --host option together with an appropriate value for CC. But this
triggers a warning:
$ ./configure --host=i686-pc-linux-gnu CC="gcc -m32 -march=i586"
configure: WARNING: if you wanted to set the --build type, don't use --host.
If a cross compiler is detected then cross compile mode will be used
..
This warning was introduced on 2000-06-30, in commit
<http://git.savannah.gnu.org/gitweb/?p=autoconf.git;a=commitdiff;h=cb2e761b8e1181f97e8e09d85526bd22078433db>
with the remark "Re-enable the old behavior of --host and --build."
This warning was meant to warn users about a changed semantics of
--build and --host. This change is now 12 years in the past; users
have had enough time to learn it. I therefore suggest to remove the
warning.
I've done lots of cross and bi-arch compilations in the last 10 years,
all with --host and without --build, and have never observed a problem
with it, except for the warning. Simply relying on config.guess is sufficient.
* lib/autoconf/general.m4 (_AC_INIT_PARSE_ARGS): Don't warn if --host
given without --build.
Bruno Haible [Mon, 23 Apr 2012 19:43:04 +0000 (21:43 +0200)]
doc: fix incorrect and incomplete doc about cross-compilation mode
* doc/autoconf.texi (Runtime): Mention the effect of the cross-
compilation mode on AC_RUN_IFELSE.
(Specifying Target Triplets): Fix description of --host's effects.
(Hosts and Cross-Compilation): Remove incorrect statement about
--host's effects.
Paul Eggert [Mon, 23 Apr 2012 17:30:05 +0000 (10:30 -0700)]
doc: document --build and cross-compilation better
* doc/autoconf.texi (Specifying Target Triplets): Mention that
specifying a build-type that differs from host-type enables
cross-compilation. Problem reported by Bruno Haible in:
http://lists.gnu.org/archive/html/autoconf-patches/2012-04/msg00009.html
Paul Eggert [Wed, 7 Mar 2012 22:22:54 +0000 (14:22 -0800)]
tests: fix port of AT_CHECK_ENV to hosts with flaky grep
* tests/local.at (AT_CHECK_ENV): Don't copy the buggy grep's
diagnostics to stderr, as that causes AT_CHECK to fail. They can
be found in the stderr-* files if this is needed for debugging.
Paul Eggert [Wed, 7 Mar 2012 06:56:39 +0000 (22:56 -0800)]
tests: port AT_CHECK_ENV to hosts with flaky grep
* tests/local.at (AT_CHECK_ENV): Don't assume that if one grep
fails, the other will too. It could be that 'grep' is flaky,
and fails somewhat at random. This would explain the problems
reported for autoconf-2.68b on FreeBSD and MacOS X, for example:
<http://lists.gnu.org/archive/html/bug-autoconf/2012-03/msg00032.html>
<http://lists.gnu.org/archive/html/bug-autoconf/2012-03/msg00035.html>
<http://lists.gnu.org/archive/html/bug-autoconf/2012-03/msg00036.html>
<http://lists.gnu.org/archive/html/bug-autoconf/2012-03/msg00044.html>
Since the perl version required in Automake::Getopt has been
recently lowered from 5.6.2 to 5.6.0, this change has the nice
effect of making autoconf compatible again with all perls in
the 5.6.x release series.
The module Automake::Struct has been removed in automake master
branch (with yesterday's commit v1.11-2055-g74a7f49 "maint: drop
'Automake::Struct' module"): since Automake now requires Perl 5.6,
that module has become obsolete, being basically just a backport
of Perl 5.6's 'Class::Struct' to Perl 5.5. With this change, we
follow suite in Autoconf, which syncs some of its internal modules
with Automake.
* lib/Autom4te/Struct.pm: Delete.
* lib/Autom4te/Makefile.am (dist_perllib_DATA): Don't list it
anymore.
* cfg.mk: Don't sync it with the Automake repository anymore.
* lib/Autom4te/Request.pm: Use 'Class::Struct' instead of
'Autom4te::Struct'.
