Akim Demaille [Fri, 17 Feb 2012 13:59:23 +0000 (14:59 +0100)]
examples: factor the test suite.
* examples/mfcalc/test, examples/calc++/test: Extract the
common bits into...
* examples/test: here.
(cwd): New.
Use it to avoid a race on the temporary directory.
Reported by Jim Meyering.
* examples/mfcalc/test, examples/calc++/test: Rename into...
* examples/mfcalc/mfcalc.test, examples/calc++/calc++.test: these.
* examples/calc++/local.mk, examples/mfcalc/local.mk,
* examples/local.mk: Adjust.
Akim Demaille [Fri, 17 Feb 2012 09:12:06 +0000 (10:12 +0100)]
examples: fix the test suites.
* examples/calc++/test, examples/mfcalc/test (me): Be more
meaningfull: include the example name.
(prog): Factor.
(run): Avoid printf, use echo.
Add missing parens.
(cleanup): New.
Call it on trap.
Remove the previous "rm" that did the cleanup.
Move into a private directory to avoid concurrency issues.
Reported by Jim Meyering.
Akim Demaille [Thu, 16 Feb 2012 14:27:03 +0000 (15:27 +0100)]
calc++: factor for other extracted tests.
* Makefile.am (TESTS, check_PROGRAMS): Initialize here.
* examples/local.mk (doc, extexi): Define here.
* examples/calc++/local.mk: Adjust accordingly.
* configure.ac: Ask for parallel-tests (for the way the logs
are handled).
* examples/calc++/test: As a consequence, always be verbose.
($prog): New.
(run): Use it.
Sort the tests in a more natural order (simplest first).
Akim Demaille [Thu, 16 Feb 2012 14:07:23 +0000 (15:07 +0100)]
doc: fix mfcalc code.
* doc/bison.texinfo (Multi-function Calc): Add missing includes.
Fix the rendering of the result: use @result and remove the
initial tabulation in the actual code.
Fix stylistic issues: avoid the , operator.
Add extexi mark-up.
* examples/extexi: Also support @smallexample.
Jim Meyering [Thu, 16 Feb 2012 10:03:31 +0000 (11:03 +0100)]
tests: avoid c++ failure due to lack of getenv decl
* tests/c++.at (Syntax error as exception): Avoid spurious failure
at least when compiling with g++-4.7.x due to lack of declaration
of getenv. Include <stdlib.h>.
Akim Demaille [Wed, 15 Feb 2012 12:36:24 +0000 (13:36 +0100)]
maint: rely on Automake for parsers.
* Makefile.am (AM_YFLAGS): Automake looks for "-d" alone.
Move other options in here.
(BISON): New.
(YACC): Use it.
(bison_SOURCES): Now that automake can see `-d' in AM_YFLAGS,
we can rely on it to compile and ship the parser header
files.
Akim Demaille [Tue, 14 Feb 2012 16:32:51 +0000 (17:32 +0100)]
calc++: rely on Automake.
Rely on $(YACC) being the bison being built, and let Automake do the
rest. It turned out to be much more difficult than anticipated, for
various reasons, including some bad behavior from Automake 1.11.2
which (i) generates calc++-parser.h instead of calc++-parser.hh, and
(ii) leaves an #include "y.tab.h" in the generated parser instead
of #include "calc++-parser.h".
The authors of Automake appear to be aware of the problem,
http://lists.gnu.org/archive/html/automake/2011-05/msg00008.html
so a simple work around will suffice for the time being.
* examples/calc++/y.tab.h, examples/calc++/calc++-parser.hh: New.
To work around Automake 1.11.2 issues.
* examples/calc++/local.mk: Remove all the rules for compilation
with bison, leave them to Automake.
So provide it with "calc++-parse.yy" as a source file.
(calc_sources_generated, calc_sources_extracted): Rename as.
(calc_generated, calc_extracted): these.
(calc_sources): New.
Fix them.
Akim Demaille [Tue, 14 Feb 2012 16:27:15 +0000 (17:27 +0100)]
maint: tidy the Makefile a bit.
* src/local.mk: Put yacc related variables together.
