maint: install minimal supported autoconf through convenience targets
To make testing of Automake in conjunction with it easier and more
reproducible. It might be nice to also have convenience targets to
actually automatically run the Automake testsuite with such autoconf
version, but that is left to later patches.
* configure.ac: AC_SUBST the definition of $required_autoconf_version.
* maint.mk (WGET): Move definition earlier.
(gnu-ftp, ac-v, ac-n, ac-p, ac-t, ac-l, ac-d): New auxiliary variables.
(fetch-minimal-autoconf, build-minimal-autoconf): New convenience
targets.
* .gitignore: Update.
And do so in a safer way, with a runtime check rather than a brittle
static maintainer check.
* t/ax/test-init.sh: Set the 'required' variable to readonly.
* syntax-checks.mk (sc_tests_required_after_defs): Remove.
(syntax_check_rules): No longer list it.
tests: use append mode to capture parallel make output
This will allow us to run the Automake testsuite forcing all the make
invocations in the test cases to run in parallel mode [1], but without
hitting the spurious failure reported in automake bug#11413.
See also:
<http://lists.gnu.org/archive/html/bug-make/2013-05/msg00135.html>
The make invocations in the test cases can be forced to run in parallel
mode by, e.g., invoking "make check" like this:
make check AM_TESTSUITE_MAKE="make -j16"
The possible spurious failures hinted above are due to the fact that
the redirected output of parallel make can racily lose lines. For
example, if GNU make (3.82) is run with -j10 on a Makefile like this:
and has its standard output redirected in overwrite-mode to a regular
file, it looses a line of that output every 15 runs or so on a Fedora
17 system with 64 ppc64 cores. Redirection in append mode does not
suffer of this problem, as explained here:
<http://lists.gnu.org/archive/html/automake-patches/2009-03/msg00073.html>
See also previous commits v1.12-63-g45c1fcd of 2012-05-05 (tests: use
append mode to capture parallel make output) and Release-1-10-280-g6426999
of 2009-03-10 (Use append mode to capture parallel test output).
* t/ax/am-test-lib.sh (run_make): Adjust and enhance.
* t/lisp8.sh: Take advantage of the enhancement, nd stop doing output
redirection for $MAKE by hand.
* t/tap-more.sh: Likewise.
* t/parallel-tests-concurrency.sh: Likewise.
To force the test scripts to invoke make in parallel mode. This should
enhance coverage of use cases (make concurrency) that are becoming more
and more important with today multicore fast machines.
* Makefile.am (check-parallel): New .PHONY target.
tests: avoid '$MAKE' redirections, use 'run_make' instead
The use 'run_make' with the -E, -O and -M option, it is more
idiomatic now. Also, this way, centralized fixes and improvements
done in 'run_make' will automatically propagate through most of
the testsuite.
* syntax-checks.mk (sc_tests_no_run_make_redirect): Also check against
'$MAKE' invocations that uses output redirections (and not only against
'run_make' invocation that do so).
* Several tests: Adjust (and few minor cosmetic changes as well, while
at it).
tests: avoid use of redirected 'run_make' invocations
Instead, properly use the -E, -O, or -M options of run_make.
The occurrences of the 'run_make' invocations using output
redirection has been found by the recently-introduced maintainer
check 'sc_tests_no_run_make_redirect'.
* t/ax/tap-summary-aux.sh: Adjust.
* t/ax/testsuite-summary-checks.sh: Likewise.
* Several other tests: Likewise.
lint: warn against redirected 'run_make' invocations
The testsuite shell function 'run_make()' has the three options
-E, -O and -M to perform the various kind of output redirections,
so the test scripts should use them instead of redirecting the
output "by hand":
See the commit message of v1.13.2-13-gc7cfa09, 2013-05-22
(tests: run_make: options to do command redirection) for an
explanation of why this is a good idea.
It's worth noting that the present patch only introduced a new
static maintainer check looking against the undesired idioms,
but doesn't yet remove usages of such idioms from the testsuite.
That will be done by follow-up patch(es).
* syntax-checks.mk (sc_tests_no_run_make_redirect): New check.
(syntax_check_rules): List it.
tests: run_make: options to do command redirection
Let's improve the API of the 'run_make()' helper shell function by
adding three new options:
-O Save the standard output from make on disk, in a regular file
named 'stdout'.
-E Save the standard error from make on disk, in a regular file
named 'stderr'.
-M Save both the standard output and standard error from make on
disk, in a regular file named 'output'. This option supersedes
both the '-O' and '-E' options.
This new API has two main advantages.
1. Its use will allow us to get rid of more cumbersome idioms
like, e.g.,
2. More importantly, using the new API we will prevent any extra output
from the shell traces of the code in run_make to be redirected along
with the make stderr (where that was redirected). This problem was
present in usages like, e.g.,
ensuring that 'stderr' won't end up containing unrelated stuff.
Note that we do not convert in bulk the old idioms and the use of
redirected 'run_make' invocations with this patch. We only convert
some occurrences, to ensure that the new implementation of 'run_make'
is sound enough. More sweeping conversions will likely be done in
follow-up patches.
* t/ax/am-test-lib.sh (run_make): Enhance and implement the extended API.
* t/tap-xfail-tests.sh: Use the new 'run_make' API.
* t/test-driver-cond.sh: Likewise.
* t/tests-environment-fd-redirect.sh: Likewise.
* t/uninstall-fail.sh: Likewise.
* t/yacc-dist-nobuild.sh: Likewise.
This makes the test logs easier to read for most shells (which do not
actually require that alias). This is especially important now that
'unset' is used in the new 'run_make()' function, and that function is
likely going to be used more and more in the future.
* t/ax/test-lib.sh (_am_unset, unset): Only define this function and
alias if "unset VAR" returns a non-zero exit status when VAR is already
unset.
tests: better idiom to override make macro defs on the cmdline
We now do so with a new wrapper shell function 'run_make()', which
relies on the use of AM_MAKEFLAGS, avoiding the use of the '-e' make
option. The use of that option (that causes the environment variables
to take precedence over the macro definitions in the Makefile) has
proved to be quite brittle in the past, causing annoying and sometimes
problematic spurious failures. This has worsened in some still
unpublished work going on in development branches. It's time to solve
the issue once and for all.
Note that we do not convert all uses of $MAKE in the testsuite right
away; we might do so in follow-up changes, with leisure, to avoid a
"patch bomb" effect (this commit is already too much of a bomb itself).
What we do in this commit is to get rid of all "$MAKE -e" invocations.
We admit that the implementation and feature-set of 'run_make()' are
far from perfect, but good enough for our current purposes. We'll
improve 'run_make()' if and when the need arises.
* syntax-checks.mk (sc_tests_no_make_e): New syntax check, guard against
the use of "$MAKE -e".
(syntax_check_rules): Add it.
(sc_tests_overriding_macros_on_cmdline): Adjust.
(lint): New, alias for 'maintainer-check', for lazy typists. Idea
backported from the 'maint' branch (Automake 1.13a).
* t/ax/am-test-lib.sh (run_make): New function. Run $MAKE with the
given command-line arguments, handling command-line override of variable
definitions in a smart way (using AM_MAKEFLAGS if a non-GNU make
implementation is detected to be in use).
(useless_vpath_rebuild): Adjust to use 'run_make', to avoid a spurious
maintainer check failure.
(yl_distcheck): Use 'run_make' rather than bare '$MAKE'.
(single_quote, append_single_quoted, is_valid_varname): New auxiliary
function, used, directly or indirectly, by it.
* Many tests: Adjust to avoid the use of "$MAKE -e", and prefer the
use of 'run_make' in few other contexts as well, where it makes sense.
Other minor fixlets while at it.
* lib/am/lisp.am (.el.elc): By initializing the 'am__dir' properly
here. For most shells, the lacking initialization, while technically
incorrect, didn't cause any issue in practice, because in those shells
"test -d" returns an exit status of 0. But with /usr/xpg4/bin/sh, the
shell complains like this: "test: argument expected", and returns a
non-zero exit status. This caused testsuite failures in several lisp
tests.
Also, while we are at it, use more proper quoting in the recipe, to
ensure a missing initialization to now be caught by more forgiving
shells as well.
* NEWS: Update.
In some shells (e.g., Solaris 10 /bin/ksh, or NetBSD 5.1 /bin/sh),
"unset VAR" returns a non-zero exit status in case the VAR variable
is already unset. This doesn't interact well with our usage of
"set -e" in the testsuite. So far, we've avoided spurious failures
by either explicitly ignoring the exit status from unset:
unset VAR || :
or explicitly ensuring that a variable is set, before trying to
unset it:
VAR=; unset VAR
But we can do better, by aliasing the 'unset' command to a custom
function that will take care of these details for us. This will
avoid us annoying spurious failures in the future, failures that
have already bitten us too much times. For an example, refer to
commit 'v1.12.2-88-g5b1dae5' of 2012-08-05 (tests: avoid tons of
spurious failures on NetBSD).
* t/ax/test-lib.sh (_am_unset): New function.
(unset): New alias to it.
(_am_exit): Adjust comments.
* t/ax/am-test-lib.sh: No need to temporary disable the 'errexit'
shell flag when unsetting variables that are potentially already
unset.
(am_process_requirements): Adjust to remove a now-useless
workaround related to unset.
* t/aclocal-macrodir.tap: Likewise.
* t/aclocal-macrodirs.tap: Likewise.
* t/auxdir-autodetect.sh: Likewise.
* t/ax/am-test-lib.sh: Likewise.
* t/ax/test-lib.sh: Likewise.
* t/check-tests-in-builddir.sh: Likewise.
* t/dist-formats.tap: Likewise.
* t/distcheck-configure-flags-am.sh: Likewise.
* t/distcheck-configure-flags.sh: Likewise.
* t/java-empty-classpath.sh: Likewise.
* t/javaflags.sh: Likewise.
* t/lflags.sh: Likewise.
* t/lflags2.sh: Likewise.
* t/lisp-flags.sh: Likewise.
* t/lisp6.sh: Likewise.
* t/missing-auxfile-stops-makefiles-creation.sh: Likewise.
* t/parallel-am.sh: Likewise.
* t/parallel-am2.sh: Likewise.
* t/parallel-am3.sh: Likewise.
* t/parallel-tests-log-override-recheck.sh: Likewise.
* t/pkg-config-macros.sh: Likewise.
* t/python-missing.sh: Likewise.
* t/python-too-old.sh: Likewise.
* t/python11.sh: Likewise.
* t/self-check-dir.tap: Likewise.
* t/self-check-report.sh: Likewise.
* t/self-check-seq.tap: Likewise.
* t/silent-configsite.sh: Likewise.
* t/suffix6c.sh: Likewise.
* t/tar-override.sh: Likewise.
* t/tests-environment-and-log-compiler.sh: Likewise.
* t/vala-configure.sh: Likewise.
* t/werror3.sh: Likewise.
* t/yflags-cmdline-override.sh: Likewise.
* t/yflags.sh: Likewise.
* t/yflags2.sh: Likewise.
tests: remove exec bit from all of them ('micro' branch)
It gives the impression that they are directly runnable, as with
"./t/foo.sh", but it has been a while since that was the case. Today,
tests are runnable only through "make check" or "./runtest".
This change is for the 'micro' branch (automake 1.13.2a). It will
soon be followed by similar patches for the 'maint' branch (automake
1.13a) and the 'master' branch (automake 1.99a).
* t/*.sh, t/*.tap: Remove executable bit.
* maint.mk (sc_tests_executable): Remove.
(syntax_check_rules): Adjust.
* gen-testsuite-part: Set permissions of generated tests to
'444' (-r--r--r--), rather than 555 (-r-xr-xr-x).
The "vala + pkg-config" doesn't interact well with cross-compilation:
<https://mail.gnome.org/archives/vala-list/2012-September/msg00094.html>
That is causing annoying failures in our testsuite when it's run in
cross-compilation mode. It's not worth trying too be overly smart and
work around these problems; just skip the affected tests instead.
* t/ax/am-test-lib.sh (require_tool): A new requirement 'valac', telling
that the test needs to compile Vala-generated C files. In particular,
this causes the test to be skipped when using a cross-compiler.
warns: don't tell AM_PROG_MKDIR_P is going to be removed
That is no longer true. For a more extended rationale, see file
'PLANS/obsolete-removed/am-prog-mkdir-p.txt' in the maint branch
(as of commit v1.13.1d-132-g90ec3fe).
* automake.in (scan_autoconf_traces): So adjust the warning message
here.
* maint.mk (announcement): Here, be prepared to handle the case
in which the first section of the NEWS file is dedicated to report
future backward-incompatibilities and/or other warnings.
Merge 'better-makeflags-recognition' and 'testing-work' into branch-1.13.2
* better-makeflags-recognition:
make flags analysis: handle more options with args
make flags analysis: use simpler variable names
make flags analysis: whitespace changes
make flags analysis: embed in a subshell
make flags analysis: be more robust
make flags analysis: cater to GNU make 3.83 (still unreleased as of now)
tests: expose weaknesses in make flags analysis
tests: improve debugging output in checks on make flags analysis
make flags analysis: refactor, to reduce code duplication
* testing-work:
maint: targets and recipes to simplify testing on real-world packages
build: preparatory refactoring
build: tiny reduction in code duplication
maint: targets and recipes to simplify testing on real-world packages
We introduce a new section in the maintainer-specific makefile that
contains recipes to test the build system of some well-known GNU
packages with the current development version of Automake. Not the
cleanest way to do so, but good enough for the moment. We'll revisit
the matter after the 1.13.2 release (which we now hope will happen
soon).
* maint.mk (git-sv-host): New.
(SV_GIT_CF, SV_GIT_AC, SV_GIT_GL): Use it to reduce code duplication.
(ALL_PACKAGES, FEW_PACKAGES): New, lists of GNU packages to try out.
(ttp-check, ttp-check-all): New targets, do the checking with said
packages.
(ttp): New, alias for 'ttp-check'.
(ttp-all): New, alias for 'ttp-check-all'.
The code was only duplicated two times, but we are soon going to
need a third occurrence, and that would be one to much.
* Makefile.am (extend_path): New.
(update_mans): Use it instead of copying & pasting its contents.
($(srcdir)/doc/amhello-1.0.tar.gz): Likewise, and minor related
adjustments.
* lib/am/header-vars.am (am__make_running_with_option): Here. Now
that we expect to be run in a subshell, we don't have to worry about
being namespace-safe. And '$foo' is much more pleasant to read than
'$am__foo' -- and pleasant code tends to be more correct.
(am__make_dryrun, am__make_keepgoing): Adjust.
make flags analysis: cater to GNU make 3.83 (still unreleased as of now)
The current development version of GNU make (that is planned to become
GNU make 3.83, sooner or later) has changed the format its $(MFLAGS)
variable slightly, removing the space between an option and its argument:
# With GNU make 3.82, compiled from official tarball:
$ make -f- <<<'all:; @echo "$$MFLAGS"' -I none
-I none
# With development version of GNU make (Git commit b5ea49b):
$ make -f- <<<'all:; @echo "$$MFLAGS"' -I none
-Inone
This was done on purpose, in order to support more easily the new
option '-O', which takes an optional argument; see:
Which was causing a spurious failure on FreeBSD. Not particularly
surprising, given how brittle "make -e" is in general ...
* t/cxx-lt-demo.sh: Instead of forcing $(CC) to be 'false' by
exporting "CC=false" in the environment and then passing the '-e'
option to make, do so by passing "CC=false" on the make command
line, both directly and using AM_MAKEFLAGS.
* fix-pr12554:
tests: avoid a spurious error with Solaris make
subdirs: don't return false positives for the '-k' option's presence
header-vars: recognize more make flags ('-k' in particular)
header-vars: simplify how make flags are determined
tests: remove dead code from t/make-dryrun.tap
header-vars: new variable $(am__running_with_option)
tests: expose bug#12554 (false positives for presence of '-k' make option)
subdirs: don't return false positives for the '-k' option's presence
This change fixes automake bug#12554.
