check: fix test-suite.log formatting with NetBSD sed
* lib/am/check.am (am__rst_section): Tweak to ensure en extra final
newline is printed also with NetBSD sed. Issue revealed by a
failure of 'test-metadata-global-result.test'.
* tests/rst-formatting.test: New test, checking the internal
variables '$(am__rst_title)' and '$(am__rst_section)'.
* tests/list-of-tests.mk: Add it.
fixup: fix test tap-signal.tap to avoid false negatives
Problem introduced by yesterday commit v1.11-1851-g79b2c6a,
"tests: avoid yet other spurious failures on NetBSD"
* tests/tap-signal.tap: Truly ensure the exit status of the make
call under test is correctly reflected in the TAP result displayed
by the corresponding 'command_ok_' invocation.
* maint:
fixup: do not tickle makefile-deps.test with the make_dryrun code
gitattributes: custom merge driver for ChangeLog not needed anymore
gitattributes: use "union" merge driver for NEWS
gitattributes: custom merge driver for ChangeLog not needed anymore
* .gitattributes: We don't maintain a version-controlled ChangeLog
file anymore, so we have no need to specify a custom merge driver
for it anymore. Update accordingly.
Peter Rosin [Tue, 14 Feb 2012 01:51:55 +0000 (02:51 +0100)]
depcomp tests: avoid redundant runs for non-libtool test
When running the dependency tests without libtool, it's
not meaningful to run the tests three-fold, the libtool
configure options are simply ignored anyway.
* tests/depcomp.sh: Only run the tests three-fold when
libtool is in use.
tests: avoid yet other spurious failures on NetBSD
* tests/tap-signal.tap: Add a "strategically placed" extra 'echo'
command and a temporary silencing of xtraces; they are required
to avoid possible garbled output with NetBSD make, which would
miss some final newlines in the expected places and thus mess up
our TAP output.
* tests/test-driver-is-distributed.test: Rewrite to avoid having to
edit configure.in; such editing was subtly botched on NetBSD (the
'AC_CONFIG_AUX_DIR' call got misplaced, ending up *before* the
'AC_INIT' call), and that caused an hard-to-track bug. Since we
are at it, extend coverage a little.
tests: avoid spurious failure with NetBSD /bin/ksh
* tests/parallel-tests-fd-redirect-exeext.test (configure.in): Tweak
so that we don't end up with the generated configure script having
two nested subshells that start with '((':
Peter Rosin [Mon, 13 Feb 2012 20:21:48 +0000 (21:21 +0100)]
tests: increase coverage for depcomp tests
On platforms requiring that no undefined symbols exist in order
to build shared libraries (e.g. Windows DLLs), you have to
explicitly declare that the libtool library you are building
does not actually have any undefined symbols, for libtool to
even try to build it as a shared library. Without that
explicit declaration, libtool falls back to a static library
only, regardless of any --enable-shared flags etc.
tests/depcomp.sh (Makefile.am): Add -no-undefined if a libtool
library (.la) is built.
(src/Makefile.am): Likewise.
* tests/dist-formats.tap: Some versions of the BSD shell wrongly
bail out when the 'errexit' shell flag is active and the left-hand
command in a "&&" list fails and that list is the *last* command
of a "case" statement. This was causing an incorrect initialization
of the '$missing_compressors' variable, and thus potential spurious
failures when non-existing programs were assumed to be available.
docs: move chapter on automake history out of main manual
The chapter on Automake history, while certainly interesting and
even fascinating, does not truly pertain to a reference manual,
so move it out from there and into its own dedicated file. This
change is made more advisable and pressing by the fact that such
and "Automake history" chapter hasn't been updated since the 1.9.6
release, so it has been becoming less faithful and useful since
then.
* doc/history.texi: New, manual on the history of the automake
package; extracted from ...
* doc/automake.texi: ... this file, with related adjustments.
* doc/Makefile.am (info_TEXINFOS): Add 'history.texi'.
(history_TEXINFOS): New, list included file 'fdl.texi'.
* Makefile.am (release-stats): Remove as obsolete.
* HACKING (Release Procedure): Don't advise anymore to run the
"release-stats" target and to update the manual with its output.
* NEWS: Update.
* .gitignore: Likewise.
depcomp tests: avoid redundant runs for libtool tests
When running libtool-oriented tests, our 'depcomp.sh' script used
to run the same checks thrice: once after having run configure
with the '--disable-shared' option, once after having run it with
the '--enable-shared' option, and once by allowing configure to
automatically select which kind of library (or libraries) to build.
This was done to offer extra coverage for the less-common depmodes
(like "aix" of "hp2") where the name of the compiler-generated
depfiles can depend on whether libtool is in use *and* on which
kind of libraries libtool is building (static, shared, or both).
About this, see commit 'Release-1-10-24-ge89ec38' of 28-03-2007,
"* lib/depcomp (aix): Rewrite depmode in the spirit of the tru64
one".
But that extra coverage came at the price of *considerably* slowing
down the depcomp libtool-oriented tests, since for each of them the
tested source tree was configured and built *twelve* times (rather
than "just" four as it is for the non-libtool case).
So, to strike a balance between coverage and speed, we now run the
libtool-oriented tests with the three different configuration modes
*only* when it is possible that the depmode that will get selected
is one of those which can actually benefit from the extra coverage.
* tests/depcomp.sh: Adjust to only run the threefold libtool tests
when that can truly be potentially useful. Add related explanatory
comments.
* tests/dist-formats.tap: The test checking whether the 'compress'
program is actually capable of compressing files relied on the
assumption that 'compress' exits with status zero if no problem
occurs; alas, this reasonable assumption doesn't hold in practice,
since 'compress' does (and is documented to) exit with status 2 if
the output is larger than the input after (attempted) compression!
Fix this by using and input that 'compress' can actually reduce in
size when compressing.
Do some related reformatting of surrounding code since we are at it.
* maint:
tests: fix a timestamp-related spurious failures
dryrun: fix regression with '$(am__make_dryrun)'
test defs: function 'is_newest' now works also with directories
Peter Rosin [Thu, 9 Feb 2012 09:25:20 +0000 (10:25 +0100)]
tests: check if 'compress' is real or fake
On Cygwin, 'compress' is provided by sharutils and is just a
dummy script that is not able to actually compress (it can
only decompress). This fake 'compress' is not usable for
our purpose - to create compressed tarballs.
* tests/dist-formats.tap (missing_compressors): Count 'compress'
as missing if it does not support the -c option.
Peter Rosin [Wed, 8 Feb 2012 14:12:40 +0000 (15:12 +0100)]
Merge branch 'msvc'
* msvc:
depcomp: try to prevent whitespace regressions
depcomp: recognize tabs as whitespace in the dashmstdout mode
depcomp: quote 'like this', not `like this'
depcomp tests: avoid spurious failures with 'dashXmstdout' depmode
* tests/gen-testsuite-part: The 'dashXmstdout' depmode cannot be
truly forced on gcc (like we were trying to do, causing spurious
failures in the process); so don't do that. This change shouldn't
actually reduce coverage anyway, since, as the comments in
'lib/depcomp' say, the 'dashXmstdout' depmode is just like
the 'dashmstdout' one, only with a different argument for the
compiler.
Peter Rosin [Wed, 8 Feb 2012 12:35:32 +0000 (13:35 +0100)]
tests: check the depmodes 'msvisualcpp' and 'msvcmsys'
* tests/defs (cygpath): New requirement, checking that cygpath
is working.
(mingw): New requirement, checking that the build system is
MSYS (in its normal MinGW mode).
* tests/gen-testsuite-part (depmodes): Add entries for depmodes
'msvisualcpp' and 'msvcmsys'.
tests: better way to compare lists in Makefile rules
With this commit, we introduce a new helper shell script for use
in the testsuite, which is meant to allow the test cases to easily
check whether two whitespace-separated lists are equal; this ability
is particularly useful to check for equality of the contents of make
variables that are expected to contain multiple whitespace-separated
words, and are defined through line continuations (or are rewritten
by automake in this way), or that contain expansion of potentially
empty variables.
Before this change, a test checking that an usage like this one:
VAR = valA
if COND1
VAR += val1 # com1
endif COND1
VAR += valC
worked as expected, couldn't use rules like:
## Doesn't work because $(VAR) expands to multiple words
verify:
test $(VAR) = "valA val1 valC"
nor like:
## Doesn't work because the final expansion of $(VAR) contains
## repeated contiguous whitespace characters (it actually
## equals "valA val1 valC", not "valA val1 valC"), and this
## is an internal detail which might change and which we don't
## want to explicitly rely on.
verify:
test "$(VAR)" = "valA val1 valC"
Instead, we had to rely on cumbersome workaround such as:
## This works, but is ugly.
verify:
test "`echo $(VAR)`" = "valA val1 valC"
or:
## This works, but is even uglier.
verify:
echo BEG: $(VAR) :END | grep "BEG: valA val1 valC :END"
Now, with the help of the new 'is' script, we can perform such a
check in a clearer and more straightforward way, as in:
## Works, and reads clearly.
verify:
is $(VAR) == valA val1 valC
* tests/is: New helper shell script, telling whether two whitespace
separated lists are equal.
* Makefile.am (EXTRA_DIST): Add it.
* tests/colneq2.test: Use the new helper script, and accordingly
get rid of older, more cumbersome idioms.
* tests/cond11.test: Likewise.
* tests/cond16.test: Likewise.
* tests/cond18.test: Likewise.
* tests/cond22.test: Likewise.
* tests/cond31.test: Likewise.
* tests/cond38.test: Likewise.
* tests/test-logs-repeated.test: Likewise.
* tests/objext-pr10128.test: Likewise.
* tests/programs-primary-rewritten.test: Likewise.
* tests/substre2.test: Likewise. Also ...
(configure.in, Makefile.am): Add a couple of hack to avoid having
to require (and run) a C compiler; accordingly ...
($required): ... remove this.
* tests/exeext4.test: Likewise.
* tests/substref.test: Likewise. Also ...
(hello.c): Use ": >" rather than "cat <<EOF" to generate it,
since it's meant to be empty anyway.
* tests/cond4.test: Use the new helper script, and accordingly
get rid of older, more cumbersome idioms. Avoid some unnecessary
uses of "make -e" since we are at it.
* tests/cond19.test: Likewise.
* tests/cond32.test: Likewise.
* tests/cond6.test: Use the new helper script, and accordingly
move some checks in the Makefile.am.
Avoid unnecessary execution of automake remake rules by manually
"touching" aclocal.m4
tests: improve and rework tests on dependency tracking
Fixes automake bug#10434. Suggestion by Peter Rosin.
The 'depcomp.tap' test case worked by trying to unconditionally
force the compiler in use by the testsuite to use, one by one, *all*
the dependency modes known by the 'depcomp' script, and, for each
such forced mode that was compatible enough with said compiler not
to cause breakage in the basic compilation rules, checking that it
was *also* good enough not to break remake rules in VPATH builds.
This seemed a good approach when this test was first introduced, as
it apparently increased coverage for the less used and less tested
dependency-tracking modes. But in the log run it turned out the
approach was actually in part too brittle, causing some annoying
spurious failures (as with the Tru64 depmode forced on GCC, see
automake bug#10434), and partly too forgiving, since, for some of
the more corner-case dependency modes, the 'depcomp' script simply
reverts to silently disabling dependency tracking when an error is
encountered (this happened e.g., with the Tru64 depmode forced on
the Sun C compiler 5.9), so that a passing test means nothing, and
only gives a false sense of security.
As Peter Rosin put it, "it's just dead wrong to assume that feeding
-M or -xM to the compiler (or whatever other random stuff 'depcomp'
might do) and not get an error is the same as dependencies magically
appearing".
So we get rid of this wrong approach, and in the process proceed to
a complete overhaul of many of the tests on automatic dependency
tracking, extending the offered coverage and rationalizing their
organization.
* tests/depcomp.sh: New helper script, used by several new
autogenerated tests.
* tests/gen-testsuite-part: Generate several tests based on the
new 'depcomp.sh' script. Emit makefile code that declares their
dependency on that script, and that extends EXTRA_DIST in order
to distribute it.
* tests/depmod.tap: Remove.
* tests/depend2.test: Remove, it has been subsumed by the new
autogenerated test 'depcomp-disabled.tap'.
* tests/depcomp3.test: Remove, it has been subsumed by the new
autogenerated test 'depmod-dashmstdout.tap'.
* tests/depcomp3.test: Remove, it has been subsumed by the new
autogenerated test 'depmod-lt-makedepend.tap'.
* tests/depcomp6.test: Remove, it has been subsumed by the new
autogenerated test 'depmod-auto.tap'.
* tests/depcomp7.test: Remove, it has been subsumed by the new
autogenerated test 'depmod-lt-auto.tap'.
* tests/depcomp5.test: Remove as obsolescent.
* test/list-of-tests.mk: Adjust.
* .gitignore: List the new autogenerated tests.
Fixes automake bug#10737. Report from Larry Daniel.
See also automake bug#8365 and commit 'v1.11-329-gd4df619' of
30-03-2011, "tests: fix timestamp-related failures".
* tests/aclocal5.test: Sleep before modifying m4 files that should
trigger remake rules. This fixes an hard-to-hit timestamp-related
race condition.
* THANKS: Update.
In commit v1.11-683-gda0964e of 05-02-2012, we introduced a new
variable '$(am__make_dryrun)' that could be used in recipes to
determine whether make is running in dry mode (e.g., as with
"make -n"). Unfortunately, the idiom we used fails to take into
account the case in which $(MAKEFLAGS) contains one or more variable
definitions whose value is a whitespace-separated list; for example,
if we invoke make as:
make check TESTS="n1.test n2.test"
then the better make implementations out there (at least modern
GNU make and BSD make) will export MAKEFLAGS to the following
value:
TESTS=n1.test\ n2.test
So a shell loop like the one we used in $(am__make_dryrun):
for flag in $$MAKEFLAGS; do ...
won't behave as expected: the shell word-splitting rules will break
up the entry for TESTS into the two distinct entries "TESTS=n1.test\"
and "n2.test", and this second entry will goad our $(am__make_dryrun)
code into thinking that make is performing a dry run!
So now we simply loop over the expanded value of $(MAKEFLAGS).
This solves the regression, but alas, is more brittle in case the
users passes on the command line a macro value containing unbalanced
" or ' quotes, or shell metacharacters like, say, '`' or '('. This
should almost never happen though, so we don't worry about it now;
we will revisit the issue only when and if we receive bug reports in
this area.
* lib/am/header-vars.am (am__make_dryrun): Fix.
* tests/make-dryrun.test: New test.
* tests/list-of-tests.mk: Add it.
test defs: avoid spurious XPASS results with NetBSD make
* tests/defs (make_can_chain_suffix_rules): Ensure to properly
declare the '.SUFFIXES:' in the temporary Makefile used to check
whether $MAKE is able to automatically chain suffix rules. This
avoids spurious XPASS results with (at least) NetBSD make in some
'suffix*.tap' tests; in a sense also, this also makes the probing
of $MAKE capabilities more faithful, since the Automake-generated
Makefiles are expected to properly declare the '.SUFFIXES:' as
well.
Several typofixes in documentation and comments, as
suggested by the "codespell.py" script. Reference:
<http://git.profusion.mobi/cgit.cgi/lucas/codespell/>
* tests/defs (rst2html): Remove, it's not needed anymore now that
the old code supporting the ".log -> .html" conversion for the
parallel-tests harness has been moved in contrib. This removal
should have actually be done in commit v1.11-1757-g0a9d201 of
17-01-2012, 'check: move ".log -> .html" conversion in contrib'.
* tests/defs (gmake, native, rst2html): Some versions of the BSD
shell wrongly bail out when the 'errexit' shell flag is active and
the left-hand command in a "&&" list fails, if that list is the
*last* command of a "for" or "while" loop. Work around this issue.
* tests/lex3.test: Likewise.
* tests/check12.test: Likewise.
* tests/yacc-basic.test: Likewise.
* tests/lex-noyywrap.test: Likewise.
* tests/parallel-tests2.test: Likewise.
* tests/lex-lib-external.test: Likewise.
* tests/check-no-test-driver.test: Likewise.
* tests/test-metadata-recheck.test: Likewise.
* tests/parallel-tests-dry-run-1.test: Likewise.
* tests/parallel-tests-dry-run-2.test: Likewise.
* tests/parallel-tests-unreadable.test: Likewise.
* tests/test-driver-custom-no-extra-driver.test: Likewise.
Problem revealed by spurious testsuite failures on a NetBSD 5.1
system and an OpenBSD 5.0 system.
test defs: function 'is_newest' now works also with directories
Reported by Jim Meyering against automake master (see bug#9147) and
by Adam Sampson against automake 1.11.3 (see bug#10730).
* tests/defs (is_newest): Call 'find' with the '-prune' option,
so that it won't descend in the directories (which could cause
spurious results).
* THANKS: Update.
parallel-tests: more uses for '$(am__make_dryrun)'
With this change we refactor some recipes of the parallel-tests
harness to make them use, where appropriate, the new internal
variable '$(am__make_dryrun)' (introduced in the 'maint' branch)
when they need to determine whether make is running in "dry mode",
i.e., with "make -n". This allows for some simplifications and
for improved uniformity.
* lib/am/check.am (check-TESTS recheck): Use $(am__make_dryrun) and
simplify accordingly.
(am--redo-logs): Delete this internal helper recipe, merged into ...
($(TEST_SUITE_LOG)): ... this, with the help of $(am__make_dryrun).
recipe accordingly.
refactor: new variable telling whether make is running in "dry mode"
We define a new variable '$(am__make_dryrun)', that can be used
in recipes to determine whether make is running in "dry mode" (as
with "make -n") or not. This is useful in rules that invoke make
recursively, and are thus executed also with "make -n" -- either
because they are declared as dependencies to '.MAKE' (NetBSD make)
or because their recipes contain the "$(MAKE)" string (GNU and
Solaris make).
* lib/am/header-vars.am (am__make_dryrun): New variable.
* lib/am/check.am (check-TESTS recheck): Use it, and simplify
recipe accordingly.
* lib/am/distdir.am (distdir): Likewise.
* lib/am/lisp.am ($(am__ELCFILES)): Likewise.
Peter Rosin [Thu, 2 Feb 2012 20:08:46 +0000 (21:08 +0100)]
tests: do not clobber the modified CC
If CC is originally a losing compiler, AM_PROG_CC_C_O will
modify CC. "$MAKE -e" will then clobber this modified CC
and reintroduce the raw losing compiler causing the test
to fail, as subdir-objects is in effect.
tests/check8.test: Drop -e from the $MAKE invocation, since
AM_COLOR_TESTS is not specified in the Makefile and -e is
therefore not needed.
tap/perl: handle missing or non-executable scripts better
This change improves how our Perl-based TAP driver handles
non-runnable test scripts (meaning they might be not executable,
or not readable, or even not exist). In particular, it makes the
driver deterministically display a clear "ERROR" result instead
of possibly dying with diagnostic from 'TAP::Parser' internals,
and prevents it from displaying spurious "missing TAP plan" errors.
Moreover, with this change, some testsuite failures present only
with newer perl versions (e.g., 5.14) are fixed. See automake
bug#10418.
* tests/tap-bad-prog.tap: When testing the perl implementation of
the TAP driver, and when the perl interpreter offers a good-enough
'IPC::Open3::open3' function, expect it not to display spurious
"missing TAP plan" diagnostic if the error is actually due to a
non-runnable test script.
* lib/tap-driver.pl (start): Removed, broken up into ...
(setup_io): ... this ...
(setup_parser): ... and this, which now tries to catch and report
errors in launching the test scripts.
(finish): New, used by both 'main' and 'setup_parser'.
(main): Adjust.
Peter Rosin [Wed, 1 Feb 2012 14:47:45 +0000 (15:47 +0100)]
tests: put AM_PROG_CC_C_O before AC_OUTPUT to help losing compilers
When AM_PROG_CC_C_O is after AC_OUTPUT, the compile script
is not used even if needed, causing testsuite fails if
libtool is not used.
* tests/depcomp8a.test: Uncomment the AM_PROG_CC_C_O macro
in its correct location, as indicated...
(configure.in): ...with this comment.
* tests/depcomp8b.test: Sync with tests/depcomp8a.test.
* lib/am/comp-vars.am: Remove. This file has stopped being useful
since commit 'Release-1-4d-206-g3334f9a' of 09-04-2001, and should
have been removed back then. But apparently only its contents were
removed, leaving the file empty but still present in the repository.
Or maybe this situation is the consequence of a minor blunder in
the conversion from CVS to Git; either way, this is not a big deal.
* maint:
tests: do not assume the object file extension is .o
tests: avoid spurious failure of 'transform2.test' on Cygwin
tests: avoid spurious failure of deleted-am.test with FreeBSD make
tests: avoid possibly undeserved PASS from check8.test
warnings: more precise category and message for one warning
release: revamp rules to tag and upload the releases
amversion: add missing dependency
hacking: update advice w.r.t. synced files
hacking: don't reference ChangeLog anymore
+ Extra non-trivial edits:
* tests/suffix8.tap: Copy in (by hand) the modifications done
to 'suffix8.test' on maint, i.e., the addition of an explicit
'.y_.obj:' suffix rule to Makefile.am.
Peter Rosin [Mon, 30 Jan 2012 22:27:06 +0000 (23:27 +0100)]
Merge branch 'msvc'
* msvc:
ar-lib: ignore the verbose modifier instead of erroring out
scripts: cherry-pick recent changes from master
tests: add missing dependency for some 'ar-lib*.test' tests
* lib/Makefile.am (install-data-hook): Correctly quote $(DESTDIR)
occurrences.
* Makefile.am (install-exec-hook, uninstall-hook): Likewise.
(autodiffs): Prefer '$(am__cd)' to plain 'cd', where warranted.
* doc/Makefile.am ($(srcdir)/amhello-1.0.tar.gz): Likewise. Also,
prefer using AC_SUBST'd $(abs_top_builddir) over obtaining it at
runtime from $(top_builddir).
* syntax-check.mk (sc_mkinstalldirs): Tweak to whitelist known
harmless occurrences of the checked-against usages.
(sc_no_for_variable_in_macro): Likewise.
* tests/CheckListOfTests.am (maintainer-check-list-of-tests): Tweak
to avoid spuriously triggering the 'sc_no_for_variable_in_macro'
maintainer check. Prefer '$(am__cd)' to plain 'cd' when warranted.
maintcheck: take advantage of some GNU make features
We can do so now that our maintainer checks require GNU make
unconditionally.
* syntax-check.mk (ams, xtests): Redefine as "immediate variables",
using the GNU make $(shell ...) builtin.
(maintainer-check-list-of-tests): Take advantage of GNU make "-C"
option. Don't use $(AM_MAKEFLAGS), we shouldn't need it with GNU
make.
build: require GNU make in order to run the maintainer checks
This is not a real regression in portability, since most maintainer
check rules already assumed GNU grep, and were anyway only meant to
run on the developers' systems (where we can safely assume the
presence of GNU make).
This change will allow us to take advantage, in future changes, of
more GNU make features, thus simplifying or optimizing some of our
maintainer rules.
* GNUmakefile: New, including 'Makefile' and 'syntax-checks.mk'.
* configure.ac (AC_CONFIG_LINKS): Link it in the builddir.
* Makefile.am (EXTRA_DIST): Distribute 'GNUmakefile' and
'syntax-checks.mk'.
Move all syntax-check rules and auxiliary variables into ...
* syntax-checks.mk: ... this new file, with some adjustments.
A small refactoring in our syntax-check to avoid code duplication
and to ensure more uniform checking. This change introduces some
new failures in syntax-check rules (both real and spurious), that
will be fixed in follow-up changes and code reorganizations.
* Makefile.am (ams): New variable, holding the list of all the
'*.am' fragments in he automake source tree.
(sc_no_brace_variable_expansions): Use it.
(sc_rm_minus_f): Likewise.
(sc_no_for_variable_in_macro): Likewise.
(sc_mkinstalldirs): Likewise.
(sc_pre_normal_post_install_uninstall): Likewise.
(sc_cd_in_backquotes): Likewise.
(sc_cd_relative_dir): Likewise.
(sc_tests_make_without_am_makeflags): Likewise.
(sc_tests_plain_egrep_fgrep): Likewise.
(sc_mkdir_p): Likewise.
Peter Rosin [Mon, 30 Jan 2012 19:57:12 +0000 (20:57 +0100)]
tests: do not assume the object file extension is .o
* tests/specflg7.test: Dig out the object file extension from the
OBJEXT makefile variable.
* tests/substref.test: Likewise.
* tests/specflg8.test: Likewise. Also check that the false-true
object is created instead of checking the true-true object twice.
* tests/suffix8.test: Add a chain rule for the case where the
object file extension is .obj.
tests: avoid spurious failure of 'transform2.test' on Cygwin
On newer Cygwin versions (at least 1.7.x), the 'transform2.test'
test has been failing spuriously; the gist is the following:
some *purposefully* rigged install rules there try something
like:
and the second install command fails (trying to overwrite the
'.../inst/bin/foo.exe' file, likely due to overly aggressive
appending of '.exe' suffix when copying/renaming Windows
executables). Since this is a Cygwin issue rather than an
Automake one (and since the use case we are testing is a really
corner-case anyway, making it unworthy to attempt to work
around it in automake proper), we simply hack the test case
to avoid the failure.
tests: avoid spurious failure of deleted-am.test with FreeBSD make
* tests/deleted-am.test: Sleep between the removal of the included
'.am' fragments and the subsequent "make" calls, to ensure that the
remake rules kick in. This is required to avoid racy spurious
failures (~ 60% of the time) with FreeBSD make.
tests: avoid possibly undeserved PASS from check8.test
* tests/check8.test: Strengthen grepping of "make check" output
where we know no problem with VPATH rewrites can take place. This
has the advantage of ensuring that we won't match also "sub/bar"
when looking for "bar" during the uncolorized tests.
tests: work around MinGW/MSYS issue in fd redirections
Some checks on $(AM_TESTS_FD_REDIRECT) were failing on MSYS, likely
because system calls like "write(9, ...)" simply doesn't work for
MinGW-compiled programs. Similar usages work for the shell scripts
though, since /bin/sh is an MSYS program and thus is a lot more
POSIX-ish than most MinGW-compiled programs.
The best fix for this issue is to separate the checks using shell
scripts as dummy test cases from the checks using compiled programs
for the same purpose, and skip these latter checks on MinGW.
This change fixes automake bug#10466. Report by Peter Rosin.
* tests/parallel-tests-fd-redirect.test: Move checks using compiled
C programs as test cases to ...
* tests/parallel-tests-fd-redirect-exeext.test: ... this new test.
* tests/list-of-tests.mk: Update.
warnings: more precise category and message for one warning
If automake detected an usage like "AC_CONFIG_FILES([./Makefile])"
in configure.ac, it warned that such an usage was unportable to
non-GNU make implementations. But the truth is actually worse:
that is actually *unportable to GNU make* itself, since it breaks
the automatic remake rules in subtle ways.
