Re: ld-elf/exclude1.s, exclude.exp

[Daniel Jacobowitz <drow at false dot org>]

On Mon, Oct 18, 2004 at 02:59:32PM -0400, Hans-Peter Nilsson wrote:
> On Sat, 16 Oct 2004, Hans-Peter Nilsson wrote:
> 
> > On Sat, 16 Oct 2004, Daniel Jacobowitz wrote:
> > > On Sat, Oct 16, 2004 at 03:36:00PM -0400, Hans-Peter Nilsson wrote:
> > > > ERROR: /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/exclude1.s: assembly failed
> > > > Why not just " .long exclude_symbol"?
> > >
> > > I figure if I pick a non-PC-relative reloc someone won't have one in
> > > the right size.  But if you think .long is more likely to work, then
> > > that's fine.
> >
> > Use of .long with a symbolic operand is already in multiple
> > places in ld-testsuite so I think that's sufficient.
> 
> Wait... I take that back: the code is going into a DSO, so an
> absolute reloc is going to get problems with targets where ld
> complains about the result being textrel: "you probably meant to
> compile with -fPIC" and similar.

I knew there a reason I wanted to use a PC-relative relocation...
however, this would work just as well if you put it into .data instead.
Does that work?

> However, I don't want "make check" to exit with error as it
> complicates my autotester.  One error-exit setter is
> "unresolved".  Is it ok to just mark the test as "unsupported"
> if assembly fails?  Another setter is "perror".  Quite
> redundantly, as callers use the function return code.  Is it ok
> to use "warning" instead?  If any answer is "no", I'll just add
> cris-*-elf to the early exit in exclude.exp.

The change to use unsupported, in any case, makes sense to me.

-- 
Daniel Jacobowitz