Spurius results for cygwin selftest.exp(was Re: [RFC] fix annoying completion bug on directories)

["Pierre Muller" <muller@ics.u-strasbg.fr>]

> >   I ran the testsuite on cygwin, and found one
> > change, but in gdb.gdb/selftest.
> >   FAIL (timeout) xgdb is at prompt
> >
> >   I suspect that this is due to a problem with the
> > [New thread %s] output, that are now default.
> > Because this output appears right after the gdb prompt,
> > and thus the match with a gdb prompt exactly at the end
> > is not found.
> 
> In these cases, in addition to analising the log file, please
> rerun the test to confirm it was spurious.  Cygwin testing
> is indeed flackier than linux.

  This is indeed a spurious FAIL that has to do with that
third thread that start on cygwin right after the command prompt is
printed out.
  Does anyone know why there is a third thread starting?
Debugging threads under cygwin is quite difficult, you never really know
where they are coming from...

  I am really surprised that now, by default, gdb will print 
debug message, while the inferior is running, without stopping it.
  For me, when gdb is started with the default options,
gdb should only output text if it stops the debuggee, otherwise
it becomes difficult for the average user to handle.
  But I understood that the [New thread... message was
already the default on Linux targets..
> 
> Just do make check RUNTESTFLAGS="failingtest.exp" a couple of
> times.  If it's always failing now, it's a regression.  If
> the failure goes aways, then also retest with with the
> patch unapplied.  You should see the flackiness too.  If
> not, it may be a regression, but you'll have to judge from
> the logs -- it may just be difficult to reproduce flackiness.

  I got it give PASS or FAIL several times,
it's really a question of how much time there is
in between the (gdb) prompt and the [New thread...] message.

  I could of of course update the selftest.exp test
to also accept the "[New thread ...]" message,
but I would like your opinions about this before,
we could easily suppress these messages by setting
print_thread_events to zero by default.


Pierre Muller
Pascal language support maintainer for GDB