The old implementation of the code descending into $(SUBDIRS)
entries used the following snippet to decide whether make is running
with the '-k' a.k.a. '--keep-going' option, and thus whether a failure
in a subdirectory should prevent the descent in the following ones:
fail= failcom='exit 1'; \
for f in x $$MAKEFLAGS; do \
case $$f in \
*=* | --[!k]*);; \
*k*) failcom='fail=yes';; \
esac; \
done
It's clear that the second pattern in the 'case' construct could possibly
match false positives, for examples in these two cases:
make check TESTS="x.test k.test"
make -I /usr/local/kool-fragments
which are somewhat unusual, but not invalid. So we need a more resilient
implementation, as we did for the detection of the '-n' flag.
This implementation is now provided by the new private macro
'$(am__make_keepgoing)' (introduced in recent commits); so we can
just us that to fix the bug.
* lib/am/subdirs.am ($(am__recursive_targets)): Use '$(am__make_keepgoing)'
instead of ad-hoc and more brittle checks.
* t/list-of-tests.mk (XFAIL_TESTS): Remove the now-passing test case
't/subdir-keep-going-pr12554.sh'.
Reported-by: Michael Daniels <mdaniels@rim.com> Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
header-vars: recognize more make flags ('-k' in particular)
* lib/am/header-vars.am (am__running_with_option): Here.
Few improvements to comments, while at it.
(am__make_keepgoing): New, tell whther make is being runt with
the '-k' option.
* t/make-keepgoing.tap: New test.
* t/list-of-tests.mk: Add it.
* t/make-dryrun.tap: Minor edits to keep it more in sync with
the new test.
* syntax-checks.mk (sc_tests_overriding_macros_on_cmdline): Here.
The test 'make-dryrun.sh' has been since long rewritten as the TAP
test 'make-dryrun.tap', and no longer spuriously triggers this
maintainer check.
header-vars: simplify how make flags are determined
Actually, son far only the '-n' option ("dry mode") was detected,
but this change will allow us to soon detect more options.
* lib/am/header-vars.am (am__running_with_option): Even when $MAKEFLAGS
appears to contain definition of variables with embedded whitespace,
use simple textual pre-processing over $MAKEFLAGS rather than tricky
recursive invocations of make to determine whether the '-n' option was
given. This is enough to correctly handle all the tricky usages covered
in the testsuite.
* t/nodep.sh: Adjust to avoid a spurious failure.
header-vars: new variable $(am__running_with_option)
This is a preparatory refactoring, needed by later patches.
No semantic change is intended.
* lib/am/header-vars.am (am__running_with_option): New, contains
shell code that determines whether the current make instance is
running with a given one-letter option (e.g., -k, -n) that takes
no argument. Actually, the only supported option at the moment
is '-n' (support for '-k' will be added soon).
(am__make_dryrun): Rewrite as a thin wrapper around
'$(am__make_running_with_option)'.
tests: expose bug#12554 (false positives for presence of '-k' make option)
The current implementation of the code descending into $(SUBDIRS)
entries uses the following snippet to decide whether make is running
with the '-k' a.k.a. '--keep-going' option, and thus whether a failure
in a subdirectory should prevent the descent in the following ones:
fail= failcom='exit 1'; \
for f in x $$MAKEFLAGS; do \
case $$f in \
*=* | --[!k]*);; \
*k*) failcom='fail=yes';; \
esac; \
done
It's clear that the second pattern in the 'case' construct can possibly
match false positives, for examples in these two cases:
make check TESTS="x.test k.test"
make -I /usr/local/kool-fragments
which are somewhat unusual, but not invalid. So we need a more resilient
implementation, as we did for the detection of the '-n' flag.
But alas, such an implementation seems quite tricky to obtain in portable
make. So for the moment we content ourselves with exposing the bug, with
the hope of being able to fix soon enough.
* t/subdir-keep-going-pr12554.sh: New test.
* t/list-of-tests.mk (handwritten_TESTS, XFAIL_TESTS): Add it.
* THANKS: Update
Reported-by: Michael Daniels <mdaniels@rim.com> Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
tar: format 'ustar' cannot support UID/GID longer than 21 bits
See automake bug#8343 and bug#13588.
POSIX 1988 'ustar' format is defined with *fixed-size* fields. There
is notably a 21 bits limit (2097151) for the UID and the GID.
