Hi Stefano, a while ago ...
* Stefano Lattarini wrote on Tue, Jan 25, 2011 at 06:38:50PM CET: > <http://lists.gnu.org/archive/html/automake-patches/2010-11/msg00152.html> > <http://lists.gnu.org/archive/html/automake-patches/2010-12/msg00006.html> > OK for master? Both are OK, with nits addressed. Thanks, Ralf > Subject: [PATCH 1/2] tests: optimize `instspc-*.test' for speed > > After the split of `instspc.test' into various generated tests, > the running time of the testsuite has noticeably increased, since > all these new generated tests must run aclocal, autoconf and > automake, whereas previously they were run only once (at the > beginning of `instspc.test'). But luckily, since the new tests > share the same input files for the autotools, this situation can > be easily worked around (at the expenses of a slight increase of > complexity for the testsuite scaffolding). > > * tests/instspc-data.test: New helper test, properly calling > the `instspc-tests.sh' script to generate input data for the > others `instspc-*.test' tests. > * tests/Makefile.am (TESTS): Add `instspc-data.test'. > ($(instspc_tests:.test=.log)): Depend on its log file. > (instspc-data.log): Depend on `instspc-tests.sh'. > * tests/instspc-tests.sh: Recognize new action `generate-data', > and use it to create hand-written and autotools-generated static > files shared by all the `instspc-*.test' tests. > When sourced by the `instspc-*.test' tests, use those previously > created files instead of recreating them from scratch. > (deindent, create_input_data): New subroutines. > Some other related changes and refactorings. > --- a/tests/Makefile.am > +++ b/tests/Makefile.am > @@ -85,8 +85,11 @@ $(instspc_tests): Makefile.am > } > $@-t > $(AM_V_at)chmod a+rx $@-t && mv -f $@-t $@ > > -# All instspc*.test tests work by sourcing the instspc-tests.sh script. > -$(instspc_tests:.test=.log): instspc-tests.sh > +# All instspc-*.test tests work by sourcing the `instspc-tests.sh' > +# script. Also, they all use shared data generated by the helper > +# test `instspc-data.test', for reasons of speed. > +instspc-data.log: instspc-tests.sh > +$(instspc_tests:.test=.log): instspc-tests.sh instspc-data.log > > MAINTAINERCLEANFILES += $(instspc_tests) > EXTRA_DIST += instspc-tests.sh > @@ -504,6 +507,7 @@ instman2.test \ > instmany.test \ > instmany-mans.test \ > instmany-python.test \ > +instspc-data.test \ > $(instspc_tests) \ > interp.test \ > interp2.test \ > --- /dev/null > +++ b/tests/instspc-data.test > +# Helper testcase which generate input data for the other test > +# `instspc-*.test'. It basically delegates the work to the helper > +# script `instspc-test.sh'. As an alternative to a helper testcase, this could also just be a helper script whose run is a prerequisite to the instspc*.log files. That way you don't have a bogus test result. E.g.: $(instspc_tests:.test=.log): instspc-tests.sh instspc-data.dir/.dirstamp instspc-data.dir/.dirstamp: srcdir=$(srcdir) $(SHELL) $(srcdir)/tests/instspc-setup or so (with the script renamed to instspc-setup). And mostlyclean-local removing instspc-data.dir. You decide whether this is worthwhile. > +# Ensure proper definition of $testsrcdir. > +. ./defs-static || exit 99 > +test -n "$testsrcdir" || exit 99 # sanity check > + > +instspc_action=generate-data > +. $testsrcdir/instspc-tests.sh > --- a/tests/instspc-tests.sh > +++ b/tests/instspc-tests.sh > @@ -94,6 +99,91 @@ define_problematic_string () > esac > } > > +# Helper subroutines for creation of input data files. > + > +deindent () "unindent"? Hmm. Both sound weird, but maybe unindent is easier to read. Might wanna have it in tests/defs? > +{ > + sed 's/^ *//' # we don't strip leading tabs -- this is deliberate! > +} > + > +create_input_data () > +{ > + mkdir sub > + > + deindent > configure.in << 'EOF' > + AC_INIT([instspc], [1.0]) > + AM_INIT_AUTOMAKE > + AC_CONFIG_FILES([Makefile]) > + AC_PROG_CC > + AC_PROG_RANLIB > + AC_OUTPUT > +EOF > + [...] > + $ACLOCAL > + $AUTOCONF > + $AUTOMAKE -a > + > + : > success This could then be .dirstamp instead. > +} > + > # Be sure to avoid interferences from the environment. > instspc_names_list='' > instspc_xfail_builds_list='' > @@ -189,102 +279,54 @@ if test x"$instspc_action" = x"generate-makefile"; then > exit 0 > fi > > -### If we are still here, we have to run a test ... > - > -# We'll need the full setup provided by `tests/defs'. Temporarly disable > +# We'll need the full setup provided by `tests/defs'. Temporarily disable > # the errexit flag, since the setup code might not be prepared to deal > # with it. > set +e > . ./defs || Exit 99 > set -e > > +# The directory set up by the `generate-data' action should contain all > +# the files we need. So remove the other files created by ./defs. And > +# check we really are in a temporary `*.dir' directory in the build tree, > +# since the last thing we want is to remove some random user files! > +test -f ../defs-static && test -f ../defs || Exit 99 > +case `pwd` in *.dir);; *) Exit 99;; esac > +rm -f * > + > +if test x"$instspc_action" = x"generate-data"; then > + # We must *not* remove the test directory, since its contents must be > + # used by following dependent tests. > + trap 'Exit $?' 0 trap '' 0 is the portable way to undo the EXIT trap, if I remember correctly. You could also just set keep_testdirs, no? > + create_input_data > + Exit 0 > +fi > Subject: [PATCH 2/2] tests: `instspc-*.test': do not create useless source > file > > * tests/instspc-tests.sh (create_input_data): Do not create > unused source file `source2.c'.
