* doc/autoconf.texi (C Compiler): Document the intent of AC_PROG_CC
better. It's not meant to check for strict conformance, only to
get the latest version with extensions.
(Running the Preprocessor, Present But Cannot Be Compiled):
Document new diagnostic wording.
* lib/autoconf/c.m4 (_AC_C_STD_TRY): Change diagnostic wording from
"checking for gcc option to accept ISO C11" to
"checkint for gcc option to enable C11 features", as this better
reflects what is actually happening.
(_AC_PROG_CC_C99): Put -qlanglvl=extc1x here ...
(_AC_PROG_CC_C11): ... rather than here, as it doesn't pass the C11
test with IBM XL C V12.1, and the point is to enable features not to
test for strict conformance.
* lib/autoconf/c.m4 (_AC_PROG_CC_C11): Add -qlanglvl=extc1x, for
IBM XL C V12.1. It shouldn't pass the C11 test yet, since it's not
documented to support _Alignas, _Alignof, UTF-8 string literals,
and duplicate typedefs, but presumably it will eventually.
* NEWS:
* doc/autoconf.texi (C Compiler, Running the Preprocessor)
(Limitations of Usual Tools, Present But Cannot Be Compiled)
(Obsolete Macros):
Document the changes described below.
* lib/autoconf/c.m4 (_AC_PROG_CC_FORCE_VERSION): Remove.
(AC_PROG_CC_C89, AC_PROG_CC_C99, AC_PROG_CC_STDC):
Just do AC_PROG_CC, but mark as obsolete. This replaces my recent
ill-advised attempt to let AC_PROG_CC_C89 and AC_PROG_CC_C99 downgrade
the version of C supported.
* doc/autoconf.texi (Limitations of Usual Tools, Volatile Objects):
Document C11 more accurately. In some cases this involves removing
some details about 'volatile', alas, since C11 changed this stuff.
Again.
Prompted by Stefano Lattarini's report of a test failure due to
a missing preselection for automake 1.12.
* tests/tools.at (autom4te preselections): Sort before diffing.
The files in lib/Autom4te/ are intentionally not synced at this point,
since automake commit v1.11-2114-g2d671e1 "perl refactor: use modern
semantics of 'open'":
<http://lists.gnu.org/archive/html/automake-patches/2012-03/msg00111.html>
would require wider adaptation of our scripts to the new XFile API, and
also exposes some latent bugs in autoconf where we use raw 'open' instead
of XFile::open. We'll take care of that in a later patches (maybe).
* build-aux/announce-gen: Resync via 'make fetch'.
* build-aux/config.guess: Likewise.
* build-aux/config.sub: Likewise.
* build-aux/gendocs.sh: Likewise.
* build-aux/gitlog-to-changelog: Likewise.
* build-aux/gnupload: Likewise.
* build-aux/texinfo.tex: Likewise.
* doc/make-stds.texi: Likewise.
* doc/standards.texi: Likewise.
* maint.mk: Likewise.
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
Issue revealed by a failure in test "35: tools.at: autom4te preselections":
<http://lists.gnu.org/archive/html/bug-autoconf/2012-09/msg00020.html>
* lib/autom4te.in: Add 'AM_PROG_MKDIR_P'.
Helped-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
According to Automake's NEWS file, it is since at least Automake 1.8,
and in autoreconf we are already assuming aclocal >= 1.8 anyway.
* bin/autoreconf.in (parse_args): Simplify a little by just assuming
the automake option '--force-missing' is supported.
($automake_supports_force_missing): Delete, no longer needed.
* NEWS: Update.
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
* bin/autoreconf.in (run_aclocal): After the previous commit, this has
become just a useless wrapper around xsystem("$aclocal .."), so get rid
of it, and inline its expansion in the two places where it was used ...
(autoreconf_current_directory): ... in here.
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
The minimal automake and aclocal version required by the "most"
conservative important real world-projects (like Gnulib and Libvirt)
is 1.9 anyway (which is the version installed on old but still
supported installations of stable Distros like RHEL 5), so this
change should be safe and justified by now.
* bin/autoreconf.in (parse_args): Simplify by just assuming the aclocal
options '--force' and '--no-force' are supported and works correctly.
($aclocal_supports_force): Delete, no longer needed.
(run_aclocal): Heavily simplify by assuming that aclocal properly creates
'aclocal.m4' as lazily as possible.
* NEWS: Update.
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
As per updated GCS recommendations.
* bin/autoconf.as, bin/autoreconf.in, bin/autoscan.in, ifnames.in,
bin/autoupdate.in: Throughout these files.
* bin/autoheader.in, bin/autom4te.in: Likewise. Also, remove some
useless escaping of the "'" single-quote characters, and reformat
some message for better line wrapping.
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
* NEWS:
* doc/autoconf.texi (C Compiler): Document the following.
* lib/autoconf/c.m4 (AC_PROG_CC): Check for the latest C version
supported, not just C89.
(_AC_C_STD_TRY): Keep track of the options we add to bring
the C compiler up to standard, so that we can undo it if the
user later requests some other C standard.
(_AC_PROG_CC_FORCE_VERSION): New macro.
(AC_PROG_CC_C89, AC_PROG_CC_C99): Use it. These macros now
have a documented side effect of changing the C version requested.
(AC_PROG_CC_C11): Remove. It wasn't useful.
(AC_PROG_CC_STDC): Now an obsolescent alias for AC_PROG_CC.
(AC_C_PROTOTYPES): Allow any standard C version, not just c89.
Don't chatter, since we don't actually run any checking code.
* lib/autoconf/types.m4 (AC_TYPE_LONG_LONG_INT)
(AC_TYPE_UNSIGNED_LONG_LONG_INT): Treat C11 like C99.
* TODO: Remove the TODO item corresponding to the above.
* doc/autoconf.texi (Default Includes, Particular Functions)
(Header Portability):
* lib/autoconf/c.m4 (AC_LANG_FUNC_LINK_TRY(C)):
* lib/autoconf/headers.m4 (AC_HEADER_STDC):
* lib/autoconf/types.m4 (_AC_CHECK_ALIGNOF):
* lib/m4sugar/m4sugar.m4 (m4_require) [comment only]:
Assume the existence of the C89 freestanding headers <float.h>,
<limits.h>, <stdarg.h>, <stddef.h>, as that's safe nowadays.
This is less likely to run into gotchas, and should fix a
cross-compilation bug with newer GCC reported by Myke Frysinger in
<http://lists.gnu.org/archive/html/bug-autoconf/2012-09/msg00001.html>.
* NEWS:
* doc/autoconf.texi (C Compiler): Document this.
(Gnulib, Function Portability, Particular Functions)
(Header Portability, Particular Headers, Defining Symbols)
(Printing Messages, Limitations of Usual Tools)
(Preprocessor Arithmetic, Volatile Objects, Exiting Portably):
Modernize wording for C11.
* lib/autoconf/c.m4 (_AC_C_C99_TEST_HEADER, _AC_C_C99_TEST_BODY):
New macros, taken from _AC_PROG_CC_C99. These are so that we can
also include the C99 tests in the C11 test program.
(_AC_PROG_CC_C99): Use them.
(_AC_PROG_CC_C11, AC_PROG_CC_C11): New macros.
(AC_PROG_CC_STDC): Prefer C11 to C99 or C89.
Test that: invalid variable names are detected, that the variable
name does not get macro expanded, that assignment to the variable
works whether as part of AC_SUBST or independently, that the last
assignment wins.
* doc/autoconf.texi (Setting Output Variables) <AC_SUBST>: Mention
that variable does not overlap with macros.
* tests/base.at (AC_SUBST): New test.
Signed-off-by: Eric Blake <eblake@redhat.com>
Consider the following:
% cat >configure.ac <<'EOF'
AC_INIT([test], [0])
m4_define([FOO], [baz])
AC_SUBST([FOO], [bar])
AC_CONFIG_FILES([test])
AC_OUTPUT
EOF
% cat >test.in <<'EOF'
@FOO@
EOF
This produces no error messages at autoconf time and none at configure
time. Nevertheless, the substituted value of FOO is the empty string,
instead of bar, as expected. Sure enough, in the output variables
section of config.log, we see FOO='' instead of FOO='bar'. Looking
at the generated configure script, we see that AC_SUBST has produced
baz=bar in the output, instead of the expected FOO=bar. But this is
the only place: everywhere else is still using FOO.
* lib/autoconf/general.m4 (AC_SUBST): Add another layer of
quoting.
* THANKS: Update.
Signed-off-by: Eric Blake <eblake@redhat.com>
Do so because future automake and aclocal versions (starting from 1.13)
drop support for 'configure.in' as the name of the Autoconf input file.
Without this patch, the Autoconf testsuite experiences some spurious
failures when run with the development version of aclocal and automake
installed early enough in $PATH.
* tests/torture.at: Rename 'configure.in' to 'configure.ac' throughout.
Remove an obsolete comment about backward-compatibility.
Helped-by: Jim Meyering <jim@meyering.net>
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
IFS may be modified temporarily when the configure script receives a
signal. Make sure the EXIT trap uses the standard value.
* lib/autoconf/general.m4: Sanitize IFS in trap.
* THANKS: Update.
Copyright-paperwork-exempt: Yes
The texinfo manual recommends avoiding the use of a trailing @c in
any macro designed to be used inline (as is the case with our ovar
and dvar macros). Furthermore, passing '@\n' in the middle of a
macro call is much different than passing '@\n' between arguments
of a @defmac for line continuation.
* doc/autoconf.texi (ovar, dvar): Don't end macro with @c, since
these macros are designed to be embedded in one-line usage.
(Fortran Compiler): Don't split @dvar.
* THANKS: Update.
Reported by Stefano Lattarini.
Signed-off-by: Eric Blake <eblake@redhat.com>
Copyright-paperwork-exempt: Yes
Otherwise, newer texinfo parses this line as a line directive, and
any error later in the file will claim to be from "conftest.c" instead
of "autoconf.texi".
* doc/autoconf.texi (Generating Sources): Avoid confusing newer
texinfo into thinking we had a line directive.
Reported by Stefano Lattarini, fix suggested by Patrice Dumas.
As reported by Paul Keir on the cygwin lists,
http://cygwin.com/ml/cygwin/2012-07/msg00263.html,
some people like to stick / in their $PATH, and if we then try
to probe $as_dir/progname for existence, we can end up causing
cygwin to have a several-second timeout per //name probe. It
is better to avoid inserting the extra slash when $as_dir is the
root directory, and simpler to code by always having a trailing
slash present than it is to strip a trailing slash. Thankfully,
_AS_PATH_WALK is an undocumented interface, and even if someone
was using it in spite of the warnings, their use of $as_dir/foo
will typically only lead to odd-looking /dir//foo probes, with
only the case of / in $PATH causing slowdowns, and only when //
is special.
There was also a minor bug where the if-not-found code of
_AS_PATH_WALK could be executed with $IFS still in the wrong state.
* lib/m4sugar/m4sh.m4 (_AS_PATH_WALK): Always end as_dir in /.
Avoid wrong IFS during if-not-found. Minor optimization to avoid
regex.
(_AS_DETECT_BETTER_SHELL, _AS_SHELL_SANITIZE): Update clients.
* lib/autotest/general.m4 (_AT_FINISH): Likewise.
* lib/autoconf/programs.m4 (_AC_CHECK_PROG, _AC_PATH_PROG)
(_AC_PATH_PROGS_FEATURE_CHECK, _AC_PATH_PROG_FLAVOR_GNU): Likewise.
Now that automake documents AM_MISSING_PROG, and given that automake
has reserved the right to change the calling conventions of 'missing',
we should not recommend a hard-coded use of 'missing --run'.
* doc/autoconf.texi (Making testsuite Scripts): Recommend
AM_MISSING_PROG when using automake, and avoid hard-coding use of
'missing' otherwise.
Automake 1.13 will be changing the semantics of 'missing'; maintaining
our own copy in version control risks problems if our version does
not match automake's expectations. As a result, gnulib no longer
mirrors 'missing'. Furthermore, gnulib originally added the
'elisp-comp' module with the explanation that autoconf uses it, but
we don't use it anywhere other than the manner in which automake
will byte-compile our .el files; since we don't document the script,
we should be just fine using the version that was installed by automake.
See more discussion in the thread starting here:
https://lists.gnu.org/archive/html/automake-patches/2012-06/msg00154.html
* cfg.mk (gnulib-update): Drop files installed by automake and no
longer present in gnulib.
* .gitattributes: Delete references to files not in git.
Cater to the Bitrig OS, an OpenBSD fork. config.guess and
config.sub have been updated upstream.
* lib/autoconf/functions.m4 (_AC_FUNC_MALLOC_IF)
(_AC_FUNC_REALLOC_IF): Bitrig inherits from BSD.
It has been years since that has been deprecated in the documentation,
in favour of 'configure.ac':
Previous versions of Autoconf promoted the name configure.in, which
is somewhat ambiguous (the tool needed to process this file is not
described by its extension), and introduces a slight confusion with
config.h.in and so on (for which '.in' means "to be processed by
configure"). Using configure.ac is now preferred.
It's now time to start giving runtime warning about the use of
'configure.in', so that support for it can be removed in future
versions of autoconf/automake.
* lib/Autom4te/Configure_ac.pm: Issue a warning in the 'obsolete'
category if 'configure.in' is detected. Since this module is synced
from Automake, this change is to be backported there (and will be
soon).
* doc/autoconf.texi: Update.
* tests/tools.at: Adjust to avoid spurious failures.
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
Paul and Eric suggested that improving the cross-compilation guesses
for targets that are glibc systems would be welcome here [1][2].
This patch modifies the cross-compilation behaviour of
AC_FUNC_CHOWN
AC_FUNC_GETGROUPS
AC_FUNC_LSTAT_FOLLOWS_SLASHED_SYMLINK
AC_FUNC_MALLOC
AC_FUNC_MMAP
AC_FUNC_REALLOC
AC_FUNC_STRCOLL
so that when the target is a glibc system (or, in case of AC_FUNC_MALLOC
and AC_FUNC_REALLOC, any known "good" Unix system), the guess is
"yes it works" rather than "guessing no".
