Re: Comments wanted! Failures in binutils-2.12 branch testsuite with gcc-3.1 branch
Haven't done full-builds since I worked for Cygnus, but some of this
might be useful anyway...
> Now, what is this _GLOBAL_OFFSET_TABLE_? Where is it? glibc?
Synthesized by the linker for any shared library. Google notes that
there was some discussion on a gcc list of whether this should have
one leading underscore vs. two, which didn't make much sense since it
is ABI-defined...
> What is 'relocation truncated to fit: R_SPARC_13 .rodata.str1.8'?
> What is 'relocation truncated to fit: R_SPARC_GOT13'?
In specific, it means that a bit of code was using a 13-bit wide
sparc-specific relative reference (the former section-relative, the
latter GOT-relative) that didn't fit in 13 bits.
More usefully, that often means that one side or the other of the
relevant reference was undefined, and so the linker defaulted one that
was too far away. I think they're both side effects of not finding
the _GLOBAL_OFFSET_TABLE_ symbol.
Note however that what I *think* is going on is that for at least some
of that, you've got some test cases that should be filtered out:
s-record format is just *not* going to support a GOT, or even
*anything* particularly advanced. Though I guess the later tests are
more real -- did you just select out a few but they *all* fail this
way?
Reply to: