Red Hat Bugzilla – Bug 150115
libxml2 build breaks with gcc4 on s390
Last modified: 2007-11-30 17:11:01 EST
From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.5) Gecko/20041215 Firefox/1.0 Red Hat/1.0-12.EL4 Description of problem: rebuilding libxml2 in beehive broke. First I get tons of aliasing warnings. Then the compile stops with an exit code 1 by gcc4 on the schemas.c file, while showing only warnings. Version-Release number of selected component (if applicable): gcc-4.0.0-0.30 How reproducible: Always Steps to Reproduce: I tried to rebuild twice failed twice 1.rebuild libxml2-2.6.17 on s390 2. 3. Actual Results: spits thousands and thousands of warnings dies in the xmlschemas,c module Expected Results: compile or at least emit an error why it could not compile Additional info:
Created attachment 111581 [details] Logs from the build
Libxml2 code seems to compile fine with gcc4-4.0.0-0.22 on i686, Daniel
FWIW, this problem also affects glib2 and gtk2, which also use a PLT-reduction technique based on aliasing.
Any chance you could try a rebuild with -no-suppress in the flags passed to libtool ... gcc? This will drop the redirection of the PIC compile error output to /dev/null, and will at least let us hazard a guess as to the actual problem.
Sorry I really don't know libtool, I have no idea how to influence this. Daniel
Ok, I managed to duplicate the error. It's the profile generation option that's triggering the bug. You may want to temporarily disable that to get a successful rebuild. As for knowing or not knowing libtool, whether you do is irrelevant to my request. I only suggested you to adjust the build machinery of your package (which presumably you're familiar with) to pass an additional flag to the libtool compiler combo when compiling this particular file. Your reply is similar to `I don't know compilers, so I can't pass flags to them'. I.e., I was not asking you to modify libtool or do anything with it, only pass whatever flags to it that you already pass, plus this one flag I suggested.
Got this on ia64 without removing the profile generation yet, it is on a different file than where it failed on s390: xpointer.c: In function 'xmlXPtrLocationSetRemove__internal_alias': xpointer.c:760: internal compiler error: in simplify_subreg, at simplify-rtx.c:3674 Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://bugzilla.redhat.com/bugzilla> for instructions. Preprocessed source stored into /tmp/cchIZg04.out file, please attach this to your bugreport. w.r.t. to your request, I never invoke libtool directly myself. I am not familiar with the auto* machinery, and have no idea how 1/ to pass libtool flags 2/ that -no-suppress is a valid libtool flag I use AM_PROG_LIBTOOL in configure.in, and have no idea what's going on behind it. Here is my level of knowledge on libtool: paphio:~/XML -> grep -i libtool configure.in Makefile.am configure.in:AC_LIBTOOL_WIN32_DLL configure.in:AM_PROG_LIBTOOL configure.in:XML_LIBTOOLLIBS="libxml2.la" configure.in:AC_SUBST(XML_LIBTOOLLIBS) paphio:~/XML -> man libtool No manual entry for libtool paphio:~/XML -> man automake No manual entry for automake paphio:~/XML -> it's a black box to me, not just libtool, the whole build, and getting things fixed in it is purely a modify-try-fail-retry process whitout any understanding of the beast. Sad but true, I control my code, but not the build. Daniel
Created attachment 111660 [details] the output of gcc4 failure on xpointer on ia64
The s390 -fprofile-use bug has been fixed upstream and will make it into -0.33 soon. I'll look at the ia64 ICE.
Okay, fixes on libxml2 side have been integrated in upstream, thanks Daniel
The IA64 ICE is PR20249.
Both the s390 and ia64 -fprofile-{generate,use} bugs should be fixed in gcc-4.0.0-0.33.
Okay I will have to regenerate libxml2 packages, probably over the week-end, then I will be able to give comparative performances figures that arjan asked :-) thanks ! Daniel