[vlc-devel] [PATCH 3/4] mediacodec: fix GetOutput timeout

[Thomas Guillem]

On Wed, Oct 29, 2014, at 19:08, Martin Storsjö wrote:
> On Wed, 29 Oct 2014, Thomas Guillem wrote:
> 
> >
> > On Wed, Oct 29, 2014, at 18:22, Martin Storsjö wrote:
> >> On Wed, 29 Oct 2014, Thomas Guillem wrote:
> >> 
> >> > ---
> >> > modules/codec/omxil/android_mediacodec.c | 2 ++
> >> > 1 file changed, 2 insertions(+)
> >> >
> >> > diff --git a/modules/codec/omxil/android_mediacodec.c b/modules/codec/omxil/android_mediacodec.c
> >> > index 8507ebc..e352266 100644
> >> > --- a/modules/codec/omxil/android_mediacodec.c
> >> > +++ b/modules/codec/omxil/android_mediacodec.c
> >> > @@ -738,6 +738,8 @@ static int GetOutput(decoder_t *p_dec, JNIEnv *env, picture_t **pp_pic, jlong ti
> >> >             (*env)->ExceptionClear(env);
> >> >             return -1;
> >> >         }
> >> > +        // timeout is only for the first try
> >> > +        timeout = 0;
> >> >
> >> >         if (index >= 0) {
> >> >             if (!p_sys->pixel_format) {
> >> > -- 
> >> > 2.1.0
> >> 
> >> Umm.. maybe? The commit message is a bit too terse, the change might be 
> >> right but I'm missing the explanation on why and in which case it
> >> matters.
> >
> > Ah yes. You can pass a timeout to GetOutput, that will pass it to the
> > dequeue function.
> > The dequeue can return an event instead of a buffer. If GetOutput handle
> > it, it'll call again the dequeue function with the same timeout. In that
> > case, we wait 2 x timeout.
> 
> Yeah, but I don't think that's too much of an issue - we only set a 
> nonzero timeout when we didn't get any input buffer, so we really want to 
> block waiting for something "useful". OTOH setting it to zero shouldn't 
> hurt more than another round of waiting for input/output buffers...
> 
> > I'll redo the commit log & comments if you agree with this commit.
> >
> > (But finally, I decided to use a timeout of 0 in the next commit)
> 
> Well if you choose to never pass any non-zero timeout in the threaded 
> setup, I guess it's simpler to just skip this patch?

I agree, let's skip it.

> 
> // Martin
> _______________________________________________
> vlc-devel mailing list
> To unsubscribe or modify your subscription options:
> https://mailman.videolan.org/listinfo/vlc-devel