Re: [rfa/testsuite] make annota1 regexps more generous

[Daniel Jacobowitz <drow@mvista.com>]

On Mon, Sep 30, 2002 at 03:24:59PM -0700, David Carlton wrote:
> I noticed today that annota1.exp seems to be generating some spurious
> FAILs on my machine, namely
> 
> FAIL: gdb.base/annota1.exp: breakpoint info
> FAIL: gdb.base/annota1.exp: backtrace from shlibrary
> FAIL: gdb.base/annota1.exp: send SIGUSR1
> FAIL: gdb.base/annota1.exp: break at 28
> 
> In all cases, there's a regexp that looks for
> ${srcdir}/${subdir}/${srcfile}, but the ${srcdir}/ component is
> missing.  It doesn't seem to me that that should cause a fail; here's
> a patch that makes the ${srcdir}/ component optional.
> 
> Because of Kevin's recent patch to the breakpoint info failure, it
> makes the most sense to me to make ${subdir}/ optional as well, given
> that he has an instance where, on one (but not all?) of those tests,
> both ${srcdir} and ${subdir} are missing.  So that's what I've done.
> (The patch looks messy, but that's just because the regexps in
> question are so big: all I'm doing is adding a few parentheses and
> question marks.)
> 
> Of course, it's possible that this really is a regression and that I'm
> not correctly understanding what those tests are looking for.  I'm
> using GCC 3.1 on Red Hat 7.3, for what that's worth.
> 
> Just out of curiosity, how many unexpected failures should I be
> getting?  I'm usually getting 100 or so, which seems like an
> unfortunately large number to me: either GDB has lots of regressions,
> or the testsuite is misdiagnosing passes as failures, or there are
> lots of FAILs that should be changed to XFAIL.  I'm hoping that many
> of them are misdiagnoses; maybe I'll spend some time looking at that
> when I'm not teaching and when I'm sick of symbol tables.

Depends on your compiler.  I had it down to a dozen or so MI failures
on x86 and a couple of XPASS's, using GCC 2.95.3.  3.2 is higher
because I need to finish a lot of C++ work... then there are some
random failures (a la pthreads; unpredictable).  And new regressions of
course.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer