glibc 2.1.94-2 breaks the postfix build and the resulting binary (fails to recognize comments in its configuration file). glibc 2.1.92-14 (from 7.0) is ok. I don't have the time to trace this down at the moment; attaching the source RPM showing the problem for now. The problem seems to be specific to x86.
Created attachment 4110 [details] Postfix SRPM that doesn't work with new glibc
Are you sure it worked against glibc-2.1.92-14? The failing program AFAIC is bin/postconf, which if I run against glibc 2.1.92-14 gives the same errors: LD_TRACE_LOADED_OBJECTS=1 /usr/src/2.1.92-14/ld-linux.so.2 --library-path /usr/src/2.1.92-14/ bin/postconf -c /var/tmp/postfix-20000924-root/etc/postfix -e 'alias_maps = hash:/etc/postfix/aliases' 'alias_database = hash:/etc/postfix/aliases' libdb.so.2 => /usr/lib/libdb.so.2 (0x4000a000) libnsl.so.1 => /usr/src/2.1.92-14/libnsl.so.1 (0x40018000) libresolv.so.2 => /usr/src/2.1.92-14/libresolv.so.2 (0x4002f000) libc.so.6 => /usr/src/2.1.92-14/libc.so.6 (0x40041000) /lib/ld-linux.so.2 => /usr/src/2.1.92-14/ld-linux.so.2 (0x80000000) ls -l /usr/src/2.1.92-14/libc.so.6 /usr/src/2.1.92-14/ld-linux.so.2 lrwxrwxrwx 1 root root 12 Oct 13 11:25 /usr/src/2.1.92-14/ld-linux.so.2 -> ld-2.1.92.so lrwxrwxrwx 1 root root 14 Oct 13 11:25 /usr/src/2.1.92-14/libc.so.6 -> libc-2.1.92.so /usr/src/2.1.92-14/ld-linux.so.2 --library-path /usr/src/2.1.92-14/ bin/postconf -c /var/tmp/postfix-20000924-root/etc/postfix -e 'alias_maps = hash:/etc/postfix/aliases' 'alias_database = hash:/etc/postfix/aliases' bin/postconf: warning: /var/tmp/postfix-20000924-root/etc/postfix/main.cf: multiple entries for key alias_maps bin/postconf: warning: /var/tmp/postfix-20000924-root/etc/postfix/main.cf: multiple entries for key alias_database bin/postconf: fatal: read /var/tmp/postfix-20000924-root/etc/postfix/main.cf: Function not implemented and if I just run postconf with the same arguments, I get the same result. The Function not implemented is total crap, postfix does not know that it can look at errno only if some function actually failed.
Yes. Like I've said, I didn't have the time to look into the problem at the moment. But I've tried compiling it on a couple of machines, and it consistently failed on every x86 machine that has been updated, and worked on all non-x86 machines and on all machines without updates. Try: zell.redhat.de (doesn't work: updated), parma.redhat.de (works: not updated), porky.devel.redhat.com (doesn't work: updated), george.devel.redhat.com (works: alpha), bart.devel.redhat.com (works: sparc) Maybe I messed up and attached the wrong source RPM, the error message I'm getting is different (complains about format errors in comment lines because it can't recognize them as such anymore). Try the one on parma or zell, that's the one I was using.
Actually it looks like gcc bug which was introduced in our gcc between gcc-2.96-54 (the one that made it into the distribution) and gcc-2.96-55. There were 15 patches added, I'll check them out. I've first tried my i386_ge_geu but it looks like that was not it, I suspect Jan Hubicka's patches. CVS head is broken as well.
It is for sure 2.96-55, because downgrading to gcc 2.96-54 even with glibc 2.1.94-3 makes postfix build succeed. The broken patch is gcc-i386-arith.patch from Jan Hubicka, to whom I spoke about it today in person, we decided on how it should be fixed and he's probably just testing the fix out.
Fixed in gcc-2.96-62.