Re: RFA: Avoid automodification in ARM call addresses

[Ian Lance Taylor <ian at airs dot com>]

Mark Mitchell <mark@codesourcery.com> writes:

> Richard Sandiford wrote:
> 
> >>Before committing your patch, can you look into fixing reload so that
> >>it at least crashes on this case rather than generating invalid code?
> >>For example, should the assert in do_output_reload check
> >>NONJUMP_INSN_P instead of JUMP_P?
> > 
> > I'm reluctant to try this myself, to be honest.  It's not the kind
> > of thing you can really test by bootstrapping & regression-testing
> > on one port.  If someone is motivated to try this, they can easily
> > back out my patch and use ARM as a testbed.
> > 
> > Sorry, I know that sounds lazy, but there only so many hours in
> > the day. ;)
> 
> Ian, can you live with that?  Or, do you think this is important enough
> that we should hold off on the patch until the reload assert gets changed?
> 
> I'm not trying to pressure you; I'm just catching up on old email, and

Sure, I can live with that.

I was more or less assuming that an ARM maintainer would approve
Richard's patch--I hadn't bothered to pull up my ARM manual to double
check his patch.

Looking at it again, it seems simpler to change the 'm' constraint to
'o', rather than changing the insn condition at all.

And I don't see why call_value_mem specifically rejects a constant
address.

Ian