real.c patch comments?

John David Anglin
Sun Jun 23 11:59:00 GMT 2002

> > I'm going to run these tests today.  I've completed the before tests.
> It does seem to improve things.  (look at gcc-testresults to see
> the two runs).  Though it's hard to tell with g++ because cc1plus
> is emitting this for main routines that don't have a return value:
>	movl $1,%r0
> 	ret
> which causes a non-zero status to be returned.  i386 and powerpc
> return a 0.  Any idea why VAX gets a 1 instead of 0?  [1 would be
> success under VMS but I hope that isn't the reason why.]

No.  This doesn't happen with 3.0.

I had a bootstrap failure in stage2 in my most recent build compiling
real.c.  The preprocessor incorrectly processed the file resulting in
NANS being defined.  DEC is 1, so this shouldn't have happened.  I was
testing a couple of patches so I decided to redo the bootstrap.  I would
suspect that there are still problems with casesi but it could be
something else.

The was a patch affecting returns on 2001-08-09.

Can you dump the rtl with "-da"?  This might provide some clues.

J. David Anglin                        
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

More information about the Gcc-patches mailing list