subdir tests: avoid an use of "make -j4", for portability
Otherwise, a spurious failure with Solaris CCS make can be triggered.
The coverage is not actually reduced, since the code path is still
covered when one runs the testsuite with AM_TESTSUITE_MAKE="make -jN"
(as should be done periodically).
subdirs: unify rules for "cleaning" and "normal" recursive targets
Before this change, the recursive invocation of cleaning targets in
the $(SUBDIRS) where done in inverse order, i.e., starting from the
last $(SUBDIRS) entry and proceeding towards the first. According
to the code comments, this was done ...
... in an attempt to alleviate a problem that can happen when
dependencies are enabled. In this case, the .P file in one
directory can depend on some automatically generated header
in an earlier directory. Since the dependencies are required
before any target is examined, make bombs.
But this comment does not apply anymore to the current implementation
of automatic dependency tracking: the '.Po' and '.Plo' files does not
depend on any C header or source file, ever!
So it seems that the distinction between "normal" and "cleaning"
recursive targets is a stale leftover of an older implementation of
the automatic dependency tracking. In fact, the Automake History
manual seems to confirm this suspect; the section "First Take on
Dependency Tracking" reads:
Because each .P file was a dependency of Makefile, this meant
that dependency tracking was done eagerly by make. For instance,
"make clean" would cause all the dependency files to be updated,
and then immediately removed. This eagerness also caused problems
with some configurations; if a certain source file could not be
compiled on a given architecture for some reason, dependency
tracking would fail, aborting the entire build.
and the following section "Dependencies As Side Effects" reads:
In this approach, the .P files were included using the -include
command, which let us create these files lazily. This avoided
the "make clean" problem.
So the distinction between "normal" and "cleaning" recursive targets
has likely been obsolete since by then already. We can thus remove
such distinction, thus reducing some complications and duplication in
our rules. Doing so, the whole testsuite still passes (both with GCC
and Sun C 5.9), even the test 'c-demo.sh', which, among the other
things, exercise the setup described in the obsolete code comment
referenced above.
Finally, note that we still keep '$(RECURSIVE_CLEAN_TARGETS)' and
'$(RECURSIVE_TARGETS)' as two distinct variables, to ensure a better
backward-compatibility for any user-defined rules that happen to use
those variables.
* NEWS: Update.
* lib/am/subdirs.am ($(RECURSIVE_CLEAN_TARGETS), $(CLEAN_TARGETS)):
Merge their recipes.
* t/subdir-distclean.sh: New test, check that "./configure && make
&& make distclean" is actually a no-op, even when conditional SUBDIRS
are involved.
* t/list-of-tests.mk: Add it.
* t/objc-megademo.sh (am_create_testdir): Define to "empty" before
including ./defs, because this test doesn't rely on the files usually
pre-set by the setup in there.
* NEWS (Warnings and deprecations): We now warn if 'configure.in' is
used instead of 'configure.ac' as autoconf input.
(Future backward-incompatibilities): Adjust.
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.
See also, in the Autoconf repository, commit 'v2.69-4-g560f16b' of
2012-05-23, "general: deprecate 'configure.in' as autoconf input".
* lib/Automake/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).
* t/help.sh: Adjust.
* t/configure.sh: Adjust and enhance.
tests: avoid spurious failures with Solaris 9 cscope program
* t/cscope.tap: Also check that the cscope program in use supports the
'-q' option; if this is not the case, some cscope usages in the generated
Makefiles would fail, so we must skip the checks involving those usages.
Some tests in the Automake testsuite already aims only at verifying
the performance, rather than the correctness, of some operations.
Still, they are somewhat shoehorned and forced into the PASS/FAIL
framework (say, with the 'ulimit' shell builtin used to verify some
operation doesn't take up too much time or memory), but that is
conceptually a stretch, and has already caused problems in practice
(see automake bug#11512 for an example).
So we start moving the "performance tests" out of the testsuite proper,
and make them run only "on demand" (when the user exports the variable
'AM_TESTSUITE_PERF' to "yes"). Ideally, we should provide those tests
with a custom runner/driver that measures and displays the relevant
performance information, but doing that correctly and with the right
APIs is definitely more difficult, so we leave it for a later step
(an hope we'll take such a step eventually).
