Re: valgrind still unhappy despite expect changes

Jorge wrote:
> On Sun, Aug 28, 2011 at 06:53:54PM +0000, corvid wrote:
> > >   I can't reproduce this with valgrind-3.6.1 on wheezy.
> > > 
> > >   No valgrind error report before or after these debug messages:
> > > 
> > >     a_Nav_cancel_expect_if_eq:: calling a_Nav_cancel_expect
> > >     Dillo: normal exit!
> > > 
> > >   I  tested by pushing a remote URL on CLI and closing the window
> > > before   the  page  data  starts  flowing  (i.e.  to  cancel  the
> > > "expect"),  and  also  with  the  cancelled  download  code path.
> > > Neither one produced an error.
> > 
> > I saw that there was another log entry the other day
> > http://starurchin.org/dillo/valgrind/7ea9fc809376ddf7dde2908e2ecf999aea274130.html
> > 
> > Do you think something sneaky could be going on, or does it seem
> > like it must come down to configuration differences somehow?
> 
>   AFAIS, either one is possible...
> 
>   Me not being able to reproduce it says it's most probably be OK
> for the codepaths I tested (FWIW, the ones I can see). If there's
> an untested codepath, well...
> 
>   OTOH,  valgrind produces some false positives (its docs say it)
> and gcc optimization can also play a role...
> 
>   So,  I'd  try to find a simple test case that produces the log.
> Then, test this problematic codepath here, and see what comes out
> of the exercise.
> 
>   Can you reproduce the bug? Do you have a simple test case?

No. I haven't really scrutinized the code at all. I've tried a little
"I wonder whether this could do it" just from looking at the traces
supplied by valgrind, but without hitting upon the trick.