So we now reveal this breakage in a new test case, and enhance
the warning by giving it a more precise and correct wording, and
by moving it from the category 'portability' to the category
'unsupported'.
* automake.in (scan_autoconf_config_files): Improve the warning.
* tests/conffile-leading-dot.test: New test.
* tests/list-of-tests.mk: Add it.
release: revamp rules to tag and upload the releases
The older Makefile rules used to create and tag the releases were
based on an approach we now consider flawed: they over-mechanized
some delicate operations that are better performed manually *and*
double-checked by a developer, and at the same time they did not
run enough safety checks.
* Makefile.am (GIT, version_rx, stable_version_rx, beta_version_rx,
match_version, git_must_have_clean_workdir, determine_release_type):
New variables.
(git-release, git-dist): Remove, they are superseded by ...
(git-tag-release, git-upload-release): ... these new targets.
* HACKING: Update and improve advices and explanations about files
in the automake repository that are now owned by automake, but
mirrored from other upstreams. Also, don't list these files
explicitly, rather point the reader to the $(FETCHFILES) variable
in Makefile.am.
* Makefile.am (FETCHFILES): Don't state that "there should be a
lot more here", as this is not true anymore today. Only 'COPYING'
must be synced by hand.
* HACKING: Don't reference the ChangeLog file anymore, since that
is now generated from the git commit messages, not maintained by
hand. So remove advice that is obsolete, and speak about "git
commit message" instead of "ChangeLog entry" for advice that is
still relevant.
This follows up on commit v1.11-665-gc5df21e of 2012-01-17,
"multilib: deprecate, will be moved to contrib". See also:
<http://lists.gnu.org/archive/html/automake-patches/2012-01/msg00109.html>
* NEWS: Update.
* automake.in ($seen_multilib): Remove.
(scan_autoconf_traces): Don't trace 'AM_ENABLE_MULTILIB', and don't
handle it anymore.
(handle_multilib): Remove.
(generate_makefile): Don't call it anymore.
* doc/automake.texi: Remove documentation about multilib support,
related macros, and helper files.
* m4/multi.m4: Delete.
* m4/Makefile.am (dist_automake_ac_DATA): Remove it.
* lib/am/multilib.am: Delete.
* lib/am/Makefile.am (dist_am_DATA): Remove it.
* contrib/multilib/multilib.am: New file, adapted from extracts of
a Makefile.in generated with automake multilib support. We did
this instead of moving and editing 'lib/am/multilib.am' because it
allows us to license this file with a liberal license that will
permit users to copy-and-paste it in non-GPLed Makefile.am files
too).
* lib/symlink-tree, lib/config-ml.in: Move ...
* contrib/multilib: ... in here.
* lib/Makefile.am (dist_script_DATA, dist_pkgvdata_DATA): Update.
* contrib/multilib/README: New file.
* contrib/Makefile.am (EXTRA_DIST): Add the files created or moved
in 'contrib/multlib'.
* tests/multilib.test: Update and enhance a little.
* tests/help-multilib.test: Likewise.
* compilers-for-testsuite:
readme: how to run the testsuite with cross-compilers
tests: no need to unset CFLAGS in tests requiring 'gcc' anymore
test defs: allow compilers to be auto-selected on user's request
test defs: substitute compilers and flags found at configure time
test defs: setup `*FLAGS' variables for GNU compilers
configure: search generic compilers for use in the tests
readme: how to run the testsuite with cross-compilers
* tests/README: Suggest a better way to run the automake testsuite
with cross-compilers -- that is, configuring the Automake source
tree with proper `--build' and `--host' configure option. And
yes, specifying both these options (not only `--host') is indeed
required to avoid spurious failures in corner cases.
When you call configure with the `--host' option but without the
`--build' option, configure tries to auto-detect whether you are
cross-compiling or not, by trying to run a generated executable.
That test might spuriously "succeed" in some corner cases (e.g.,
Cygwin is able to run non-Cygwin apps). In fact, generally, it
can be the case that a cross-compilation is not detected as a
cross anymore just because someone has installed an emulator; as
an example, think of what can happen on a GNU/Linux system that
is configured (through the use of the binfmt_misc kernel module)
to execute PE executables (compiled for MS-DOS or Windows) through
Wine *automatically*. In conclusion, configure needs to be used
as recommended in the documentation (i.e., by specifying *both*
`--host' and `--build' instead of just one of them) to not have
the build fall into any of a number of weird traps.
* tests/defs (cross_compiling): Improve comments.
tests: no need to unset CFLAGS in tests requiring 'gcc' anymore
* tests/ccnoco.test: Since this test have "gcc" in $required,
there is no need to manually nullify the CFLAGS variable, since
now `tests/defs' should automatically re-define that to a value
appropriate for gcc.
* tests/ccnoco3.test: Likewise.
test defs: allow compilers to be auto-selected on user's request
In the recent changes, by pre-setting the required compilers for
the `configure' scripts launched by the test scripts, we have
slightly reduced the coverage of autoconf/automake code aimed at
automatically detecting said compilers. This commit restore such
coverage, by allowing the user to instruct the testsuite *not* to
preset the testsuite compiler(s).
* tests/defs (cc): If $CC is set to the special value "autodetect"
or "autodetected", don't export the configure-detected CC, CFLAGS
and CPPFLAGS variables; rather, unset them.
(c++): Likewise, but for CXX, CXXFLAGS and CPPFLAGS instead.
(fortran): Likewise, but for FC and FCFLAGS instead.
