[pulseaudio-discuss] [PATCH] stream: Return error in case a client peeks to early

[Tanu Kaskinen]

On Tue, 2012-11-06 at 08:36 +0100, David Henningsson wrote:
> On 11/05/2012 07:13 PM, Tanu Kaskinen wrote:
> > On Sun, 2012-11-04 at 23:09 +0100, David Henningsson wrote:
> >> On 11/04/2012 02:22 PM, Tanu Kaskinen wrote:
> >>> On Sat, 2012-11-03 at 19:51 +0100, David Henningsson wrote:
> >>>> Returning NULL seems to be the right thing to do here, even if
> >>>> gnome-control-center does not handle that very well IIRC. So we might
> >>>> need an additional patch in g-c-c.
> >>>> So assuming I commit a patch doing that. If somebody else wants to add
> >>>> logic to figure out how large the hole is, that could be discussed
> >>>> separately.
> >>>> Any objections?
> >>>
> >>> It's not clear what you meant by "add logic to figure out how large the
> >>> hole is". Add to where? pa_stream_peek() or gnome-control-center?
> >>
> >> I was referring to your earlier comment "An addition: the client
> >> probably wants to know how large the hole is.", i e pa_stream_peek.
> >>
> >>> To me, reporting the hole length in the "nbytes" parameter of
> >>> pa_stream_peek() seems like the right thing to do, so I hope your patch
> >>> will do this.
> >>
> >> It does not: I was just suggesting to discuss that separately.
> >
> > Ok. Well, I'm suggesting to not make these issues separate, since it's a
> > small thing to decide and implement, and doing a separate fix would be
> > "fixing a fix".
> 
> I don't have an issue with step-by-step improvements. I see this as such 
> (if the second step is to be seen as an improvement).
> 
> > Does someone have an issue with returning the hole size
> > in the nbytes argument?
> 
> Not if you're volunteering to do the work. All I care about is that PA 
> clients can crash for this reason, and I'd like that to stop happening.

Cool, I certainly volunteer to do the work. It should be trivial, so
I'll probably send a patch today.

> Note: gnome-control-center needs a corresponding fix either way, as it 
> currently checks the length rather than the NULL pointer, IIRC.

Mmm, I guess I should take this task too.

-- 
Tanu