This is important because some of these macros are used in Gnulib, and
in case of "guessing no" Gnulib provides extra workaround code, and
1) Generally, when targetting embedded systems, code size should be
minimized,
2) In [3], unnecessary workaround code will look like a Glibc bug.
This patch also changes the configure output to "guessing yes" or
"guessing no" in a case where the ac_cv_* variable is undocumented.
[1] http://lists.gnu.org/archive/html/bug-gnulib/2012-05/msg00010.html
[2] http://lists.gnu.org/archive/html/bug-gnulib/2012-05/msg00013.html
[3] http://sourceware.org/glibc/wiki/Testing/Gnulib
This patch was tested with the simple configure.ac file
=============== configure.ac ==================
AC_INIT([dummy], [0])
AC_FUNC_CHOWN
AC_FUNC_GETGROUPS
AC_FUNC_LSTAT_FOLLOWS_SLASHED_SYMLINK
AC_FUNC_MALLOC
AC_FUNC_MMAP
AC_FUNC_REALLOC
AC_FUNC_STRCOLL
AC_OUTPUT
===============================================
* lib/autoconf/functions.m4 (AC_FUNC_CHOWN): Require AC_CANONICAL_HOST.
When cross-compiling to a glibc system, guess yes.
(AC_FUNC_GETGROUPS): Likewise.
(AC_FUNC_LSTAT_FOLLOWS_SLASHED_SYMLINK): Likewise.
(AC_FUNC_STRCOLL): Likewise.
(_AC_FUNC_MALLOC_IF): Require AC_CANONICAL_HOST. When cross-compiling
to a know Unix system other than AIX and OSF/1, guess yes.
(_AC_FUNC_REALLOC_IF): Likewise.
(AC_FUNC_MMAP): Require AC_CANONICAL_HOST. When cross-compiling to a
system with a Linux kernel, guess yes.
Obviously, I didn't fully test commit f35498d.
* doc/autoconf.texi (Limitations of Builtins): Add a comma.
Signed-off-by: Eric Blake <eblake@redhat.com>
At 2.68b, I asked whether anyone would miss .gz and .bz2 formats.
Consensus was overwhelming that .gz still holds a place in people's
hearts, in spite of .xz compressing to smaller files, but no one
was able to make a convincing argument for .bz2.
* configure.ac (AM_INIT_AUTOMAKE): Drop bzip2; xz wins hands down.
Signed-off-by: Eric Blake <eblake@redhat.com>
The files in lib/Autom4te/ are intentionally not synced at this
point, since this recent Automake patch:
https://lists.gnu.org/archive/html/automake-patches/2012-03/msg00111.html
was buggy regarding '-' as stdout, and also exposes some latent
bugs in autoconf where we use raw 'open' instead of XFile::open.
* build-aux/announce-gen: Resync via 'make fetch'.
* build-aux/config.sub: Likewise.
* build-aux/git-version-gen: Likewise.
* build-aux/gnupload: Likewise.
* build-aux/move-if-change: Likewise.
* build-aux/texinfo.tex: Likewise.
* doc/standards.texi: Likewise.
* maint.mk: Likewise.
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.
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.
Signed-off-by: Eric Blake <eblake@redhat.com>
* 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.
Signed-off-by: Eric Blake <eblake@redhat.com>
* lib/Autom4te/General.pm: This file is *not* used by Automake;
adjust comments accordingly.
Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
* 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.
This reduces the implicit requirement from 5.6.2 back to 5.6,
while raising the explicit requirement to match the actual code.
* configure.ac (PERL): Fail up front if perl is too old.
* NEWS: Document this.
* README: Likewise.
* README-hacking: Likewise.
* lib/Autom4te/ChannelDefs.pm: Bump requirement.
* lib/Autom4te/General.pm: Relax requirement.
Recent ksh is noisy:
$ ksh -nc '``'; echo $?
ksh: warning: line 1: `...` obsolete, use $(...)
0
* tests/local.at (AT_CHECK_SHELL_SYNTAX): Ignore noisy ksh on.
Reported by Martin Zaun.
Signed-off-by: Eric Blake <eblake@redhat.com>
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.
* maint.mk: Resync via 'make fetch'.
* lib/Autom4te/Channels.pm: Likewise.
* lib/Autom4te/Configure_ac.pm: Likewise.
* lib/Autom4te/FileUtils.pm: Likewise.
* lib/Autom4te/Getopt.pm: Likewise.
* lib/Autom4te/XFile.pm: Likewise.
