[PATCH] Fix PR middle-end/15486

[Eric Botcazou]

> > I just realized that this might not be totally correct.  We might need to
> > clear DECL_SECTION_NAME when asm_emit_uninitialised returns early.
>
> Why?

Because invoking again resolve_unique_section on the decl will be a no-op.  
Now the second invocation takes into account the 'reloc' parameter and I 
think there are cases in which it also must be taken into account for decls 
that were tentatively emitted by asm_emit_uninitialised, mostly those 
benefiting of flag_zero_initialized_in_bss.

Of course the modified patch doesn't get prettier...

-- 
Eric Botcazou