Paul Eggert [Sun, 4 Mar 2012 05:28:01 +0000 (21:28 -0800)]
tests: port AS_TR_SH and AS_TR_CPP test to Solaris 8 wc
* tests/m4sh.at (AS@&t@_TR_SH and AS@&t@_TR_CPP): Do not assume
that "wc -l" outputs only digits; on Solaris 8 it also outputs
blanks and POSIX allows this.
Paul Eggert [Sun, 4 Mar 2012 02:47:36 +0000 (18:47 -0800)]
tests: fix "#/bin/sh" typo
* tests/m4sh.at (AS@&t@_EXECUTABLE): "#!/bin/sh", not "#/bin/sh".
Typo reported by Tim Rice in:
http://lists.gnu.org/archive/html/autoconf-patches/2012-03/msg00009.html
Paul Eggert [Sun, 4 Mar 2012 02:42:21 +0000 (18:42 -0800)]
tests: port AS_EXECUTABLE_P test to Solaris 8 /bin/sh
* tests/m4sh.at (AS@&t@_EXECUTABLE): Treat any nonzero exit
status as failure. This is needed for Solaris 8 /bin/sh,
where executing a nonexecutable file causes the shell
to say the file had exit status 1.
configure: don't infloop when re-executing with $CONFIG_SHELL
It turns out our guard against infinite recursion wasn't good
enough when shells without $LINENO support were involved, since
the creation-and-sourcing of configure.lineno broke the guard's
expectations. Reports by Tim Rice and Paul Eggert.
* lib/m4sugar/m4sh.m4 (_AS_LINENO_PREPARE): Export '_as_can_reexec'
to "no" before sourcing the just-created configure.lineno.
Eric Blake [Sat, 25 Feb 2012 04:54:39 +0000 (21:54 -0700)]
m4sh: make AS_EXECUTABLE_P public
In the process of making it public, factor it into a reusable
function. This makes constructs like AC_CHECK_PROGRAM smaller,
as well as making libtool's naughty use of $as_executable_p safer.
* lib/m4sugar/m4sh.m4 (_AS_TEST_PREPARE): Add a shell function.
(AS_EXECUTABLE_P): Forward to shell function.
* doc/autoconf.texi (Common Shell Constructs): Document it.
* NEWS: Mention this.
Eric Blake [Sat, 25 Feb 2012 04:09:34 +0000 (21:09 -0700)]
m4sh: require that 'test -x' works
4.3BSD is no longer a reasonable portability target; and we are
pretty sure that these days we can find at least one shell on any
platform that supports 'test -x'. Drop a horribly unsafe use of
eval as a result. :)
Libtool still uses $as_executable_p without so much as calling
either AS_TEST_X or AS_EXECUTABLE_P; even though the latter has
existed, although undocumented, since at least 2.59; furthermore,
libtool uses it in a context where filtering out directories
would have been desirable. Shame on them.
* lib/m4sugar/m4sh.m4 (_AS_TEST_X_WORKS): New probe.
(AS_SHELL_SANITIZE, AS_INIT): Use it in shell searching.
(AS_TEST_X, AS_EXECUTABLE_P): Simplify.
Eric Blake [Sat, 25 Feb 2012 03:45:35 +0000 (20:45 -0700)]
doc: mention more pitfalls of file mode tests
4.3BSD is museum-ware now, so we can assume that test -x exists;
however, we still can't assume that it always does what we want.
* doc/autoconf.texi (Limitations of Builtins) <test (files)>:
Treat 'test -x' as mostly portable, but mention problems with
root user, ACLs, and TOCTTOU races.
Paul Eggert [Fri, 10 Feb 2012 07:33:29 +0000 (23:33 -0800)]
maint: replace FSF snail-mail addresses with URLs
* tests/erlang.at, tests/go.at, tests/statesave.m4:
Replace FSF snail mail addresses with URLs, as per GNU coding
standards, and for consistency with other tests.