(AUTOMAKE_OPTIONS): Move to...
* Makefile.am: here.
Remove an old Emacs mode request which disables Automake support.
* src/local.mk (YACC, AM_YFLAGS): Move to...
* Makefile.am: here, as they will be used by other local.mks.
Akim Demaille [Mon, 13 Feb 2012 17:23:45 +0000 (18:23 +0100)]
maint: de-recurse the handling of examples
The directory was still using a local Makefile.am because it provides
"scoped" Make variables: these examples are not meant to use the same
CPPFLAGS etc. If we were to use the same -I set, we'd pick up
gnulib's stdio.h for instance, which we do not want for these simple
examples.
Yet, as a result, the dependencies are less accurate, there is code
duplication, etc. This is especially perceptible when trying to
extract more examples from the documentation, as will be done in
forthcoming changes.
In order to make the tuning of CPPFLAGS easier, discard the predefined
-I from Automake.
* examples/calc++/Makefile.am: Rename as...
* examples/calc++/local.mk: this.
Adjust the paths which are now rooted in top_srcdir/top_builddir.
Handle BISON_CXX_WORKS here, instead of the too crude previous
approach that completely discarded the whole directory.
($(BISON)): Remove now useless bouncing recipe.
(calc___CPPFLAGS): New.
Stay away from -Ilib.
* Makefile.am, configure.ac, examples/local.mk,
* examples/calc++/test: Adjust.
Akim Demaille [Thu, 30 Jun 2011 12:10:21 +0000 (14:10 +0200)]
lalr1.cc: also handle syntax_error when calling yylex.
* data/lalr1.cc (parse): Catch syntax_error around yylex and
forward them to errlab1.
* tests/c++.at (Syntax error as exception): Check support for
syntax exceptions raised by the scanner.
* NEWS, doc/bison.texinfo: Document it.
Akim Demaille [Thu, 9 Feb 2012 13:09:23 +0000 (14:09 +0100)]
maint: more silent rules.
* tests/local.mk (TESTSUITE_AT): Include plackage.m4.
Adjust dependencies.
Make testsuite.at its first argument.
(package.m4): Be silent.
(testsuite): Be silent.
Use $<.
Akim Demaille [Thu, 9 Feb 2012 13:02:22 +0000 (14:02 +0100)]
skeletons: simplify the protections against "unused" warnings.
* data/c.m4 (b4_parse_param_use): Also accept optional arguments
to "use".
Simplify callers.
* data/glr.c (yyuserAction): Simplify use of b4_parse_param_use.
(yy_reduce_print): Don't use b4_parse_param_use, as all the arguments
_are_ used.
* data/lalr1.cc (YY_SYMBOL_PRINT): Even when disabled, "use" the
symbol argument.
This neutralizes a warning in yypush_ when there are no symbols
with a semantic values.
(yy_destroy_): Remove useless "use" of yymsg.
Akim Demaille [Wed, 8 Feb 2012 09:28:58 +0000 (10:28 +0100)]
use a more consistent quoting style.
See <http://lists.gnu.org/archive/html/bug-bison/2012-01/msg00120.html>.
Use quotearg as often as possible instead of leaving the choice of
the quotes to the translators. Use shorter messages. Factor similar
messages to a single format, to make localization easier.
* src/files.c, src/getargs.c, src/muscle-tab.c, src/reader.c
* src/scan-code.l, src/scan-gram.l, src/symtab.c:
Use quote() or quotearg_colon() on printf arguments instead of
quotes in the format string.
* data/bison.m4: Keep sync with the changes in muscle-tab.c.
Akim Demaille [Wed, 8 Feb 2012 09:28:58 +0000 (10:28 +0100)]
use a more consistent quoting style.
See <http://lists.gnu.org/archive/html/bug-bison/2012-01/msg00120.html>.
Use quotearg as often as possible instead of leaving the choice of
the quotes to the translators. Use shorter messages. Factor similar
messages to a single format, to make localization easier.
* src/files.c, src/getargs.c, src/muscle-tab.c, src/reader.c
* src/scan-code.l, src/scan-gram.l, src/symtab.c:
Use quote() or quotearg_colon() on printf arguments instead of
quotes in the format string.
* data/bison.m4: Keep sync with the changes in muscle-tab.c.
Akim Demaille [Mon, 30 Jan 2012 15:21:13 +0000 (16:21 +0100)]
maint: more silent-rules.
* doc/local.mk, src/local.mk, examples/calc++/Makefile.am: Use
$(AM_V_GEN) and $(AM_V_at) where appropriate.
(cherry picked from commit f67c0a1c35cbc342e1c1b2201aa563b17c26e3b1)
Jim Meyering [Sun, 29 Jan 2012 11:50:32 +0000 (12:50 +0100)]
do not ignore errors like ENOSPC,EIO when writing to stdout
Standard output was never explicitly closed, so we could not
detect failure. Thus, bison would ignore the errors of writing
to a full file system and getting an I/O error on write, but only
for standard output, e.g., for --print-localedir, --print-datadir,
--help and some verbose output.
Now, "bison --print-datadir > /dev/full" reports the write failure:
bison: write error: No space left on device
Before, it would exit 0 with no diagnostic, implying success.
This is not an issue for "--output=-" or the other FILE-accepting
command-line options, because unlike most other GNU programs,
an output file argument of "-" is treated as the literal "./-",
rather than standard output.
* bootstrap.conf (gnulib_modules): Add closeout.
* src/main.c: Include "closeout.h".
Use atexit to ensure we close stdout.
* .gitignore: Ignore new files pulled in via gnulib-tool.
(cherry picked from commit acb5895680611f8beb497b41694d9686f2932c50)
Jim Meyering [Sun, 29 Jan 2012 11:50:32 +0000 (12:50 +0100)]
do not ignore errors like ENOSPC,EIO when writing to stdout
Standard output was never explicitly closed, so we could not
detect failure. Thus, bison would ignore the errors of writing
to a full file system and getting an I/O error on write, but only
for standard output, e.g., for --print-localedir, --print-datadir,
--help and some verbose output.
Now, "bison --print-datadir > /dev/full" reports the write failure:
bison: write error: No space left on device
Before, it would exit 0 with no diagnostic, implying success.
This is not an issue for "--output=-" or the other FILE-accepting
command-line options, because unlike most other GNU programs,
an output file argument of "-" is treated as the literal "./-",
rather than standard output.
* bootstrap.conf (gnulib_modules): Add closeout.
* src/main.c: Include "closeout.h".
Use atexit to ensure we close stdout.
* .gitignore: Ignore new files pulled in via gnulib-tool.
Akim Demaille [Wed, 25 Jan 2012 15:57:58 +0000 (16:57 +0100)]
yacc: fix YYBACKUP.
Reported by David Kastrup:
https://lists.gnu.org/archive/html/bug-bison/2011-10/msg00002.html.
* data/yacc.c (YYBACKUP): Accept rhs size.
Restore the proper state value.
* TODO (YYBACKUP): Make it...
* tests/actions.at: a new test case.
* NEWS, THANKS: Update.
(cherry picked from commit d115aad9112fb4e2fe1b268c9db7390732d39539)
Akim Demaille [Wed, 25 Jan 2012 12:04:43 +0000 (13:04 +0100)]
maint: fix --gcc-warnings support.
* configure.ac: Use enable_gcc_warnings instead of enableval,
which is valid only with AC_ARG_ENABLE.
(cherry picked from commit 1890a2a816dab86c23cc1d0af8fac3986335deb7)
Akim Demaille [Wed, 25 Jan 2012 15:57:58 +0000 (16:57 +0100)]
yacc: fix YYBACKUP.
Reported by David Kastrup:
https://lists.gnu.org/archive/html/bug-bison/2011-10/msg00002.html.
* data/yacc.c (YYBACKUP): Accept rhs size.
Restore the proper state value.
* TODO (YYBACKUP): Make it...
* tests/actions.at: a new test case.
* NEWS, THANKS: Update.
Paul Eggert [Tue, 24 Jan 2012 21:54:12 +0000 (13:54 -0800)]
tests: port to Solaris 10 'diff -u'
* tests/regression.at (parse-gram.y: LALR = IELR): Port to Solaris 10,
where "diff -u X X" outputs "No differences encountered"
instead of outputting nothing. Reported by Tomohiro Suzuki in
<http://lists.gnu.org/archive/html/bug-bison/2012-01/msg00101.html>.
(cherry picked from commit 071ca141ec7987a5eca782227ce70db434a206ab)
Jim Meyering [Mon, 23 Jan 2012 10:47:46 +0000 (11:47 +0100)]
build: avoid possibly-replaced fprintf in liby-source, yyerror.c
* lib/yyerror.c (yyerror): Use fputs and fputc rather than fprintf
with a mere "%s\n" format. Always return 0 now, on the assumption
that the return value was never used anyway.
Don't include <config.h> after all. This avoids a problem
reported by Thiru Ramakrishnan in
http://lists.gnu.org/archive/html/help-bison/2011-11/msg00000.html
* cfg.mk: Exempt lib/yyerror.c from the sc_require_config_h_first test.
* THANKS: Update.
(cherry picked from commit fb9eea88856d73a5f25295a52f6d4df7fabc565b)
Paul Eggert [Tue, 24 Jan 2012 21:54:12 +0000 (13:54 -0800)]
tests: port to Solaris 10 'diff -u'
* tests/regression.at (parse-gram.y: LALR = IELR): Port to Solaris 10,
where "diff -u X X" outputs "No differences encountered"
instead of outputting nothing. Reported by Tomohiro Suzuki in
<http://lists.gnu.org/archive/html/bug-bison/2012-01/msg00101.html>.
Jim Meyering [Mon, 16 Jan 2012 18:20:50 +0000 (19:20 +0100)]
quote consistently and make tests pass with new quoting from gnulib
Updating to gnulib pulled in new quote and quotarg modules,
by which quoting is now done like 'this' rather than `this'.
That change induces many "make check" test failures. This change
adapts code and tests so that "make check" passes once again.
* src/scan-code.l: Quote like 'this', not like `this'.
* src/scan-gram.l: Likewise.
* src/symtab.c: Likewise.
* tests/actions.at: Adjust tests to match.
* tests/input.at: Likewise.
* tests/named-refs.at: Likewise.
* tests/output.at: Likewise.
* tests/regression.at: Likewise.
* lib/.gitignore: Regenerate.
* m4/.gitignore: Likewise.
(cherry picked from commit ae93e4e4b8e78396a7707de3dcdd47aca7a46dec)
Jim Meyering [Mon, 23 Jan 2012 10:47:46 +0000 (11:47 +0100)]
build: avoid possibly-replaced fprintf in liby-source, yyerror.c
* lib/yyerror.c (yyerror): Use fputs and fputc rather than fprintf
with a mere "%s\n" format. Always return 0 now, on the assumption
that the return value was never used anyway.
Don't include <config.h> after all. This avoids a problem
reported by Thiru Ramakrishnan in
http://lists.gnu.org/archive/html/help-bison/2011-11/msg00000.html
* cfg.mk: Exempt lib/yyerror.c from the sc_require_config_h_first test.
* THANKS: Update.
Jim Meyering [Mon, 16 Jan 2012 11:29:19 +0000 (12:29 +0100)]
build: manually update bootstrap from gnulib, and adapt
Updating to the latest bootstrap from gnulib involves more of a
change than usual, and updating to the latest gnulib would involve
its own set of challenges with the upcoming quoting changes, so
we update bootstrap manually and separately.
* bootstrap: Update from gnulib.
* lib/Makefile.am: Initialize more variables to empty, so that gnulib.mk
can append to them with "+=".
* bootstrap.conf (gnulib_mk_hook): Remove. No longer honored.
(gnulib_tool_option_extras): Generate gnulib.mk.
(cherry picked from commit 85ad742d4e94345d5ac07755f6722344cfd49603)
Jim Meyering [Mon, 16 Jan 2012 11:18:46 +0000 (12:18 +0100)]
maint: include <config.h> first
* cfg.mk (exclude_file_name_regexp--sc_require_config_h_first):
Exempt data/glr.c and data/yacc.c from the include-config.h-first
requirement.
(cherry picked from commit 77f114dd6e50ab39e0962f39ad9c2f2e1d347acf)
Jim Meyering [Mon, 16 Jan 2012 10:58:41 +0000 (11:58 +0100)]
maint: placate the space-TAB syntax-check
* cfg.mk (exclude_file_name_regexp--sc_space_tab): Exempt
tests/input.at and tests/c++.at, since they appear to use
SP-TAB sequences deliberately.
* OChangeLog: Remove space-before-TAB.
(cherry picked from commit aebc4303615a4431acaa54e42380feb6b63e7f12)
Jim Meyering [Sat, 21 Jan 2012 21:43:09 +0000 (22:43 +0100)]
build: generalize etc/prefix-gnulib-mk
This script hard-coded "libbison" and lib/gnulib.mk.
Adjust the script to require a --lib-name=$gnulib_name option
and a FILE argument like lib/$gnulib_mk.
Also add support for --help and --version.
* etc/prefix-gnulib-mk: Generalize.
* bootstrap.conf (bootstrap_post_import_hook): Update its invocation.
Jim Meyering [Mon, 16 Jan 2012 10:47:05 +0000 (11:47 +0100)]
maint: get gpl-3.0 from gnulib
* bootstrap.conf (gnulib_modules): Add gpl-3.0.
* doc/gpl-3.0.texi: Remove from version control, now that
we get it via gnulib.
* doc/.gitignore: Ignore it.
(cherry picked from commit b87414a05e43ed270d5a88d365d0139f43acb34d)
Jim Meyering [Mon, 16 Jan 2012 18:20:50 +0000 (19:20 +0100)]
quote consistently and make tests pass with new quoting from gnulib
Updating to gnulib pulled in new quote and quotarg modules,
by which quoting is now done like 'this' rather than `this'.
That change induces many "make check" test failures. This change
adapts code and tests so that "make check" passes once again.
* src/scan-code.l: Quote like 'this', not like `this'.
* src/scan-gram.l: Likewise.
* src/symtab.c: Likewise.
* tests/actions.at: Adjust tests to match.
* tests/input.at: Likewise.
* tests/named-refs.at: Likewise.
* tests/output.at: Likewise.
* tests/regression.at: Likewise.
* lib/.gitignore: Regenerate.
* m4/.gitignore: Likewise.
Jim Meyering [Mon, 16 Jan 2012 11:29:19 +0000 (12:29 +0100)]
build: manually update bootstrap from gnulib, and adapt
Updating to the latest bootstrap from gnulib involves more of a
change than usual, and updating to the latest gnulib would involve
its own set of challenges with the upcoming quoting changes, so
we update bootstrap manually and separately.
* bootstrap: Update from gnulib.
* Makefile.am: Initialize more variables to empty, so that gnulib.mk
can append to them with "+=".
* bootstrap.conf (gnulib_mk_hook): Remove. No longer honored.
(bootstrap_post_import_hook): Instead, run the same command,
etc/prefix-gnulib-mk lib/$gnulib_mk, via slightly different API.
Temporarily disable "bootstrap_sync=true".
* etc/prefix-gnulib-mk: Don't prepend "lib/" to tokens like -I$(...
or "\".
Jim Meyering [Mon, 16 Jan 2012 10:58:41 +0000 (11:58 +0100)]
maint: placate the space-TAB syntax-check
* cfg.mk (exclude_file_name_regexp--sc_space_tab): Exempt
tests/input.at and tests/c++.at, since they appear to use
SP-TAB sequences deliberately.
* OChangeLog: Remove space-before-TAB.
Jim Meyering [Mon, 16 Jan 2012 10:47:05 +0000 (11:47 +0100)]
maint: get gpl-3.0 from gnulib
* bootstrap.conf (gnulib_modules): Add gpl-3.0.
* doc/gpl-3.0.texi: Remove from version control, now that
we get it via gnulib.
* doc/.gitignore: Ignore it.