Re: [RFA/TESTSUITE] annota1.exp fixes

[David Carlton <carlton@math.stanford.edu>]

On Thu, 10 Apr 2003 13:34:42 -0400, Elena Zannoni <ezannoni@redhat.com> said:

> Hmm, I've reread the threads and I am none the wiser. We didn't ask the 
> emacs people, like Eli suggested.

No. :-(

> I have a revised patch, which works for my problem, but doesn't help
> you.  I was wondering if using something like matching on
> (${longversion}|${shortversion}) would work.

It's not the end of the world if my problem never gets solved: I'm
used to seeing those failures, after all.  It's a strange situation:
my srcdir is just '.', so ${srcdir}/${subdir}/${srcfile} really is
equivalent to ${subdir}/${srcfile}, GCC is just noticing that it can
leave out the './' at the start when generating debug info.  (I'm
pretty sure I checked at the time and made sure that it's GCC that's
getting rid of the './', not GDB.)

So I'm happy enough with your patch; on the other hand, it will cause
Kevin to start seeing failures with the IRIX compiler again, if I'm
reading it right.  How about this as a compromise: file a GDB external
suspended PR mentioning that some compilers don't always put full path
names in the debug info, have a pass branch with your new regexp, and
have an xfail branch associated to that PR with your original more
generous regexp?  (I'm not sure we do this in the testsuite anywhere
yet, but my understanding is that xfail takes an optional second
argument, which is a PR number.)  I think that's a good compromise:
the test should remain useful, because if GDB ever regresses on this
issue then presumably somebody will notice a pass->xfail transition.
And we'll only be saying 'pass' in situations where we're sure we're
doing the right thing.

David Carlton
carlton@math.stanford.edu