[pulseaudio-discuss] [RFC][PATCH] improve missing handling in memblockq (was Re: stream wedged in non-playing state)

[Arun Raghavan]

On Fri, 22 Jul 2016, at 03:24 PM, Arun Raghavan wrote:
> On Fri, 22 Jul 2016, at 02:45 PM, Arun Raghavan wrote:
> > 
> > 
> > On Thu, 30 Jun 2016, at 02:43 AM, Ulrich Eckhardt wrote:
> > > Pierre Ossman wrote:
> > > > Triggering the bug requires a specific amount of data to be drained,
> > > > which is difficult to achieve in any sensible manner just by being a
> > > > client. :/
> > > > 
> > > > I added the attached test though. It doesn't test the full scope of
> > > > the bug as it doesn't include the native protocol side of things, but
> > > > it should verify correct minreq behaviour in the core (which in turn
> > > > should avoid bugs further out).
> > > 
> > > Hi!
> > > 
> > > I recently tried to look at this report, too, and couldn't quite
> > > understand what initially caused the issue, so I started writing tests
> > > for various aspects of the memblockq code. The results so far are
> > > available at https://github.com/UlrichEckhardt/pulseaudio in the
> > > feature/pa_memblockq_tests branch, in case anyone wants to take a look
> > > or pull them.
> > > 
> > > There is one commit ("Add a test for missing data behaviour") there,
> > > which shows an oddity in the behaviour already: If you exceed the
> > > target length and then drain some data, these drained bytes are
> > > reported as missing via pop_missing(), even if the current level still
> > > exceeds the target length, which seems wrong.
> > > 
> > > I also tried your attached testcase (had to adjust it slightly) and
> > > could also reproduce faults with it. For convenience I uploaded it to
> > > the feature/pa_memblockq_pop_missing branch there.
> > 
> > Thanks for this, and I'll take a look at the other tests you added as
> > well. I'm pushing your patch reattributed to Pierre (since the majority
> > of code in there is still his), and with attribution to you in the
> > comment.
> 
> I think those tests are something we should definitely add. I've pushed
> all but the last which, as you pointed out, fails. Will take a look at
> that now.

With Pierre's patches, there is no "static" vs. "dynamic" missing any
more. Both pa_memblockq_missing() and pa_memblockq_pop_missing() are
consistent. I'm fixing up your test to reflect that and pushing it.

-- Arun