Mixed sparc -m32/-m64 testsuite failures

[Kaveh R. Ghazi]

 > From: Richard Henderson <rth@redhat.com>
 > 
 > On Sat, Apr 13, 2002 at 11:57:10AM -0700, David S. Miller wrote:
 > > Richard, if his platform can run 64-bit binaries, why is he
 > > getting this situation? :-)
 > 
 > It can.  IIRC, he's got one of each.
 > r~

No no, I test one platform that used to "config.guess" as sparc-* and
now after Richards change to config.guess it "guesses" as sparcv9-*.
But that's irrelevant other than the fact that tests which used to get
bypassed because they required sparcv9-* now get run.

I run the testsuite with extra passes added in RUNTESTFLAGS to test
-m64.  Look at the top of gcc.dg/20001101-1.c:

 > /* { dg-do run { target sparcv9-*-* sparc64-*-* } } */
 > /* { dg-options "-O2 -m32 -mcpu=ultrasparc -mvis" } */

See it specifies its own -m32, and when my RUNTESTFLAGS adds -m64 you
have both options listed and it spews diagnostics which I showed in
the original report:
http://gcc.gnu.org/ml/gcc-patches/2002-04/msg00714.html

>>So if I wasn't clear in my original email, what I'm reporting is a
bug in command line option handling.<<

One of these two options should override the other, when both are
passed it currently behaves as if both were in effect.  Contrast this
with -O0 followed by -O2 which does the right thing.

According to Richard, there's no way to correctly handle this in
specs.  Is that understanding correct?

		--Kaveh
--
Kaveh R. Ghazi			Director of Systems Architecture
ghazi@caip.rutgers.edu		Qwest Global Services