Tom Rini tom_rini@mentor.com says (in bug#8343):
When the user has a UID or GID that is larger than the ustar format
supports, pax does not error out gracefully in some cases (FC13).
Marc Herbert <marc.herbert@intel.com> adds (in bug#8343):
When "configure" is run by a user with an UID bigger than 21 bits,
BSD pax 3.4 aborts when trying to create the 'conftest.tar' test
archive and leaves an empty or corrupted conftest.tar file behind.
In the next step, pax tries to extract this incomplete or corrupted
archive and this *** hangs the whole ./configure script ***.
Note: GNU cpio 2.9 pretends to pass the test but it is a LIE: it
silently truncates any big UID to its lower 21 bits. I don't know
what can be the consequences of this lie.
I think there is currently a design issue in automake/m4/tar.m4
considering that a ustar archive should should *never* succeed when
./configure is run from a big user ID.
Months later, Petr Hracek <phracek@redhat.com> reports a similar issue
(in bug#13588) for Fedora 17:
I am trying to solve problem in case a user is created with big
UID and during configuration pax hangs with message
ATTENTION! pax archive volume change required.
Ready for archive volume: 1
Input archive name or "." to quit pax.
Archive name >
Time to fix this issue, on the line of a preliminary patch provided by
Petr Hracek in bug#13588. The final patch ended up being remarkably
different from that original proposition, though.
* m4/tar.m4 (_AM_PROG_TAR): If the UID or GID of the current user is
too high (> 2097151), the 'ustar' format cannot work. Adjust checks
accordingly. Some related code reordering and clean-up.
* t/tar-ustar-id-too-high.sh: New test.
* t/list-of-tests.mk: Add it.
* t/tar.sh: While at it, tweak and enhance a little.
* t/tar2.sh: Likewise.
* t/tar3.sh: Likewise.
* t/tar-override.sh: Likewise.
* NEWS: Update.
* THANKS: Likewise.
Helped-by: Pavel Raiskup <praiskup@redhat.com> Helped-by: Petr Hracek <phracek@redhat.com> Helped-by: Marc Herbert <marc.herbert@intel.com> Helped-by: Tom Rini <tom_rini@mentor.com> Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
* fix-part-pr13832:
tests: avoid possible autotools caching issues (automake bug#13832)
* fix-pr13760:
dry-run: don't get confused by '-I' option
dry-run: with GNU make, prefer $(MFLAGS) over $(MAKEFLAGS)
header vars: can determine whether we are running under GNU make
* fix-doc-pr14019:
docs: issues with configure substitutions in TESTS
* news-wording-improve:
NEWS: improve wording for automake bug#13514 fix
docs: issues with configure substitutions in TESTS
Motivated by automake bug#14019.
* doc/automake.texi: Currently, when the parallel test harness is in use,
configure substitutions in TESTS definitions can only work if they expand
to tests that ends with a suffix listed in TEST_EXTENSIONS. Document this
limitation.
Fixes automake bug#13760 for non-GNU make implementations that still
support the option '-I'. So far, the only such make implementation
are FreeBSD (8.x) make and NetBSD (5.x) make.
* lib/am/header-vars.am (am__make_dryrun): If a non-GNU make is being
used, try to handle the '-I' option in $MAKEFLAGS correctly. For GNU
make, that is already done by the proper use of the $MFLAGS variable.
dry-run: with GNU make, prefer $(MFLAGS) over $(MAKEFLAGS)
Fixes automake bug#13760 for GNU make.
* lib/am/header-vars.am (am__make_dryrun): If GNU make is being used, rely
on the contents of the $(MFLAGS) variable rather than of the $(MAKEFLAGS)
to decide whther make is being executed in "dry run" mode. Not only this
makes the code possibly faster and less brittle, but also fixes automake
bug#13760 (at least when GNU make is in use).
* t/make-dryrun.tap: Adjust: some tests that were xfailing now pass.
header vars: can determine whether we are running under GNU make
This is mostly a preparatory patch in view of future changes.
* lib/am/header-vars.am (am__is_gnu_make): New, contains shell code that
determines whether we are running under GNU make.
* t/make-is-gnu.sh: New test.
* t/list-of-tests.mk: Add it.
projects / thirdparty / automake.git / log
