Support start trimming post-decoding. Use it with FFmpegDemuxer.

media/base/audio_buffer_unittest.cc

Index: media/base/audio_buffer_unittest.cc
diff --git a/media/base/audio_buffer_unittest.cc b/media/base/audio_buffer_unittest.cc
index c40c076bd5e8ee73c224df4db6b972f0c00fea84..09da83e5fa45973c2ab8da63f6c60b2ee31c3068 100644
--- a/media/base/audio_buffer_unittest.cc
+++ b/media/base/audio_buffer_unittest.cc
@@ -358,4 +358,49 @@ TEST(AudioBufferTest, Trim) {
   EXPECT_EQ(buffer->duration(), base::TimeDelta::FromSeconds(0));
 }
 
+TEST(AudioBufferTest, TrimRange) {
+  const ChannelLayout channel_layout = CHANNEL_LAYOUT_4_0;
+  const int channels = ChannelLayoutToChannelCount(channel_layout);
+  const int frames = 100;
+  const base::TimeDelta start_time;
+  const base::TimeDelta duration = base::TimeDelta::FromSeconds(frames);

[wolenetz, 2014/04/29 19:18:08]

Do you really intend duration to be 100 seconds, even though 100 frames at 44100
is only ~0.002268 seconds? If so, please document this maybe in the |duration|
variable name like "overestimated_duration".

[DaleCurtis, 2014/04/30 22:04:04]

Fixed here https://codereview.chromium.org/261533002/ and in new tests.

+  const float step = 1.0f;
+  scoped_refptr<AudioBuffer> buffer =
+      MakeAudioBuffer<float>(kSampleFormatPlanarF32,

[wolenetz, 2014/04/29 19:18:08]

TrimRange does fancy mem ops depending on planar vs interleaved. Please test
both.

[DaleCurtis, 2014/04/30 22:04:04]

Done.

+                             channel_layout,
+                             channels,
+                             kSampleRate,
+                             0,
+                             step,
+                             frames,
+                             start_time,
+                             duration);
+  EXPECT_EQ(frames, buffer->frame_count());
+  EXPECT_EQ(start_time, buffer->timestamp());
+  EXPECT_EQ(frames, buffer->duration().InSeconds());
+
+  scoped_ptr<AudioBus> bus = AudioBus::Create(channels, frames);
+
+  const int trim_start = 25;
+  const int trim_length = 50;
+
+  // Verify the first 25 frames before trimming.
+  buffer->ReadFrames(trim_start, 0, 0, bus.get());
+  VerifyResult(bus->channel(0), trim_start, 0, step);

[wolenetz, 2014/04/29 19:18:08]

TrimRange's fancy mem ops could use additional verification: VerifyResult of all
4 channels please.

[DaleCurtis, 2014/04/30 22:04:04]

Done.

+
+  // Trim frames from the middle of the buffer.
+  buffer->TrimRange(trim_start, trim_start + trim_length);
+  EXPECT_EQ(buffer->frame_count(), frames - trim_length);
+  EXPECT_EQ(buffer->timestamp(), start_time);
+  EXPECT_EQ(buffer->duration(), base::TimeDelta::FromSeconds(50));

[wolenetz, 2014/04/29 19:18:08]

I don't understand why an AudioBuffer could be given a duration that is
out-of-whack versus its frame count and sample rate. Is there a use case that
justifies this allowance?

[DaleCurtis, 2014/04/29 20:46:35]

Generally the duration will only be off by a microsecond or so; it was required
before AudioBuffer's had sample rates.  Now that they do, we can probably rework
this a bit.

I suspect removing the ability to set_duration() would break a lot of tests
which perform the equivalent of something like
ASSERT_EQ(timestamp_helper_duration, buffer_duration).

I'll see about a separate CL which removes set_duration().

[DaleCurtis, 2014/04/30 22:04:04]

Fixed here: https://codereview.chromium.org/261533002/

+
+  // Verify the first 25 frames.
+  buffer->ReadFrames(trim_start, 0, 0, bus.get());
+  VerifyResult(bus->channel(0), trim_start, 0, step);

[wolenetz, 2014/04/29 19:18:08]

ditto: verify all channels

[DaleCurtis, 2014/04/30 22:04:04]

Done.

+
+  // Verify the next 25 frames.
+  buffer->ReadFrames(trim_start, trim_start, 0, bus.get());
+  VerifyResult(bus->channel(0), trim_start, trim_start + trim_length, step);

[wolenetz, 2014/04/29 19:18:08]

ditto

[DaleCurtis, 2014/04/30 22:04:04]

Done.

+}
+
 }  // namespace media