(fortran77): Likewise, but for F77 and FFLAGS instead.
(require_compiler_): New function, to reduce code duplication.
test defs: substitute compilers and flags found at configure time
* tests/Makefile.am (do_subst): Also substitute CC, CXX, F77, FC,
CPPFLAGS, CFLAGS, CXXFLAGS, FCFLAGS and FFLAGS.
* tests/defs-static.in: Define those variables, allowing for
overrides from the environment.
* tests/defs (for tool in $required): Export (subsets of) those
variables when the stuff in `$required' calls for it.
Add related explanatory comments.
test defs: setup `*FLAGS' variables for GNU compilers
* configure.ac: Setup some `*FLAGS' variables for use by the
GNU compilers in our testsuite. For example, use `GNU_CFLAGS'
instead of `CFLAGS', and so on for similar variables. This
is especially useful in case the compilers found or defined
at configure time are not the GNU ones.
* tests/defs-static.in: Initialize those same variables with
the values set at configure time, but allowing overrides
from the environment.
* tests/Makefile.am (do_subst): Process configure-style
substitutions of those variables (e.g., `@GNU_CFLAGS@').
* tests/defs: When a GNU compiler is required, override the
corresponding generic `*FLAGS' variable with the GNU-specific
variant (e.g., redefine `$CFLAGS' to take the value of
`$GNU_CFLAGS').
configure: search generic compilers for use in the tests
* configure.ac: Look for "generic" C, C++ and Fortran compilers,
with the aim of starting to use them in the testsuite (this will
be done in future changes). This is more tricky than it seems,
since we don't want to abort the whole configure script even if
no one of those compilers is available (after all, they're only
needed by the testsuite, not to build automake), but currently
autoconf doesn't offer an easy way to obtain this behaviour.
We prefer non-GNU compilers to the GNU ones, to ensure better
coverage "in the wild".
* tests/aclocal-install-absdir.test: Do not set the 'errexit' shell
flag, as it is already set by 'tests/defs'.
* tests/distcheck-pr10470.test: Likewise.
* tests/objext-pr10128.test: Likewise.
* tests/parallel-tests-dry-run-1.test: Likewise.
* tests/silent-nested-vars.test: Likewise.
* tests/tar-override.test: Likewise.
* tests/vala-mix.test: Likewise.
* tests/vala-vpath.test: Likewise.
* maint:
tests: fix spurious failure due to autom4te caching
vala: fix name of temporary file used in vala rules
vala tests: add missing 'valac' requirement, and other minor fixlets
news: fix grammaro
tests: fix spurious failure due to autom4te caching
* tests/lzma.test: Remove stale autom4te.cache directories, to
prevent racy, spurious failures (using 'aclocal --force' was
not enough, since the cache was still picked up by the following
automake call).
dist tests: missing 'compress' program was causing spurious failures
Fixes automake bug#10575.
The compress(1) tool is becoming anachronistic, and thus is not
installed by default on modern distros (e.g., Fedora 16). We
can't thus unconditionally assume it exists on every reasonable
portability target.
* tests/dist-formats.tap ($missing_compressors): When defining
this, don't assume anymore that compress(1) is unconditionally
available.
(Parallel compression): Skip this sub-test if compress(1) is
unavailable.
Since we are at it, fix a couple of unrelated buglets: a typo
in a test name (s/distcheck/ark-exists/), and some attempts to
remove directories with "rm -f".
vala: fix name of temporary file used in vala rules
Unquoted `@' characters in a "..." string in the automake script
were causing slightly wrong rules to be emitted in the generated
Makefile.in; i.e., rules like:
vala tests: add missing 'valac' requirement, and other minor fixlets
See also automake bug#10575.
* tests/vala-mix.test ($required): Add 'valac'; this will avoid
spurious failures on systems lacking a Vala compiler.
Add some explicative comments for a couple of non-obvious make
calls.
Make grepping of "make -n" slightly stricter, to avoid potential
false positives.
* maint:
fixup: distribute 'contrib/multilib/multi.m4'
multilib: deprecate, will be moved to contrib
fixlet: flags for Fortran77 compiler are in FFLAGS, not F77FLAGS
cosmetics: fix a botched comment in a maintainer check
As of 2012-01-17, according to Google codesarch, almost no active
package is using the 'multilib' feature offered by automake.
The only major exception seems to be GCC... But on a closer look,
it become clear that GCC basically carries its own version of
multilib support. In fact, Automake syncs its 'config-ml.in' and
'symlink-tree' scripts from GCC; and the GCC repository contains a
version of the 'multi.m4' file that is *more* updated than the one
in the automake repository (the former having being modified the
last time in 2008, the latter only in 2006).
The 'multilib' feature was anyway hardly documented at all, only
being briefly cited in the manual as an "obscure feature", "still
experimental", that was only for users "familiar with multilibs"
and which "can debug problems they might encounter". We expect
such users to be motivated and knowledgeable enough to make the
minor adjustments required to start using the contrib version of
multilib, if they really need to.
* NEWS (Future backward incompatibility): Update.
* doc/automake.texi: Deprecate multilib support. State that it
will be removed from automake core in the next major release.
* m4/multi.m4 (AM_ENABLE_MULTILIB): Deprecate. If called, now
gives a proper warning in the 'obsolete' category (while still
retaining its former behaviour for the rest).
* tests/multilib.test: Update.
* contrib/multilib/multi.m4: New, verbatim copy of the earlier
version of multi.m4, without the new deprecation warning.
* Makefile.am (fetch): Don't sync the 'config-ml.in' file nor
the 'symlink-tree' script from GCC SVN repository anymore.
(FETCHFILES): Adjust.
(WGET_GCC): Remove, it's not needed anymore.
fixlet: flags for Fortran77 compiler are in FFLAGS, not F77FLAGS
This change fixes automake bug#10555.
Note that the bug was a minor one, since it didn't affect the
compilation rules generated by automake, but only only the "hints"
printed by automake in some error messages (e.g., "The usual way
to define `FFLAGS' is to add AC_PROG_F77 to configure.ac").
* lib/Automake/Variable.pm (%_ac_macro_for_var): The code generated
by AC_PROG_F77 uses FFLAGS, not F77FLAGS, as the variable where to
look for switches for the Fortran 77 compiler: adjust accordingly.
cmdline parsing: move into a dedicated perl module
With this change, we delegate most of the automake and aclocal code
for command-line options parsing to a new module "Automake::Getopt".
This allows better code sharing between automake and aclocal, and
also with Autoconf, which will sync the new module from us. See
also autoconf commit 'v2.68-120-gf4be358' (2012-01-17, "getopt: new
Autom4te::Getopt module"), and this mailing list discussion:
<http://lists.gnu.org/archive/html/autoconf-patches/2012-01/msg00033.html>
This change might interact with the behaviour described in automake
bug#7434; for example, starting from now, "automake -Wfoo --version"
will cause automake to emit diagnostic like "unknown warning
category 'foo'" before actually printing the version number and
exiting. This is not a big deal in practice, and the code sharing
and simplifications introduced by this patch is certainly worth it.
Still, we should revisited the issue in the future.
* lib/Automake/Getopt.pm: New module, basically a slightly-edited
copy of the 'lib/Autom4te/Getopt.pm' file from the autoconf devel
repository (commit v2.68-120-gf4be358). It defines and exports ...
(parse_options): ... this new function.
* automake.in (parse_arguments): Use the new function.
* aclocal.in (parse_arguments): Likewise.
* lib/Automake/Makefile.am (dist_perllib_DATA): Add the new file.
* tests/getopt.test: Remove.
* tests/list-of-tests.mk: Update.
tests: don't use custom TAP diagnostic in our own tests
We simply don't need it (yet), we are not actually using it (the
set up for using it is there in the Makefile, but no test script
makes actual use of that).
Two tests checking "make -n" behaviour with the parallel-tests
harness has been introduced in the Automake repository in the past:
one of them ('parallel-tests-dryrun.test') in the maint branch, the
other one ('parallel-tests-dry-run.test') in the master branch.
Needless to say, their too-much-similar names can be a great source
of confusion now. Moreover, the two tests have been written during
distinct refactorings, and are meant to exercise different code
paths and/or usage scenarios, so that no one of them subsumes the
other, and they should be both kept.
Thus we simply rename both of them to avoid future confusion.
* tests/parallel-tests-dryrun.test: Renamed ...
* tests/parallel-tests-dry-run-1.test: .. to this.
* tests/parallel-tests-dry-run.test: Renamed ...
* tests/parallel-tests-dry-run-2.test: .. to this.
Add cross-reference among them.
* tests/list-of-tests.mk: Adjust.
* Makefile.am (recheck): Now that all the test scripts are in
the 'tests/' directory, its recipe can be greatly simplified,
as we don't need to deal with recursion in multiple subdirectories
anymore.
* maint:
cosmetics: move CheckListOfTests.am into tests/
tests: move all under the same hierarchy ('tests/' directory)
gitlog-to-changelog: update from upstream
changelog: don't cluster multiple entries under the same "date line"
+ Extra non-trivial edits:
* tests/Makefile.am (XFAIL_TESTS): Update with the xfailing tests
that were in 'lib/Automake/tests'. Add proper "FIXME" comment.
After the previous change 'v1.11-660-gfbeda3d', the makefile
fragment 'CheckListOfTests.am' is used only by the Makefile.am
in 'tests'; so keeping them two nearer makes sense. It also help
in reducing potential confusion, since (after that same change)
the whole Automake testsuite is expected to be (and remain)
contained into the 'tests' subdirectory.
tests: move all under the same hierarchy ('tests/' directory)
This simplifies the organization of the Automake source tree and
reduces the (lamentably high) number of Makefiles in the Automake
build system by one. It also makes the maintainer check that
verifies the consistency of list of tests more self-contained and
simpler. Finally, it might be a first step forward the transition
to a non-recursive build system for automake (if we ever decide to
go down that road fully some day).
* lib/Automake/tests: All the '*.pl' tests in here moved ...
* tests/pm: ... into this new directory.
* lib/Automake/tests/Makefile.am: Remove, its meaningful
contents moved ...
* tests/Makefile.am: ... here, with obvious adjustments.
(test_subdirs): New variable, for the sake of the recipe
of 'maintainer-check-list-of-tests'.
* CheckListOfTests (maintainer-check-list-of-tests): Enhance
its recipe to make it able to deal with test script residing
in subdirectories.
* Makefile.am (maintainer-check-list-of-tests): Simplified.
(TEST_SUBDIRS): Remove, no more needed.
* tests/list-of-tests.mk (perl_TESTS): New variable, lists
the '.pl' tests just moved into 'tests/pm'.
(handwritten_TESTS): Add the contents of '$(perl_TESTS)'.
* lib/Automake/Makefile.am (SUBDIRS): Remove.
* configure.ac (AC_CONFIG_FILES): Update.
* .gitignore: Adjust.
projects / thirdparty / automake.git / log
