don't assume pointer cast to unsigned long is a valid initializer

[Geoff Keating]

> Cc: mark@codesourcery.com, rth@redhat.com, gcc-patches@gcc.gnu.org
> From: Alexandre Oliva <aoliva@redhat.com>
> Organization: GCC Team, Red Hat
> Date: 03 Mar 2003 17:10:49 -0300

> On Mar  3, 2003, Geoff Keating <geoffk@geoffk.org> wrote:
> 
> > You're saying it doesn't work.  I said it should work.
> 
> I've already explained that there are ports in which it can't possibly
> work,

Could you name some of these ports?  I've named a number of ports on
which it should work, and on which I think this test could differ from
20011114-1.c.  For reference, here's the list again:

- All mips-elf platforms (due to R_MIPS_64)
- avr-elf (R_AVR_32)
- powerpc-XCOFF
- sparc-elf (R_SPARC_64)
- xstormy16-elf (R_XSTORMY16_32)
- h8300-elf (R_H8_DIR32_X)

> and on most ports your test is redundant with another test we
> already have.  You're trying to force the world to accept your notion
> that every post must have a relocation that is as wide as a long.

No!  I'm pointing out that most ports already have a relocation that
is as wide as a long, and that in fact I haven't yet found any port
that doesn't support this.  I note that you haven't named one either.

Maybe if you could post the name of the port that you originally found
this failing on, we could look at that port and decide whether or not
it could support this if GCC was fixed.  

At the moment, even if we do find a port that can't support this, it
looks like most ports can support this functionality and so we should
default it to being run, rather than not run.

-- 
- Geoffrey Keating <geoffk@geoffk.org>