Hi Kevin .. On Mon, Jul 08, 2002 at 11:44:33AM -0400, Kevin.Hendricks wrote: >So we have the *exact* same symbols you recompiled your glibc -2.2.5 >with the libgcc-compat patch from the libc-alpha mailing list (and used >a gcc-3.X compiler to do it .. right). ehm ... the libgcc-compat patch is not included into the glibc .. because there was much to do, the most changes are done but not all .. so I decided to leave it out. But this patch is used, if you compile glibc with gcc-3.1, and that is not the case in debian this time. >If so I can't understand why mozilla has missing symbols issues with the >build. >That is very strange. >So to recap - we have 3 problems to squash >1. my original build fails with sigill on your machine but when >cppuhelper is recompiled with -O0 the sigills go away (I have updated >from gcc 3.1 to gcc 3.1.1-pre to try and stomp this one out but Daniel >seems to think this is a bug in our new gcc3 bridge code that I have to >track down >2. once past #1, when mozilla address book stuff starts up you get a >segfault or a missing >symbol problem. The issue here is glibc 2.2.5 related and has 4 >possibilities: >- stock glibc-2.2.5 built with gcc-2.95.X this is the glibc in debian :) >I seem to be using glibc-2.2.5 with libgcc compat patch built with gcc >3.X We have to build glibc-2.2.5 build with gcc-2.95, because gcc-3.x is not supoorted in debian yet. >But this seems to introduce a versioned and non-versioned symbols >related to __udivdi3 that somehow appear in the mozilla libraries >libnspr4.so, libxpcom.so, and libmozjs.so. yes .. >The question is what is the correct way to have everyone be compatible >with binaries built with gcc-3.X on the different types of glibc 2.2.5 >combinations? I can't believe, that this is so difficult, because, glibc is not build with g++ .. so there should not be any incompatibilities .. but .. *wait* if you build your glibc with gcc-3.x, you will pointed out, that you will break any compatibility to other programs, which linked against the glibc, which is build with gcc-2.95. OK, I will say, the OpenOffice.org team has to build OpenOffice.org twice. .... >Do you have access to a glibc-2.2.5 without libgcc-compat patch built >with gcc-2.95.X that we could compile Mozilla 1.0 on and check what >versions of that symbol exist in those mozilla libraries? Whatever they >are, we have to be the same since the majority of systems out there are >using that combination I would expect. The debian-glibc is that, what you are looking for. The mozilla-binaries can be found here: http://bourbaki.math.uni-kiel.de/~jpalic/OpenOffice.org/ >3. Your cppu build with debug=TRUE fails since the alignment issue of >structure inherited from a parent is showing a different alignment in a >structure than the code wants. >Hopefully you have just somehow improperly defined #define >MAX_ALIGNMENT_4 in cppu somehow. If not, then something has changed >seriously about alignment of inherited structures inside structures that >follow doulbes, ints, etc I will try that, when my OOo build has finished! *build* Regards jan -- .''`. Jan-Hendrik Palic | : :' : ** Debian GNU/ Linux ** | ** OpenOffice.org ** ,.. ,.. `. `' http://www.debian.org | http://www.openoffice.org ,: ..` ` `- jan.palic@linux-debian.de | ' ` `
Attachment:
pgpjekacyMSML.pgp
Description: PGP signature