* t/cond29.sh: Move ...
* t/perf/cond.sh: ... here, and adjust.
* t/testsuite-recheck-speed.sh: Move ...
* t/perf/testsuite-recheck.sh: ... here.
* t/testsuite-summary-speed.sh: Move ...
* t/perf/testsuite-summary.sh: ... here.
* t/list-of-tests.mk (perf_TESTS): New variable, listing the tests in
the 't/perf' directory.
(handwritten_TESTS): Adjust.
* defs: Skip any tests in the 't/perf/' subdirectory unless the
'AM_TESTSUITE_PERF' variable is set to "yes" or "y".
* .gitignore: Update.
aclocal: declare function prototypes, do not use '&' in function calls
This change will also fix automake bug#11543 (from a report by Matt
Burgess).
* aclocal.in: Declare prototypes for almost all functions early, before
any actual function definition (but omit the prototype for the dynamically
generated '&search' function). Add prototypes to any function definition.
Remove '&' from function invocations (i.e., simply use "func(ARGS..)"
instead of "&func(ARGS...)").
* THANKS, NEWS: Update.
news: support for configure.in will be dropped in future automake versions
The use of configure.in as Autoconf input has been deprecated for a
very long time in the Autoconf documentation, and the next version of
Autoconf (2.70) will start warning about it ar runtime as well (see
commit 'v2.69-4-g560f16b' or 2012-05-23, "general: deprecate
'configure.in' as autoconf input", in the Autoconf's git repository).
* syntax-checks.mk (sc_no_brace_variable_expansions): Do not complain
about ${MAKEVAR} usages in comments: they might be legitimate. This
change is not strictly required for mainline automake, but will help
in the Automake-NG branch.
* sc_tests_here_document_format: Relax a bit, so that usages like
"cout << "string" << endl;" in the test scripts (usages which can be
used in here documents defining C++ sources) are not flagged as uses
of bad delimiters for here documents. This change is not strictly
required for mainline automake, but will help in the Automake-NG
branch, and will anyway be more future proof w.r.t. possible future
testsuite additions.
(sc_tests_Exit_not_exit): Do not whitelist lines containing $PERL
explicitly (that was only required for one test); instead ...
* t/ext2.sh: ... use "exit (1)" instead of "exit 1" in the perl
invocation of the guilty script.
* Makefile.am (git-tag-release): Do not check that the version number
NEWS is updated w.r.t. $(VERSION); given the new way we manage NEWS,
that would cause gratuitous spurious failures.
* HACKING: Update.
tests: use 'parallel-tests' Automake option by default
This will help our transition from 'serial-tests' to 'parallel-tests'
as the default test suite driver enabled by a TESTS assignment in the
input Makefile.am. Note that that change of default will only take
place in master, though.
* defs: Pass the 'parallel-tests' option to the AM_INIT_AUTOMAKE
invocation in the created 'configure.ac' stub, unless the variable
'am_serial_tests' is set to "yes". Don't pay attention anymore to
the 'am_parallel_tests' variable, that's obsolete now.
* defs-static.in: Warn if the 'am_serial_tests' variable is set in the
environment; conversely, don't warn anymore about 'am_parallel_tests'
being set in the environment.
* Makefile.am (AM_TESTS_ENVIRONMENT): Nullify the 'am_serial_tests'
variable instead of the now-obsolete 'am_parallel_tests' one.
* syntax-checks.mk (sc_tests_obsolete_variables): Also warn against
uses of 'am_parallel_tests', which is now deprecated in favor of
'am_serial_tests'. Similarly, if a use of 'parallel_tests' is seen,
suggest using 'am_serial_tests' instead, not 'am_parallel_tests'.
* gen-testsuite-part: Now that we use the 'parallel-tests' by default
in our tests, we need to completely change the logic and semantics of
generation of sibling tests for those tests that check the Automake
generated testsuite harness itself. Do that, and give a complete
explanation of the new logic and semantics in the relevant comments.
* t/README: Update.
* Lots of test cases: Adjust.
doc: fix location of Automake's own test-suite.log file
* doc/automake.texi (Reporting Bugs): The global testsuite log of
Automake is now (basically since we have removed make recursion from
the Automake's build system) saved in the file 'test-suite.log', not
in the file 'tests/test-suite.log'. Adjust accordingly.
Nikolai Weibull [Wed, 16 May 2012 16:16:41 +0000 (18:16 +0200)]
ylwrap: preserve subdirectories in "#line" munging
If Automake is used in non-recursive mode and one of the inputs is a
yacc file, for example, "src/grammar.y", ylwrap will remove too many
directories from the output file when it adjusts the paths in it.
This results in #line directives referring to "grammar.y" instead of
"src/grammar.y".
This is a result of $input_rx simply taking all the directory
components of the absolute input path and removing them.
One solution is to store the path passed to ylwrap and replace
$input_rx with it. This is what we do.
Suggestion and initial patch (without tests) by Nikolai Weibull:
<http://lists.gnu.org/archive/html/automake/2012-05/msg00013.html>
Final patch by Stefano Lattarini.
* lib/ylwrap ($input_sub_rx): New.
When munging the #line directives, substitute '$input_rx' with it,
instead of stripping it altogether.
Adjust comments.
* t/yacc-line.sh, t/lex-line: Adjust and extend.
* NEWS, THANKS: Update.
This commit should cause no semantic change in the ylwrap behaviour.
It will only be needed in light of a future change. See:
<http://lists.gnu.org/archive/html/automake/2012-05/msg00013.html>
* lib/ylwrap (get_dirname, quote_for_sed): New functions, factoring
out some non-trivial code. Use them where appropriate.
yacc tests: avoid a spurious failure with Sun C++ 5.9
* t/yacc-cxx.sh (parse1.yy): When calling the 'getchar' function declared
in the #included <cstdio>, fully qualify it as std::getchar', to avoid
Sun C++ 5.9 erroring out with:
"parse1.yy", line 5: Error: The function "getc" must have a prototype.
* t/yacc-clean-cxx.sh (sub1/parsefoo.yxx): Likewise.
* t/self-check-reexec.tap: When searching for a suitable non-Bash
shells, be sure to reject any shell that is Zsh "in disguise" (as
can be found on some Debian systems, where /bin/ksh can be symlinked
to /bin/zsh4). This is required because our testsuite does not
support older versions of Zsh, and that was causing the test to fail
in the setup just described.
tests: fix spurious failures due to dpkg install-info
* t/install-info-dir.sh: Skip some checks if 'install-info' is the one
from dpkg, not the one from GNU info; the former might try to create
files in '/var/backups/', causing spurious failures like this for any
non-root user:
install-info(.../install-info-dir.dir/_inst/info/foo.info): \
creating new section `Dummy utilities'
cp: cannot create regular file `/var/backups/infodir.bak': \
Permission denied
install-info(.../install-info-dir.dir/_inst/info/foo.info): \
could not backup .../install-info-dir.dir/_inst/info/dir in \
/var/backups/infodir.bak: No such file or directory
* t/parallel-tests-fork-bomb.sh: Export LANG, LANGUAGE and LC_ALL to
'C', to ensure error messages will be in English, and thus avoiding
possible false negatives in our grepping of them.
* testsuite-saner-shell:
tests: fix a spurious failure with dash
test defs: fix indentation (cosmetic change)
tests: remove obsolete uses of $sh_errexit_works
configure: search a sturdy POSIX shell to be used in the testsuite
tests: shell running test scripts is now named AM_TEST_RUNNER_SHELL
After the last changes, configure will ensure that the shell
selected to run the test scripts can correctly propagate exit
status to the exit trap when 'set -e' is in effect.
* configure.ac (sh_errexit_works): Do not AC_SUBST it anymore.
* defs-static.in (sh_errexit_works): Do not initialize
from the AC_SUBST value anymore.
* defs (trap): Trap the EXIT signal unconditionally.
* t/self-check-explicit-skips.sh: Do not skip the test if
'$sh_errexit_works' is != "yes", this check doesn't make
sense anymore.
* t/self-check-cleanup.tap: Likewise.
* t/self-check-exit.tap: Assume the exit trap is always
installed by ./defs.
configure: search a sturdy POSIX shell to be used in the testsuite
* configure.ac: Add code (partially inspired to checks in gnulib's
'tests/init.sh') to search for a good-enough, not-buggy POSIX/XSI
shell to be used in our testsuite. Accordingly AC_SUBSTitute the
variable 'AM_TEST_RUNNER_SHELL'.
* NEWS: Update.
Since commit v1.11-2128-g8e02441 of 2012-04-06, "maint: no more make
recursion in Automake's build system" we haven't been distributing the
testsuite README file in our release tarballs. Oops.
init: warn against obsolete usage of AM_INIT_AUTOMAKE
Support for the two- and three-arguments invocation forms of the
AM_INIT_AUTOMAKE macro, as in:
AM_INIT_AUTOMAKE($PACKAGE, $VERSION)
or:
AM_INIT_AUTOMAKE($PACKAGE, $VERSION, NODEFINE)
will be removed in the next major Automake release (1.13).
Such usages have already been deprecated in the documentation
starting from commit v1.11-2015-ge99690a of 23-02-2012 "docs,
news: document planned removal of obsolete macros and features".
We now start giving runtime warnings as well (in the 'obsolete'
category).
* NEWS: Update.
* m4/init.m4 (AM_INIT_AUTOMAKE): Report the two- and three-arguments
form invocation.
* automake.in (scan_autoconf_traces): Likewise.
* doc/automake.texi: Minor adjustments. Add an @anchor to the
location where it's described how to modernize outdated invocation
of AM_INIT_AUTOMAKE, so that it can be referenced from automake
warning/error messages.
* t/aminit-moreargs-deprecation.sh: New test.
* tests/list-of-tests.mk: Add it.
* tests/ac-output-old.tap: Adjust by calling automake with the
warnings in the 'obsolete' category disabled.
* t/backcompat.test: Likewise.
* t/backcompat3.test: Likewise.
* t/backcompat5.test: Likewise.
* t/backcompat6.test: Likewise.
* t/version.test: Likewise.
* t/version2.test: Likewise.
* t/pr2.test: Modernize style of AC_INIT and AM_INIT_AUTOMAKE
invocations, and use proper m4 quoting.
* t/pr87.test: Likewise.
* t/confsub.test: Likewise.
* t/install2.test: Likewise.
The comments in 'lib/autoconf/autoconf.m4' from Autoconf 2.69 read:
We discourage the use of the non prefixed macro names: M4sugar maps
all the builtins into 'm4_'. Autoconf has been converted to these
names too. But users may still depend upon these, so reestablish
them.
Alas, in our init.m4 files, we were still using m4 builtins without
prefixing them with "m4_". Not nice for a package that is one of the
major clients of Autoconf, and an integral part of the GNU Autotools!
Fix the issue.
* m4/init.m4 (AM_INIT_AUTOMAKE): Use 'm4_define' and 'm4_defn' instead
of 'define' and 'defn'.
Nick Alcock [Fri, 4 May 2012 14:14:43 +0000 (15:14 +0100)]
docs: fix clumsy grammar in the scripts-based testsuite chapter
* doc/automake.texi (Scripts-based Testsuites): "We'll have later" is
a rare example of English in the Automake manual clearly not written
by a native English-speaker: while comprehensible, it can be better
formulated.
Copyright-paperwork-exempt: yes Signed-off-by: Nick Alcock <nick.alcock@oracle.com> Signed-off-by: Stefano Lattarini <stefano.lattarini@gmail.com>
tests: use append mode to capture parallel make output
This should fix the spurious failure reported in automake bug#11413.
This is due to the fact that the redirected output of parallel make
can racily loose lines. For example, if GNU make (3.82) is run 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 and Linux 3.3.1. Redirection in append
mode does not suffer of this problem.
See also similar commit 'Release-1-10-280-g6426999' of 2009-03-10,
"Use append mode to capture parallel test output", which tackled a
similar problem for 't/parallel-tests3.sh' and 't/lisp8.sh'.
* t/tap-more.sh: Use append mode for output from "make -j", to avoid
dropped lines.
* t/parallel-tests3.sh, t/lisp8.sh: Enhance comments.
parallel-tests: separate different logs with an empty line
After commit v1.12-21-g5eeb366 "parallel-tests: optimize global log
creation", the log files copied in the global log are not anymore
separated by a blank line, causing unclear output like this:
FAIL: test1
===========
output from test 1
FAIL: test2
===========
output from test 2
where we would want something like this:
FAIL: test1
===========
output from test 1
FAIL: test2
===========
output from test 2
Fix this regression.
* lib/am/check.am (am__create_global_log): Print an extra empty line
after having displayed the content of a '.log' file.
* t/objcxx-deps.sh, t/objcxx-minidemo.sh: In C++ files, use
"std::cout <<", not "cout <<", to avoid triggering a spurious
failure by the syntax check 'sc_tests_here_document_format'.
Cherry-picked from yesterday's commit v1.12-81-g7f3f467, that
had erroneously been applied to 'master' only rather than to
'maint'.
* syntax-checks.mk: Move ...
(sc_tests_plain_aclocal,
sc_tests_plain_autoconf,
sc_tests_plain_perl,
sc_tests_plain_autoupdate,
sc_tests_plain_automake,
sc_tests_plain_make,
sc_tests_plain_autoheader,
sc_tests_plain_autoreconf,
sc_tests_plain_autom4te): ... all these targets ...
(sc_tests_plain_egrep_fgrep): ... and part of this target ...
(sc_tests_plain_check_rules): ... into this variable, and
reimplement their recipes ...
($(sc_tests_plain_check_rules)): ... with this static pattern
rule. Enhance the coverage they offer a little since we are
at it.
(sc_m4_am_plain_egrep_fgrep): New rule, the part of the old
'sc_tests_plain_egrep_fgrep' rule that checked for use of
'fgrep' and 'egrep' in *.am and *.m4 fragments (rather than
in test scripts).
(syntax_check_rules): Update.
* t/dist-formats.tap: Fix bad use of "make" instead of "$MAKE"
revealed by the extra coverage.
* t/cond5.sh, t/auxdir-autodetect.sh: Cosmetic changes to avoid
spuriously triggering the maintainer checks due to the
enhanced coverage.
tests: avoid spurious failure on missing Obj C/C++ compiler
* t/objc-megademo.sh: Arrange the test to SKIP, not FAIL, if there is
no Objective C or no Objective C++ compiler. Since we are at it, remove
some checks that were actually testing Autoconf (not Automake) behaviour.
* objc++:
objc, objc++: test support for compilation flags
objc, objc++: add stress test
objc, objc++: test automatic dependency tracking
objc: reorganize basic tests
objc++, objc: add first semantic tests
objc++: add first basic test
news: announce initial support for Objective C++
objc++: test support for '.mm' suffix in _SOURCES entries
objc++: add documentation
objc++: initial support for Objective C++
tests: fix spurious failure due to autom4te caching
Reported by Peter Breitenlohner :
<http://lists.gnu.org/archive/html/automake-patches/2012-05/msg00045.html>
* t/suffix.sh: 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).
* t/objc-megademo.sh: New test, trying out a package using all
of C, C++, Objective C and Objective C++ at the same time.
* t/list-of-tests.mk: Add it.
* t/objc.sh, t/objc2.sh: Removed, merged into ...
* t/objc-basic.sh: ... this new test.
* t/objcxx-basic.sh: Add reference to new sister test.
* t/list-of-tests.mk: Update.
Merge branch 'parallel-tests-performance' into maint
* parallel-tests-performance:
parallel-tests: save few forks when possible
parallel-tests: optimize global log creation
parallel-tests: optimize 'recheck' target for speed
coverage: add performance tests on some parallel-tests aspects
aclocal: error out again on unrecognized arguments
Starting from commit v1.11-662-g52246cc of 2102-02-18, "cmdline parsing:
move into a dedicated perl module", aclocal has been silently ignoring
non-option arguments instead of correctly reporting them.
Fix this regression.
* t/aclocal.sh: Update to catch the regression.
* aclocal.in (parse_arguments): Explicitly reject non-option arguments.
* NEWS: Update.
bootstrap: consistently use correct copyright year
Suggested by Peter Johansson in the discussion about automake bug#11356:
<http://debbugs.gnu.org/cgi/bugreport.cgi?bug=11356#41>
* boostrap.sh (dosubst): Don't initialize the substituted value for the
release year from the current year (using `date +%Y`); instead, do it
through ...
($RELEASE_YEAR): ... this new statically-defined variable. To ensure it
won't get out-of-sync, it will be automatically updated ...
* Makefile.am (update-copyright): ... by this target's recipe.
Since are at it, fix a botched output redirection for an error message,
i.e., use ">&2" instead of ">&1".
* lib/am/check.am (am__check_pre): Save some forks, by analyzing more
carefully the value of '$@' to decide how to extract its dirname part,
and whether that should be created as a directory.
Autoconf 2.65 is almost three years old now, so requiring it is
acceptable. Also, the soon-to-appear support for Objective C++
in Automake will be simplified if we can assume that Autoconf
version (which is the one that introduced support for Objective
C++).
* NEWS (Planned future backward-incompatibilities): Automake 1.13
will require Autoconf 2.65 or later.
readme: update INSTALL to latest version from Autoconf
* INSTALL: Updated to the version coming with Autoconf 2.69. This
is warranted now that we use and require that same Autoconf version
in our build system.
color-tests: coloring can be forced on non-ANSI terminals as well
Before this change, colorization of testsuite output was suppressed
whenever the terminal was recognized to be a "dumb" one, incapable
of handling ANSI coloring (i.e., when the environment variable TERM
had a value of "dumb"). This happened even when the AM_COLOR_TESTS
variable was set to a value of "always".
Such a behaviour was suboptimal and slightly confusing; in fact, if
a user wants to force coloring of testsuite output that is being
redirected to a regular file, he should be able to do so even if his
terminal is not capable of handling ANSI colors -- in fact, such
terminal wouldn't be involved with the testsuite output in any way,
so why should it be allowed to influence it?
Thus, we now enable coloring of test output whenever AM_COLOR_TESTS
is set to "always", irrespective of the value of the TERM environment
variable.
* NEWS: Update.
* lib/am/check.am [%?COLOR%] (am__tty_colors): Activate colorization
of testsuite output whenever AM_COLOR_TESTS has the value of "always".
* t/ax/tap-summary-aux.sh: Export the TERM environment variable to
"dumb" when forcing colorization of the testsuite output; this should
*not* prevent such colorization from taking place, and we want to
check that this expectation really holds.
* t/ax/testsuite-summary-checks.sh: Likewise.
* t/color.sh: Likewise, and adjust some grepping checks.
* t/tap-color.sh: Likewise. Also, remove redundant "make check"
invocation since we are at it.
* t/color2.sh: Likewise, and check that exporting TERM=dumb actually
prevents testsuite output colorization when AM_COLOR_TESTS is unset.
* t/parallel-tests-reset-term.sh: Relax, to prevent it from failing
spuriously due to the new semantic.
cygnus: warn when it's used (flagged as 'obsolete' now)
Support for "Cygnus-style" tree as enabled by the 'cygnus'
option is now reported as obsolete when the warnings in the
'obsolete' category are enabled.
See automake bug#11034.
* automake.in (check_cygnus): Report the use of 'cygnus' mode as
obsolete.
* NEWS: Update.
* docs/automake.texi: Minor adjustments.
* t/cygnus-deprecation.sh: New test.
* t/list-of-tests.mk: Add it.
* tests/cygnus-check-without-all.sh: Adjust by calling automake
with the warnings in the 'obsolete' category disabled.
* t/cygnus-requires-maintainer-mode.sh: Likewise.
* t/cygnus-dependency-tracking.sh: Likewise.
* t/cygnus-imply-foreign.sh: Likewise.
* t/cygnus-no-installinfo.sh: Likewise.
* t/cygnus-no-dist.sh: Likewise.
* t/clean2.sh: Likewise.
* t/flavor.sh: Likewise.
* t/txinfo5.sh: Likewise.
* t/txinfo5b.sh: Likewise.
* t/library.sh: Remove use of 'cygnus' and 'dejagnu' options,
and of AM_MAINTAINER_MODE macro.