Eric Blake [Mon, 23 Jan 2012 17:48:03 +0000 (10:48 -0700)]
INSTALL: convert to '' quoting, drop blank line at end
This is allowed by recent GNU Coding Standards changes, and
mirrors recent gnulib changes:
https://lists.gnu.org/archive/html/bug-gnulib/2012-01/msg00267.html
https://lists.gnu.org/archive/html/bug-gnulib/2012-01/msg00298.html
I've confirmed that after these changes, the INSTALL generated and
installed by autoconf matches the INSTALL.ISO in gnulib.
* Makefile.am ($(srcdir)/INSTALL): Match gnulib INSTALL.ISO.
Jim Meyering [Sat, 21 Jan 2012 15:12:45 +0000 (16:12 +0100)]
maint: convert .x-sc_* into exclude_file_name_regexp--sc_* exemptions
Many of the .x-sc_* exemptions were no long necessary. Remove those
files and instead, provide exemptions via variable definitions in
cfg.mk to address the few remaining exceptions.
* .x-sc_prohibit_atoi_atof: Remove file.
* .x-sc_space_tab: Likewise.
* .x-sc_sun_os_names: Likewise.
* .x-sc_trailing_blank: Likewise.
* .x-sc_two_space_separator_in_usage: Likewise.
* .x-sc_useless_cpp_parens: Likewise.
* cfg.mk: Add minimal exemptions.
* cfg.mk: Add minimal exemptions.
* doc/standards.texi (Standard C): Address the sole useless-cpp-parens
violation in this file:
-#if defined (__STDC__) || defined (WINDOWSNT)
+#if defined __STDC__ || defined WINDOWSNT
With that, the only remaining offender is config.guess, whose name
is now listed in cfg.mk.
Suggested by Eric Blake.
Paul Eggert [Mon, 23 Jan 2012 08:10:55 +0000 (00:10 -0800)]
doc: work around mingw-w64 alloca problem with example
* doc/autoconf.texi (Particular Functions): In example code for
alloca, do not re-#define alloca. This works around a mingw-w64
problem reported by Vincent Torri in
<http://lists.gnu.org/archive/html/autoconf/2012-01/msg00018.html>.
Jim Meyering [Sat, 21 Jan 2012 11:05:02 +0000 (12:05 +0100)]
maint: fix or disable failing syntax-check rules
* cfg.mk (local-checks-to-skip): List failing tests, so we skip
them, for now.
(old_NEWS_hash): Update.
* doc/autoconf.texi: Per suggestion from Eric Blake, obfuscate
the first word of "Filesystem Hierarchy Standard" as File@/system
so it continues to render as one word, yet doesn't trigger the
syntax-check prohibition.
Jim Meyering [Tue, 17 Jan 2012 13:22:02 +0000 (14:22 +0100)]
maint: also sync maint.mk and useless-if-before-free from gnulib
* cfg.mk (gnulib-update): Add them to the list.
* maint.mk: Update from gnulib.
* build-aux/gitlog-to-changelog: Likewise.
* build-aux/useless-if-before-free: New file, from gnulib.
* doc/gnu-oids.texi: Update.
maint: make position of gnulib checkout configurable
Some gnulib-related tools (most prominently, the gnulib-provided
'bootstrap' script) allow the user to define the position of his
gnulib's repository checkout through the use of the 'GNULIB_SRCDIR'
environment variable. We should do the same, for consistency and
to easily support slightly unusual layouts in developers' source
trees.
* cfg.mk (gnulib_dir): Define to "$GNULIB_SRCDIR" if that's set,
and to default value of "'$(abs_srcdir)'/../gnulib" otherwise.
Update comments.
* lib/Autom4te/Getopt.am: The master copy of this file has
been moved to the Automake repository (see Automake commit
'v1.11-662-g52246cc' 2012-01-18, "cmdline parsing: move into
a dedicated perl module"). So we now we sync it from there,
by listing it ...
* cfg.mk (autom4te_files): ... in this variable.
build: fix automake error due to missing ChangeLog
Apparently, Automake does not accept the '$(srcdir)/ChangeLog'
target in Makefile.am as a declaration that ChangeLog is
automatically generated (and thus does not need to exist at
automake time). One has to use a *literal* 'ChangeLog' target.
Problem introduced in commit v2.68-118-g6ed5195 of 2012-01-17,
"maint: generate ChangeLog from git log".
* Makefile.ma ($(srcdir)/ChangeLog): Renamed ...
(ChangeLog): ... to this.
* lib/Autom4te/General.pm (getopt): Move the guts of its
implementation ...
* lib/Autom4te/Getopt.pm (parse_options): .. into this function
in the new Autom4te::Getopt module. This will make it simpler
for the implementation to be shared with other projects (right
now, Automake).
* lib/Automake/Makefile.am (dist_perllib_DATA): Add the new
module.
Jim Meyering [Tue, 17 Jan 2012 11:22:59 +0000 (12:22 +0100)]
tests: avoid spurious failure for each gnu-fortran-using test
The tests compare pre- and post-run lists of envvars, which must
be the same, modulo a list of known, filtered-out exceptions.
However, when running fortran-checking tests with GNU fortran, each
would fail due to the post-run addition of the GFC symbol added in v2.68-97-gbd962ac.
* tests/local.at (AT_CHECK_ENV): Add GFC to the list of symbols
that we ignore in pre-/post-run environment diffs.
Following the practice set by various other GNU projects, we start
to automatically generate the ChangeLog file from the git commit
messages. This will avoid duplication (as the ChangeLog entries
were always inserted both in the git commit message and in the
version-controlled ChangeLog file), and potential problems with
spurious merge conflicts (which, although greatly mitigated by
Bruno Haible's `git-merge-changelog' helper program, have never
been completely solved).
* ChangeLog: Moved ...
* ChangeLog.3: ... to this.
* build-aux/gitlog-to-changelog: New script, synced from gnulib.
* cfg.mk (gnulib-update): Also sync gitlog-to-changelog.
* Makefile.am (gen-ChangeLog): New .PHONY rule, generate the
ChangeLog for distribution.
(dist-hook): Depend on it.
($(srcdir)/ChangeLog): New dummy rule, to pacify automake "gnu"
strictness. Creates a dummy ChangeLog, that will be overridden
by the proper one at distribution time.
(gen_start_date): New variable, the date starting from which the
git log entries are to be copied in the generated ChangeLog.
(EXTRA_DIST): Add ChangeLog.3 and gitlog-to-changelog.
* configure.ac (AC_CONFIG_SRCDIR): Use 'lib/autoconf/autoconf.m4'
instead of 'ChangeLog' as the sentinel file.
* .gitignore: Add ChangeLog.
getopt: remove hack for special handling of "-" argument
Older versions of Getopt::Long acted bogusly and died when they
where configured with the 'bundling' flag and an argument '-' was
seen on the command line they were parsing. That is no longer
the case though, and has not been for quite a long time: the bug
is no longer present in the 5.6.2 version of perl and the 2.25
version of Getopt::Long (and today, the latest versions of perl
and Getopt::Long are respectively 5.14.2 and 2.38). The obsolete
workaround for that Getopt::Long bug can thus be removed from our
'getopt' function.
It is also worth noting that such a workaround was quite buggy
and brittle itself; for example, a command like this:
"autom4te --output -"
would have caused the incorrect diagnostic:
"autom4te: option `--output' requires an argument"
Much worse, a command like this:
"autom4te --language=autoconf --output - configure.ac"
would have caused the standard input of autom4te to be processed
and copied into the 'configure.ac' file, deleting its pre-existing
content! Surely not what a user would have expected.
After this change, a command like this:
autom4te --language=autoconf --output - - <configure.ac >out
works as expected, processing the input from 'configure.ac' and
writing it to the 'out' file.
* lib/Autom4te/General.pm (use): Require perl version 5.6.2.
(getopt): Remove the old workaround.
getopt: fix diagnostic for missing mandatory option argument
Before this change, an incorrect command line usage:
"autom4te --output"
triggered broken diagnostic like:
"autom4te: unrecognized option `--output'"
instead of the expected and correct:
"autom4te: option `--output' requires an argument"
* lib/Autom4te/General.pm (getopt): Give correct diagnostic in
case of usage errors due to missing arguments for options for
which they are mandatory. Code basically copied from automake's
'parse_arguments' private subroutine.
Paul Eggert [Thu, 5 Jan 2012 20:32:12 +0000 (12:32 -0800)]
doc: mention Bash 2.03 bug with backslash-newline
* doc/autoconf.texi (Invoking the Shell): New section.
(Backslash-Newline-Empty): Rename from Backslash-Newline-Newline.
Mention problem with Bash 2.03.
Paul Eggert [Mon, 2 Jan 2012 19:22:25 +0000 (11:22 -0800)]
autoconf: remove " -link" and ")" from xlf output
* lib/autoconf/fortran.m4 (_AC_PROG_FC_V_OUTPUT):
Also remove " -link" and trailing ")" from xlf output.
Problem and fix reported by Thomas Jahns in
<http://lists.gnu.org/archive/html/bug-autoconf/2012-01/msg00000.html>.
configure: will re-execute with $CONFIG_SHELL, if it's set
* lib/m4sugar/general.m4 (_AS_DETECT_BETTER_SHELL): Define the macro
`_AS_FORCE_REEXEC_WITH_CONFIG_SHELL' to `yes', so that the code in
`_AS_DETECT_BETTER_SHELL' will cause autoconf-generated configure
scripts to always re-execute themselves with $CONFIG_SHELL, if it's
set in the environment.
* doc/autoconf.texi (config.status Invocation): Update.
* doc/install.texi (Defining Variables): Likewise.
* NEWS: Likewise.
* tests/m4sh.at: Add tests for the new semantics in ...
(Configure re-execs self with CONFIG_SHELL): ... this new
test group.
m4sh: allow forced re-execution with $CONFIG_SHELL, if it's set
* lib/m4sugar/m4sh.m4 (_AS_DETECT_BETTER_SHELL): If the m4sh client
has defined the macro `_AS_FORCE_REEXEC_WITH_CONFIG_SHELL' to
"yes", emit code to always re-execute the current script with
$CONFIG_SHELL, if that's set.
* tests/m4sh.at: Add tests for the new and old semantics, in ...
(Re-exec with CONFIG_SHELL, Forced re-exec with CONFIG_SHELL): ...
these new test groups.
Paul Eggert [Mon, 26 Dec 2011 08:34:13 +0000 (00:34 -0800)]
AS_LN_S: fall back on 'cp -pR' (not 'cp -p') if 'ln -s' fails
This works better for symlinks to directories.
Problem reported by Eli Zaretskii via Werner Lemberg in
<http://lists.gnu.org/archive/html/bug-autoconf/2011-12/msg00006.html>.
* NEWS:
* doc/autoconf.texi (Particular Programs): Document this.
* lib/m4sugar/m4sh.m4 (_AS_LN_S_PREPARE): Implement this.
Paul Eggert [Thu, 8 Dec 2011 06:41:04 +0000 (22:41 -0800)]
AC_LANG_BOOL_COMPILE_TRY(C): port to g++ with warnings
* lib/autoconf/c.m4 (AC_LANG_BOOL_COMPILE_TRY(C)): Use the
array as well as setting it, to pacify g++. Reported by
Werner Lemberg in
<http://lists.gnu.org/archive/html/autoconf/2011-12/msg00005.html>.
fortran: define $GFC to "yes" if $FC is a GNU compiler
* lib/autoconf/fortran.m4 (AC_PROG_FC): Define `$GFC' to "yes" if
the detected fortran compiler is a GNU compiler, define it to the
empty string otherwise.
This is mostly for consistency for what is done for the C, C++
and Fortran 77 compilers.
* doc/automake.texi: Update.
docs: some fixlets in section about shell signal handling
* doc/autoconf.texi (Signal handling): Rename ...
(Signal Handling): ... to this, for consistency with other node
names. Fix some typos and grammaros. Add more URL references
in comments.
docs: korn shells can have $? > 256 for signal-terminated children
Some Korn shells, when a child process dies due to signal number
n, can leave in $? an exit status of 256+n, instead of the more
common 128+n. See also Austin Group issue 0000051:
<http://www.austingroupbugs.net/view.php?id=51>
* doc/autoconf.texi (Signal handling): Document the described Korn
Shell behaviour, and some of its possible shortcomings.
Eric Blake [Mon, 26 Sep 2011 18:01:11 +0000 (12:01 -0600)]
docs: relax documentation license by dropping cover text
See https://lists.gnu.org/archive/html/bug-diffutils/2011-08/msg00022.html
for precedence in diffutils. The autoconf manual, as of this commit,
was still barely below 400 pages.
* doc/autoconf.texi (copying): Drop front- and back-cover texts.
* NEWS: Document this.
Reported by Brian Gough.
docs: signal-related bugs and incompatibilities for the shells
* doc/autoconf.texi (Signal handling): New paragraph.
(@menu at "Portable Shell", @detailmenu): Update.
Motivated by recent discussion on the bug-autoconf list, as well
as work in the automake testsuite:
<https://lists.gnu.org/archive/html/bug-autoconf/2011-09/msg00003.html>
<https://lists.gnu.org/archive/html/bug-autoconf/2011-09/msg00004.html>
<http://lists.gnu.org/archive/html/automake-patches/2011-09/msg00066.html>
docs: more details about make VPATH rewriting woes
* doc/autoconf.texi (Automatic Rule Rewriting): Solaris make
VPATH rewriting applies to any whitespace-separated word in a
rule, so it might apply also to shell variables, functions
and keywords (and automake has already tripped on this once);
document this, with an example. Since we are at it, do some
minor reformatting of existing text.
Paul Eggert [Thu, 1 Sep 2011 06:16:06 +0000 (23:16 -0700)]
AC_C_CONST: don't reject gcc -Werror -Wall
* lib/autoconf/c.m4 (AC_C_CONST): Don't reject gcc when it is used
with -Werror -Wall during configuring. It's unwise to use GCC
that way, but apparently enough people do it nowadays that it's an
issue. These days nobody uses the old compilers that the old
tests reject, so we can't test this fix against them, but it's
more important to work with modern GCC (even when misused) than to
work with no-longer-used compilers. Problem reported by Shevek in
<http://lists.gnu.org/archive/html/bug-autoconf/2008-11/msg00007.html>
and raised again by Dan Kegel in
<http://lists.gnu.org/archive/html/bug-autoconf/2011-08/msg00020.html>.
Motivated by automake bug#9245:
<http://debbugs.gnu.org/cgi/bugreport.cgi?bug=9245>
and FreeBSD PR bin/159730:
<http://www.freebsd.org/cgi/query-pr.cgi?pr=159730>
* doc/autoconf.texi (Parallel Make): Document other BSD make
incompatibilities. Reorganize the existing related documentation
accordingly.
Eric Blake [Fri, 22 Jul 2011 13:03:47 +0000 (07:03 -0600)]
docs: fix minor doc problems
* doc/autoconf.texi (Why Not Imake): Fix grammar.
(autoreconf Invocation): Fix short option for --version.
* THANKS: Update.
Reported by Christophe Jarry and Russ Allbery.
Paul Eggert [Sat, 18 Jun 2011 21:39:57 +0000 (14:39 -0700)]
* lib/autoconf/functions.m4 (_AC_LIBOBJ_ALLOCA): Try to outsmart GCC.
Problem with stack-detection code reported by Andy Wingo in
<http://lists.gnu.org/archive/html/bug-autoconf/2011-06/msg00032.html>.
This fix is imported from gnulib's c-stack module.
projects / thirdparty / autoconf.git / log
