Re: MIPS md_apply_fix()(?) problem.

[Daniel Jacobowitz <drow at mvista dot com>]

On Mon, Nov 19, 2001 at 03:58:33PM -0800, Ian Lance Taylor wrote:
> Alan Modra <amodra@bigpond.net.au> writes:
> 
> > On Mon, Nov 19, 2001 at 04:27:10PM +0000, Nick Clifton wrote:
> > > 
> > > Then I was going to add a new bitfield to the fixup which would
> > > fixup_segment would set if it had added in the symbol's value before
> > > calling md_apply_fix3.  Backends could then examine this flag and undo
> > > the addition if they wanted to.  In fact it might be better to have a
> > > target specific macro that fixup_segment uses to check to see whether
> > > it should add in the symbol's value in the first place.
> > 
> > I'm of the opinion that there's a serious problem with the md_apply_fix3
> > interface.
> >   a) We have hacks and tweaks inside a function to undo the caller's
> >      actions.  That shouldn't be necessary.
> >   b) Comments like the following, taken from tc-ppc.c
> >       /* FIXME: Why '+'?  Better yet, what exactly is '*valuep'
> > 	 supposed to be?  I think this is related to various similar
> > 	 FIXMEs in tc-i386.c and tc-sparc.c.  */
> > 
> > Of course, pointing out architectural problems is a little different
> > to fixing them, and I haven't studied the code with a view to rewriting
> > it.  :)
> 
> See bfd/doc/bfdint.texi, the node ``BFD relocation handling.''

Wearing my reluctant-release-manager hat:

This sounds like a good thing to tackle at the beginning of a
development cycle and a bad thing to tackle at the end.  I'd like to
ask that no one get adventurous (at least about committing related
patches :) for a little while; hopefully we can branch relatively soon.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer