Re: Avoid a spurious "*ABS*" symbol (was: Re: stage one of gas reloc rewrite)

[Alan Modra <amodra at bigpond dot net dot au>]

On Sun, Sep 01, 2002 at 06:22:13PM -0400, Hans-Peter Nilsson wrote:
> I'm confused; it looks like the code and comment has things

Sorry for the lack of clarity.  The comment is in adjust_reloc_syms,
but the place where we really need to add the symbol is in
fixup_segment, on pcrel fixups.

> - Do you (Alan and other people) agree this (handling fx_addsy
>   NULL for fixups for constant expressions -- usually fx_pcrel)
>   is the right route?

Well, in some ways having the abs symbol is a hack.  However, the
common case is that relocations have a symbol.  By not putting a
symbol on pcrel fixups, where we do want a reloc, we will probably
require code in lots of places to test for this case.  I'm inclined to
say no, unless you can prove me wrong about the extra code.

> - What amount of testing would be needed?  Would it suffice that
>   none of the targets you (Alan) mentioned have test-suite
>   regressions?  Or what else would be needed?

That set of targets give fairly good test coverage, modulo how good
the actual target tests are!

> Some alternatives I have *not* tried:
[snip]

I'd be inclined to strip out the symbol using obj_frob_symbol, unless
no object format needs it.  In which case it can be stripped out by
adding a test near the place where EMIT_SECTION_SYMBOLS is used.

I haven't properly researched whether some object formats need the
symbol in the object file.  I just tried getting rid of it in some
early versions of the patch and ran into problems.  With COFF, from
memory.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre