g++.dg/other/static11.C regression

[Zack Weinberg]

Richard Henderson <rth@twiddle.net> writes:

> On Sun, Oct 19, 2003 at 12:48:03PM -0700, Zack Weinberg wrote:
>> I think Jan's solution is the right one.  The assembler will object to
>> fake internal register names/numbers, so we don't lose the ability to
>> detect this bug.
>
> I'm not sure that's correct.  I think that in the case of
>
> 	mov AP, %eax
>
> that it'll just be interpreted as a symbol.

It comes out as "%whatever" in AT&T mode, but I think you're right in
Intel mode.  There's also the related problem that the entry in
REGISTER_NAMES for ARG_POINTER_REGNUM is "".

What do you think of having print_reg alter its behavior depending on
whether file == asm_out_file?  That lets us restore the abort for the
case where it matters; we could also restore the printout of the
internal register number.  Although I confess I like 

            (clobber (reg:CC %eflags))

better than

            (clobber (reg:CC 17 %eflags))

zw