Group AVDA output surface into AVDASurfaceBundle.

media/gpu/avda_surface_bundle.cc

+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "media/gpu/avda_surface_bundle.h"
+
+namespace media {
+
+AVDASurfaceBundle::AVDASurfaceBundle(int id) : surface_id(id) {}

[watk, 2017/02/22 20:38:56]

s/id/surface_id/ to match header

[liberato (no reviews please), 2017/02/23 18:18:46]

Done.

+
+AVDASurfaceBundle::~AVDASurfaceBundle() {
+  // Explicitly free the surface first, just to be sure.

[watk, 2017/02/22 20:38:56]

Unclear what you want to be sure of here. That it's deleted before the ST?

[liberato (no reviews please), 2017/02/23 18:18:46]

yes, done.

+  surface = gl::ScopedJavaSurface();
+  // TODO(liberato): should we ReleaseSurfaceTexture here, in case this isn't
+  // the last reference?
+  surface_texture = nullptr;

[watk, 2017/02/22 20:38:56]

Yes I think so. If you drop the bundle you're dropping the ability to render to
the surface, so there's no reason not to ReleaseST(). Codec images might outlive
AVDA for an arbitrary time.

[liberato (no reviews please), 2017/02/23 18:18:46]

i'm not sure that it's safe, the more that i think about it.  the consumer might
have a back buffer that it wants to swap.  i'll update the comment.

[watk, 2017/02/23 20:03:47]

But this is what we've always done right? The PostTask you deleted from
ActualDestroy().

We want to release the back-buffer memory so we don't have a single frame
keeping a full ST alive. It doesn't really matter at that point if there's one
image with its buffer in the back buffer; we only drop at most one frame.

We could potentially have the images call ReleaseSurfaceTexture() after the last
one calls updateTexImage(). But that's probably not worth the complication.

[liberato (no reviews please), 2017/03/06 22:59:36]

ah, really good catch!  done.

+}
+
+}  // namespace media