| Summary: | [4.0 Regression] libjava bootstrap failure on Tru64 UNIX: CPPFLAGS changed in libltdl | ||
|---|---|---|---|
| Product: | gcc | Reporter: | Rainer Orth <ro> |
| Component: | libgcj | Assignee: | Not yet assigned to anyone <unassigned> |
| Status: | RESOLVED FIXED | ||
| Severity: | critical | CC: | gcc-bugs, gerald, kcook, sean, tromey |
| Priority: | P2 | Keywords: | build |
| Version: | 4.0.0 | ||
| Target Milestone: | 4.0.0 | ||
| Host: | Target: | ||
| Build: | Known to work: | ||
| Known to fail: | Last reconfirmed: | 2004-11-15 12:17:43 | |
| Bug Depends on: | |||
| Bug Blocks: | 17574 | ||
| Attachments: | Base target CPPFLAGS on CPPFLAGS_FOR_TARGET | ||
|
Description
Rainer Orth
2004-10-29 17:51:42 UTC
I'm seeing this in general, on *all* platforms, if I set CFLAGS to a string ending with a blank: export CFLAGS="-pipe ", for example. I think this is autoconf/automake/libtool bug (it is a quoting problem I assume). I tried reproducing this on x86 FC2.
I did:
rm i686-pc-linux-gnu/libjava/{config.cache,Makefile}
make CFLAGS_FOR_TARGET='-g -O2 ' configure-target-libjava
Then:
cd i686-pc-linux-gnu/libjava/libltdl/
./config.status --recheck
This worked fine. And before running this last step I looked
in config.cache and libltdl/config.status to see how CFLAGS was
set. It all looked ok to me.
Perhaps it is a ksh quoting bug. Or perhaps my approach to reproducing
it is incorrect somehow.
It would be worth digging through your config.status and config.cache
files to see where the extra space occurs and where it does not. This
will help in isolating the bug, e.g. if the space is in config.status
then the bug probably occurs during the actual invocation; if the space
is in config.cache but not config.status then it may be stripped when
creating config.status, etc.
Subject: Re: [4.0 Regression] libjava bootstrap failure on Tru64 UNIX: CPPFLAGS changed in libltdl tromey at gcc dot gnu dot org writes: > Perhaps it is a ksh quoting bug. Or perhaps my approach to reproducing > it is incorrect somehow. To make sure the problem isn't shell dependent, I ran a bootstrap on alpha-dec-osf5.1b with CONFIG_SHELL set to bash: the same problem occured as with /bin/ksh. > It would be worth digging through your config.status and config.cache > files to see where the extra space occurs and where it does not. This > will help in isolating the bug, e.g. if the space is in config.status > then the bug probably occurs during the actual invocation; if the space > is in config.cache but not config.status then it may be stripped when > creating config.status, etc. In the alpha-dec-osf5.1b build mentioned above, I find in alpha-dec-osf5.1b/libjava/config.cache: ac_cv_env_CPPFLAGS_value='-O2 -g -O2 -mieee' i.e. the extra space occurs. In config.status, there is with options \"'--cache-file=./config.cache' '--host=alpha-dec-osf5.1b' '--build=alpha-dec-osf5.1b' '--enable-multilib' '--prefix=/vol/gcc' '--with-local-prefix=/vol/gcc' '--disable-nls' '--disable-libmudflap' '--with-gcc-version-trigger=/vol/gnu/src/gcc/gcc-dist/gcc/version.c' '--enable-languages=c,c++,java,objc' '--program-transform-name=s,y,y,' '--srcdir=/vol/gnu/src/gcc/gcc-dist/libjava' '--with-target-subdir=alpha-dec-osf5.1b' 'CPPFLAGS=-O2 -g -O2 -mieee' 'build_alias=alpha-dec-osf5.1b' 'host_alias=alpha-dec-osf5.1b' 'target_alias=alpha-dec-osf5.1b' --enable-ltdl-convenience --with-auxdir=/vol/gnu/src/gcc/gcc-dist\" echo "running /usr/local/bin/bash /vol/gnu/src/gcc/gcc-dist/libjava/configure " '--cache-file=./config.cache' '--host=alpha-dec-osf5.1b' '--build=alpha-dec-osf5.1b' '--enable-multilib' '--prefix=/vol/gcc' '--with-local-prefix=/vol/gcc' '--disable-nls' '--disable-libmudflap' '--with-gcc-version-trigger=/vol/gnu/src/gcc/gcc-dist/gcc/version.c' '--enable-languages=c,c++,java,objc' '--program-transform-name=s,y,y,' '--srcdir=/vol/gnu/src/gcc/gcc-dist/libjava' '--with-target-subdir=alpha-dec-osf5.1b' 'CPPFLAGS=-O2 -g -O2 -mieee' 'build_alias=alpha-dec-osf5.1b' 'host_alias=alpha-dec-osf5.1b' 'target_alias=alpha-dec-osf5.1b' --enable-ltdl-convenience --with-auxdir=/vol/gnu/src/gcc/gcc-dist $ac_configure_extra_args " --no-create --no-recursion" >&6 exec /usr/local/bin/bash /vol/gnu/src/gcc/gcc-dist/libjava/configure '--cache-file=./config.cache' '--host=alpha-dec-osf5.1b' '--build=alpha-dec-osf5.1b' '--enable-multilib' '--prefix=/vol/gcc' '--with-local-prefix=/vol/gcc' '--disable-nls' '--disable-libmudflap' '--with-gcc-version-trigger=/vol/gnu/src/gcc/gcc-dist/gcc/version.c' '--enable-languages=c,c++,java,objc' '--program-transform-name=s,y,y,' '--srcdir=/vol/gnu/src/gcc/gcc-dist/libjava' '--with-target-subdir=alpha-dec-osf5.1b' 'CPPFLAGS=-O2 -g -O2 -mieee' 'build_alias=alpha-dec-osf5.1b' 'host_alias=alpha-dec-osf5.1b' 'target_alias=alpha-dec-osf5.1b' --enable-ltdl-convenience --with-auxdir=/vol/gnu/src/gcc/gcc-dist $ac_configure_extra_args --no-create --no-recursion ac_configure_args="--enable-multilib '--cache-file=./config.cache' '--host=alpha-dec-osf5.1b' '--build=alpha-dec-osf5.1b' '--enable-multilib' '--prefix=/vol/gcc' '--with-local-prefix=/vol/gcc' '--disable-nls' '--disable-libmudflap' '--with-gcc-version-trigger=/vol/gnu/src/gcc/gcc-dist/gcc/version.c' '--enable-languages=c,c++,java,objc' '--program-transform-name=s,y,y,' '--srcdir=/vol/gnu/src/gcc/gcc-dist/libjava' '--with-target-subdir=alpha-dec-osf5.1b' 'CPPFLAGS=-O2 -g -O2 -mieee' 'build_alias=alpha-dec-osf5.1b' 'host_alias=alpha-dec-osf5.1b' 'target_alias=alpha-dec-osf5.1b' --enable-ltdl-convenience --with-auxdir=/vol/gnu/src/gcc/gcc-dist" s,@CFLAGS@,-O2 -g -O2 -mieee,;t t s,@CXXFLAGS@,-g -O2 -mieee,;t t s,@LIBGCJ_CFLAGS@, -mieee,;t t s,@LIBGCJ_CXXFLAGS@, -mieee,;t t s,@LIBGCJ_JAVAFLAGS@, -mieee,;t t s,@CPPFLAGS@,-O2 -g -O2 -mieee,;t t s,@IEEESPEC@,-mieee,;t t I.e. the extra space is present here, too. In libltdl/config.log, I find $ /vol/gnu/src/gcc/gcc-dist/libjava/libltdl/configure --prefix=/vol/gcc --cache-file=./config.cache --host=alpha-dec-osf5.1b --build=alpha-dec-osf5.1b --enable-multilib --prefix=/vol/gcc --with-local-prefix=/vol/gcc --disable-nls --disable-libmudflap --with-gcc-version-trigger=/vol/gnu/src/gcc/gcc-dist/gcc/version.c--enable-languages=c,c++,java,objc --program-transform-name=s,y,y, --srcdir=/vol/gnu/src/gcc/gcc-dist/libjava --with-target-subdir=alpha-dec-osf5.1b CPPFLAGS=-O2 -g -O2 -mieee build_alias=alpha-dec-osf5.1b host_alias=alpha-dec-osf5.1b target_alias=alpha-dec-osf5.1b --enable-ltdl-convenience --with-auxdir=/vol/gnu/src/gcc/gcc-dist --cache-file=.././config.cache --srcdir=/vol/gnu/src/gcc/gcc-dist/libjava/libltdl Here, the extra space is gone. Rainer I'm up to three bugs here of which two are very related. The first bug is that somewhere during bootstrap an extra space is getting tacked on CFLAGS. I'll try to track down where that happens. I'm pretty sure that some mistake is happening in the toplevel Makefile.