This patch series stems from a discussion on bug-automake:
[<http://comments.gmane.org/gmane.comp.sysutils.automake.bugs/4892>]
In short, some make implementations tend to use their built-in ".y => .o"
and ".l => .o" rules instead of respectively the ".y => .c => .o" and
".l => .c => .o" implicit rule chains provided by the automake-generated
Makefiles. This problem can be solved by always declaring more
dependencies *explicitly* in the generated Makefile.in.
The patch n.4 in the series does this. The other patches provide new and
enhanced tests to verify the functionalities affected by the change.
-*-*-*-
It might be worth reporting the extensive comments in the automake scripts
regarding dependencies/rules generation. Note that the following comments
are the updated, extended version provided by the propsed patch series:
# If the object file has been renamed (because per-target
# flags are used) we cannot compile the file with an
# inference rule: we need an explicit rule.
#
# If the source is in a subdirectory and the object is in
# the current directory, we also need an explicit rule.
#
# If both source and object files are in a subdirectory
# (this happens when the subdir-objects option is used),
# then the inference will work.
#
# [CUT]
#
# Using inference rules for subdir-objects has been tested
# with GNU make, Solaris make, Ultrix make, BSD make,
# HP-UX make, and OSF1 make successfully.
#
# Unfortunately, explicit *dependencies* might still be needed
# in the generated Makefile.in, even when explicit *rules* are
# *not* required. Otherwise, in particular situations, some
# make implementations (e.g. Heirloom make) might end up using
# their built-in implicit rules, rather then the rules' chains
# inferred from Automake-generated generic suffix rules.
#
# An example will help to clarify this.
#
# Let's say we have a C file `foo.c' generated from a Yacc
# input `foo.y'. If the Makefile contains no target which
# depends *explicitly* from foo.c, Heirloom make will end up
# using its builtin `.y => .o' rule (calling $(YACC) and $(CC)
# directly, and obviously ignoring the $(AM_CFLAGS) variable,
# the ylwarp script, etc.), instead of the `.y => .c => .o'
# rule chain generated by Automake (thanks to the two suffix
# rules `.y.c' and `.c.o').
#
# So, for consistency and simplicity, we always declare
# *explicit* dependencies, while avoiding to output
# unnecessary explicit *rules*.
-*-*-*-
Most relevant bits of the original discussion on bug-automake are reported
here, just for the record ...
Stefano Lattarini:
> I encounterd the following failure in silent5.test when trying out the
> Automake testsuite using Heirloom make as $MAKE:
>> [...]
>> rm -f baz5.c
>> flex -t baz5.l > baz5.c
>> gcc -g -O2 -c -o baz5.o baz5.c
>> rm -f baz5.c
>> CC baz6.o
>> [...]
>> LEX baz5.c
>> CC ba2-baz5.o
>> [...]
>> rm -f foo5.c
>> flex -t foo5.l > foo5.c
>> gcc -g -O2 -c -o foo5.o foo5.c
>> rm -f foo5.c
>> [...]
> [CUT]
> The failure is not spurious, and seems due to the fact that heirloom
> make is overly verbose in rules involving `flex' even when silent-rules
> are enabled.
Ralf Wildenhues:
> That looks like heirloom-make uses its own internal rule for baz5.c and
> for foo5.c. That would be problematic because ylwrap is not used, and
> other problems. With a non-flex lex you'd see more test failures I
> suppose.
> [CUT]
> I'm guessing this has to do with chains of inference rules not being
> detected or so.
Ralf Wildenhues:
> Yeah, the make has a .l.o rule that triggers before our .l.c and .c.o
> rule chains.
> I guess this could be worked around by adding explicit rules (at least
> that's what SUSv3 recommends), maybe explicit dependencies without
> rules suffice.
Stefano Lattarini:
> Yes, explicit dependencies without rules suffice.
> If I venture to write a patch to solve this problem, will you
> consider it?
Ralf Wildenhues:
> Yes, if it's not dangerous or regresses other situations.
-*-*-*-
Patch series' summary:
Stefano Lattarini (6):
Extend, fix and improve tests on Yacc/Lex support.
New tests: Heirloom make and Lex/Yacc rules.
Add new tests on Lex and Yacc.
automake.in: improve declaration of dependencies in generated Makefile.in
New tests on Lex and Yacc dependencies declarations.
More tests on AM_PROG_LEX, AC_PROG_LEX and @lex...@.
ChangeLog | 77 +++++++++++
automake.in | 114 ++++++++++------
tests/Makefile.am | 22 +++-
tests/Makefile.in | 22 +++-
tests/defs.in | 7 +
tests/exeext3.test | 10 +-
tests/lex10.test | 26 ++++
tests/lex11.test | 26 ++++
tests/lex12.test | 49 +++++++
tests/lex13.test | 70 ++++++++++
tests/lex2.test | 2 +
tests/lex3.test | 28 ++--
tests/lex4.test | 2 +
tests/lex5.test | 21 +--
tests/lex6.test | 54 ++++++++
tests/lex7.test | 152 +++++++++++++++++++++
tests/lex8.test | 134 ++++++++++++++++++
tests/lex9.test | 102 ++++++++++++++
tests/lexcpp.test | 46 ++++++
tests/lexlib.test | 55 ++++++++
tests/lexvpath.test | 117 ++++++++++++++++
tests/suffix3.test | 26 +++--
tests/yacc.test | 7 +-
tests/yacc10.test | 164 ++++++++++++++++++++++
tests/yacc11.test | 145 ++++++++++++++++++++
tests/yacc12.test | 26 ++++
tests/yacc13.test | 26 ++++
tests/yacc2.test | 15 +-
tests/yacc3.test | 15 +-
tests/yacc4.test | 18 +--
tests/yacc5.test | 24 ++--
tests/yacc6.test | 14 +-
tests/yacc7.test | 8 +-
tests/yacc8.test | 21 ++--
tests/yacc9.test | 56 ++++++++
tests/yaccpp.test | 21 ++-
tests/yaccvpath.test | 27 ++--
tests/yl-static-checks.sh | 333 +++++++++++++++++++++++++++++++++++++++++++++
38 files changed, 1920 insertions(+), 162 deletions(-)
create mode 100755 tests/lex10.test
create mode 100755 tests/lex11.test
create mode 100755 tests/lex12.test
create mode 100755 tests/lex13.test
create mode 100755 tests/lex6.test
create mode 100755 tests/lex7.test
create mode 100755 tests/lex8.test
create mode 100755 tests/lex9.test
create mode 100755 tests/lexcpp.test
create mode 100755 tests/lexlib.test
create mode 100755 tests/lexvpath.test
create mode 100755 tests/yacc10.test
create mode 100755 tests/yacc11.test
create mode 100755 tests/yacc12.test
create mode 100755 tests/yacc13.test
create mode 100755 tests/yacc9.test
create mode 100644 tests/yl-static-checks.sh
-*-*-*-
Patches will follow soon.
Regards,
Stefano
