Re: [PATCH v2] PR ld/20828: Move symbol version processing ahead of GC symbol sweep

[Alan Modra <amodra at gmail dot com>]

On Fri, Feb 24, 2017 at 02:28:07PM +0000, Maciej W. Rozycki wrote:
> On Fri, 24 Feb 2017, Alan Modra wrote:
> 
> > > OK to commit.  If we find corner cases where the rearranged code
> > > regresses we'll handle that as we find them.
> > 
> > These seem to be fallout from the above change.
> > 
> > arc-linux-uclibc  +FAIL: ld-elf/pr13195
> > arc-linux-uclibc  +FAIL: ld-elf/pr19698
> > arc-linux-uclibc  +FAIL: Symbol export class test (auxiliary shared object)
> > arc-linux-uclibc  +FAIL: Symbol export class test (final shared object)
> 
>  I'll have a look.  Unfortunately I've only added `arc-linux' to my target 
> list as I was verifying my most recent `readelf' fixes, so I didn't catch 
> these as PR ld/20828 regressions. :(

The problem is elf_arc_size_dynamic_sections allocates section
contents for all dynamic sections.  That's naughty because
.gnu.version_d now already has version defs.  Those versions are
therefore lost so they do not drain from the strtab, resulting in
assertion failures.  The arc backend needs fixing.  I'll do it
tomorrow.

-- 
Alan Modra
Australia Development Lab, IBM