Re: Versioning mess proved!!!

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

On Sat, 30 Sep 2000, Ulrich Drepper wrote:
> Jack Howarth <howarth@fuse.net> writes:
> > 6) now rebuild the zlib srpm on the glibc 2.1.94 machine and install it.
> > Attempt to run the ./example
> > created in step 5. You will get a segfault.
>
> I cannot reproduce this.  I followed your instructions and it works
> just fine for me.  On x86 that is.  You haven't mentioned which
> architecture you tried it on but I assume PPC.
>
> There are two things to do:
>
> - you have the binary, just use the debugger.  At least get a backtrace.
>
> - run
> 	readelf -a <BINARY>
>
>   on both executables and diff the output.
>
>
> I don't have access to a PPC machine in the moment and therefore
> cannot help.

Hmm, it segfaults in call_gmon_start cause __gmon_start__ gets relocated and 
thus is no longer NULL.
The problem seems to be that __gmon_start__ ends up as an untyped weak in 
glibc22:

murx22:  0001e994  08114 R_PPC_GLOB_DAT        00000000  __gmon_start__       
     + 0
murx22:   129: 00000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__
murx22:   248: 00000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__

whereas on glibc21 this was weak function:

murx21:  0001debc  08215 R_PPC_JMP_SLOT        0000b670  __gmon_start__       
     + 0
murx21:   130: 0000b670     0 FUNC    WEAK   DEFAULT   14 __gmon_start__
murx21:   256: 0000b670     0 FUNC    WEAK   DEFAULT   14 __gmon_start__

AFAICS this is generated from sysdeps/generic/initfini.c into csu/initfini.s, 
which only gets 

/*@_fini_EPILOG_ENDS*/
 
/*@TRAILER_BEGINS*/
        .weak   __gmon_start__
        .ident  "GCC: (GNU) 2.96 20000928 (experimental)"

at the end. The result is the same whether I use gcc-2.95.3 or gcc-2.96 as of 
yesterday. Any idea how this can be fixed?

Franz.