896374 - [A/V] Video codec is blocked, and no video file can be played.

Status: RESOLVED FIXED

(Firefox OS Graveyard :: Gaia::Video, defect, P1)

Description

[leo.bugzilla.gecko]

STR

1. Play video file on device 1.
2. Send any video file from device 2 to device 1.
3. In device 1, receive file and try to play it from the notification, while first one is still playing.
4. Device 1 remains black screen with spinner. Then, push pause/start 2~3 times.
5. Push back button.

Then, video frame is stopped and I can hear only audio.
Progress bar isn't changed.
After that, I cannot play any video file until turn off the device.

Comment 1

[John Hu [:johnhu][:johu][:醬糊小弟]]

This bug is already reported in bug 889191. After investigation, the cause is related to bug 887650. I cc alive so that he can test them together.

Comment 2

[leo.bugzilla.gecko]

(In reply to John Hu [:johnhu] from comment #1)
> This bug is already reported in bug 889191. After investigation, the cause
> is related to bug 887650. I cc alive so that he can test them together.

I'm not sure that this problem is related to bug 889191 and bug 887650.
Maybe, this problem will be fixed by that patch.

But I think we have to handle another problem here.

Bug 889191 and bug 887650 is about UX problem.
However, the main issue here is "After this problem, every video file cannot be played at all".

When push back button in STR 5 in comment #0, 
Video codec is suddenly stopped without any state change.
And even though I push back button, it doesn't release A/V codec. 

So, no media file start to play because it cannot use any H/W codec.

Comment 3

[Sotaro Ikeda [:sotaro]]

Sorry, I made this bug at MediaResourceManagerServic e:-(

Comment 4

[Sotaro Ikeda [:sotaro]]

Somehow, following code does not work.
>   while (it != mVideoCodecRequestQueue.end()) {

Comment 5

[Sotaro Ikeda [:sotaro]]

I become clear that I hit android::Vector's bug. Following code does not work correctly. It assumes the Vector uses same arryBuffer. But it is not always correct. There is a case that the arryBuffer is reallocated. In this case it does not work.
 > it = mVideoCodecRequestQueue.erase(it);

Comment 6

[Sotaro Ikeda [:sotaro]]

Attachment: patch - fix inifinite loop in MediaResourceManagerService::cancelClientLocked()

Changed not to use the iterator returned from mVideoCodecRequestQueue.erase(it). Current gecko continues to check all mVideoCodecRequestQueue's items even when the item is found. It is not necessary and changed to break when the item is found.

Comment 7

[Sotaro Ikeda [:sotaro]]

Confirmed the patch works on v1.1 leo.

Comment 8

[leo.bugzilla.gecko]

I also checked that the patch works.
And if you change vector from android to STL. It also works :)

Thanks!!.

Comment 9

[Sotaro Ikeda [:sotaro]]

Thanks for the confirmation. I did not choose STL bacause of Bug 850576. When we need to use the following in future, It's better to change to STL.

> iterator returned from mVideoCodecRequestQueue.erase(it).

Comment 10

[Sotaro Ikeda [:sotaro]]

Attachment: patch v2 - fix inifinite loop in MediaResourceManagerService::cancelClientLocked()

Add header to a patch. Carry "chris.double: review+"

Comment 11

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/projects/birch/rev/6c5f376186a0

Comment 12

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/6c5f376186a0

Comment 13

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-b2g18/rev/7a8815ce373f

Comment 14

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-b2g18_v1_1_0_hd/rev/7a8815ce373f