With this change, the time required to execute the test case
'testsuite-summary-speed.sh' has dropped as follows:
+ Slow Debian i686 system, 1 core x 1.5 GHz, 768 MB of RAM,
GNU make 3.81:
- 1 run: 4 minutes => 30/40 seconds
+ Fast Solaris 10 i686 system, 4 cores x 3 GHz, 20 GB of RAM,
Solaris CCS make:
- 2 runs: 6 minutes => 30 seconds
+ Very fast Fedora ppc64 system, 64 cores x 3.5 GHz, 64 GB of RAM,
GNU make 3.82:
- 5 runs: 7 minutes => 1 minute and 10 seconds
* lib/am/check.am ($(TEST_SUITE_LOG)): Optimize for speed, by avoiding
lots of forks with the help of ...
(am__create_global_log): ... this new internal variable, basically
defining a smart awk program, and ...
(am__global_test_result_rx, am__copy_in_global_log_rx): ... these new
internal variables, used by the one above.
(am__rst_section): Remove as obsolete
* t/rst-formatting.trs: Remove.
* t/list-of-tests.mk: Update.
* t/testsuite-summary-count-many.sh: Improve its stress testing by
using 1 million tests rather than just 5 thousands: we can afford
this with our new optimization, and still have the test completed
in a reasonable time.
parallel-tests: optimize 'recheck' target for speed
With this change, the time required to execute the test case
'testsuite-recheck-speed.sh' has dropped as follows:
+ Slow Debian i686 system, 1 core x 1.5 GHz, 768 MB of RAM,
GNU make 3.81:
- 1 run: 6 minutes => 40 seconds
+ Fast Solaris 10 i686 system, 4 cores x 3 GHz, 20 GB of RAM,
Solaris CCS make:
- 4 runs: 3 minutes => 1 minute
+ Very fast Fedora ppc64 system, 64 cores x 3.5 GHz, 64 GB of RAM,
GNU make 3.82:
- 5 runs: 4 minutes => 1 minute 30 seconds
* lib/am/check.am (recheck): Optimize for speed, by avoiding lots of
forks with the help of ...
(am__list_recheck_tests): ... this new internal variable, basically
defining a proper awk program, and ...
(am__recheck_rx): ... this other new internal variable, used by the
one above.
* t/test-trs-recover2.sh: Relax by not checking for a very corner
case ('.log' and '.trs' files both unreadable) that we don't handle
anymore.
coverage: add performance tests on some parallel-tests aspects
* t/testsuite-recheck-speed.sh: New test.
* t/testsuite-summary-speed.sh: Likewise.
* t/list-of-tests.mk: Update.
The new tests are not meant to PASS/FAIL, but rather to help us to
obtain quantitative measurements of the performance improvements
offered by soon-to-appear optimization patches.
Timing before the optimizations implemented by following patches:
+ Slow Debian i686 system, 1 core x 1.5 GHz, 768 MB of RAM,
GNU make 3.81:
- testsuite-recheck-speed.sh: 1 run, ~ 6 minutes
- testsuite-summary-speed.sh: 1 run, ~ 4 minutes
+ Fast Solaris 10 i686 system, 4 cores x 3 GHz, 20 GB of RAM,
Solaris CCS make:
- testsuite-recheck-speed.sh: 4 runs, ~ 3 minutes
- testsuite-summary-speed.sh: 2 runs, ~ 6 minutes
+ Very fast Fedora ppc64 system, 64 cores x 3.5 GHz, 64 GB of RAM,
GNU make 3.82:
- testsuite-recheck-speed.sh: 5 runs, ~ 4 minutes
- testsuite-summary-speed.sh: 5 runs, ~ 7 minutes
AM_PROG_MKDIR_P: deprecate, to be removed in Automake 1.13
Support for the obsolescent 'AM_PROG_MKDIR_P' macro (and its
output variable '$(mkdir_p)') has already been deprecated in
the documentation. It's now time to make its use actively
emit warnings in the 'obsolete' category as well.
* automake.in (scan_autoconf_traces): Warn about 'AM_PROG_MKDIR_P'
obsolescent if that macro is seen.
* m4/mkdirp.m4 (AM_PROG_MKDIR_P): Warn about its own obsolescence.
* m4/init.m4: Require 'AC_PROG_MKDIR_P', not 'AM_PROG_MKDIR_P'.
* t/mkdirp-deprecation.sh: New test.