* wizardy. Obviously this is not right, but on the other hand it really shouldn't matter much. Usually, this extra space doesn't matter due to another bug shown by Gerald's testcase which shows that at some invocation of a submake an intentional trailing space is gets stripped from CFLAGS. I do think I know why this bogus space causes bootstrap failures on Alpha and that is the config/mt-alphaieee which appends " -mieee" to CFLAGS, this means our wayward trailing space in CFLAGS is now longer a trailing space and so it doesn't get stripped when it is calls a submake at some point. So what does all this have to do with CPPFLAGS? Well, this bug reports exposes that we do have a more troubling problem in that CPPFLAGS is getting assigned "-O2 -g -O2". CPPFLAGS is for the preproccesor and only should have things like "-I ../dir" or "-DMACRO". This came from this patch here by Sean McNeil: http://gcc.gnu.org/ml/gcc-patches/2003-02/msg01736.html I think DJ rightfully questioned this patch at the time. It's clearly not correct to abuse CPPFLAGS that way IMO. I propose that patch be reverted. If one wants to pass something to libgfortran, use FCFLAGS (which actually may need to get added to the toplevel to pass down to libgfortran). For C++ use CXXFLAGS. That's what those two flags are for. Great, but why should that matter you ask? Because autoconf considers CPPFLAGS to be a precious (unchanging) variable, but not CFLAGS. So, reverting Sean's patch should fix both problems even with the first two bugs. Finally, I'm pretty sure that this bug is misfiled, since is a bug within our build machinery. But we don't have category for that. This can happen else where so removing target. Created attachment 7954 [details]
Base target CPPFLAGS on CPPFLAGS_FOR_TARGET
Could someone please verify that this patch fixes the Alpha bootstrap bug.
I can confirm that the patch addresses the issue from comment #1, which is a good reason to go for that patch (even in case the alpha issue might not be addressed). Thanks! Subject: Re: [4.0 Regression] libjava bootstrap failure on Tru64 UNIX: CPPFLAGS changed in libltdl kcook at gcc dot gnu dot org writes: > Could someone please verify that this patch fixes the Alpha bootstrap bug. It does: I've successfully bootstrapped mainline on alpha-dec-osf5.1b with this patch: http://gcc.gnu.org/ml/gcc-testresults/2005-01/msg00655.html While there are many C++ execution failures (which seem to be easily fixable), the Java results are quite similar to those from July 2004 (when I last successfully bootstrapped on Tru64 UNIX V5.1B). Thanks alot. Rainer Subject: Bug 18222 CVSROOT: /cvs/gcc Module name: gcc Changes by: kcook@gcc.gnu.org 2005-01-17 23:50:27 Modified files: . : ChangeLog Makefile.def Makefile.in Makefile.tpl Log message: 2005-01-17 Kelley Cook <kcook@gcc.gnu.org> PR bootstrap/18222 * Makefile.def: Pass CPPFLAGS_FOR_TARGET. * Makefile.tpl: Define target CPPFLAGS on CPPFLAGS_FOR_TARGET. * Makefile.in: Regenerate. Patches: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/ChangeLog.diff?cvsroot=gcc&r1=1.1045&r2=1.1046 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/Makefile.def.diff?cvsroot=gcc&r1=1.45&r2=1.46 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/Makefile.in.diff?cvsroot=gcc&r1=1.253&r2=1.254 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/Makefile.tpl.diff?cvsroot=gcc&r1=1.123&r2=1.124 Fixed. |