* t/list-of-tests.mk: Add it.
* syntax-check.mk (sc_mkdir_p): Delete this check as obsolete.
(syntax_check_rules): Remove it.
* t/gettext-macros.sh: Ensure the gettext-requiring tests will
call aclocal and automake with the 'obsolete' warnings disabled,
since even recent versions of gettext used the now-deprecated
'AM_PROG_MKDIR_P' m4 macro.
* NEWS: Update.
fixup: avoid unconditional re-bootstrapping on "make dist"
Since our 'bootstrap' script is distributed, the 'distdir' target depends
on it. But in our GNUmakefile, we also have a 'bootstrap' target declared
.PHONY, which when called re-bootstrap the Automake distribution. Thus,
whenever we run "make dist", GNU make sees it must remake the 'bootstrap'
target (as it is a dependency of the 'distdir' target), and thus ends up
re-bootstrapping all the package (because of the .PHONY 'bootstrap' target
in GNUmakefile).
We fix this issue by renaming our bootstrap script to 'bootstrap.sh'.
build: ensure release year in copyright notice is up-to-date
From a suggestion by Eric Blake. See automake bug#11356.
This is a follow-up to previous patch 'v1.12-12-gb99b5be'.
* configure.ac (RELEASE_YEAR): New AC_SUBST'd variable, should hold
the value of the current year.
* Makefile.am (update-copyright): Be sure to also update the
definition of 'RELEASE_YEAR' in configure.ac.
* lib/Automake/Config.in ($RELEASE_YEAR): New exported variable,
initialized from the value substituted for '@RELEASE_YEAR@'.
(@EXPORT): Add it.
* automake.in, aclocal.in: Use '$RELEASE_YEAR' (which will be
substituted at make time) instead of hard-coding the release
year. This should ensure the copyright range in the version
message and in the generated files (Makefile.in and aclocal.m4)
are automatically kept up-to-date.
* bootstrap (dosubst): Update, also substitute '@RELEASE_YEAR@'.
* THANKS: Use spaces, not tabs, to separate a reporter's name from
his e-mail address. Also, increase such spacing, to allow a more
consistent formatting with longer names.
readme: explicitly state we use ranges in copyright years
Reported by Jack Kelly in automake bug#11356.
* README: Explicitly state we use ranges in copyright years; this is
mandated by the GNU Coding Standards for each package that wants to
actually use ranges in copyright years.
Dependencies in the Automake build system are not completely specified
(see for example the commit log of recent commit 'v1.12-10-gab14841',
"build: avoid too greedy rebuilds in the testsuite"). In fact, some of
them cannot even be; for example, Makefile is generated at configure
time from Makefile.in, which should be regenerated by our bleeding-edge
automake script, which is generated by out Makefile -- specifying the
complete chain of dependencies here would bring to a circular dependency
issue.
For this reason, before testing or deploying a change, we are often
forced to perform a full re-bootstrap of the Automake package, to ensure
all our files are actually up-to-date. Until now, this has to be done
manually, thus causing wasted keystrokes and more possibilities of error.
With this change, we introduce a new 'bootstrap' make target to
automatize all the (easy) steps of this re-bootstrapping (plus some
minor bells & whistles since we are at it).
* GNUmakefile: Rewrite to allow an easy bootstrapping and clean rebuild
of the whole package, in particular with the help of ...
(bootstrap): ... this new target.
automake, aclocal: update copyright year in output
Fixes automake bug#11356.
* aclocal.in, automake.in: Add copyright year '2012' in the version
message. Ditto for the copyright notice in the generated Makefile.in
or aclocal.m4 files; and for those, since we are at it, prefer using
a range for copyright years, rather than listing each copyright year
individually.
maint: don't use "union" merge driver for NEWS anymore
The use of the "union" merge driver to avoid conflicts in NEWS has
proven to be problematic, as it often silently produced mis-merges
(instead of spurious conflicts, but a spurious conflict is better
than a mis-merge). Also, our new branching policy (see automake
bug#11153 for a description) should greatly mitigate the problem of
merge conflicts in NEWS. For further discussion, see:
<http://lists.gnu.org/archive/html/automake-patches/2012-04/msg00148.html>
projects / thirdparty / automake.git / log
