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

media/base/audio_discard_helper_unittest.cc

Index: media/base/audio_discard_helper_unittest.cc
diff --git a/media/base/audio_discard_helper_unittest.cc b/media/base/audio_discard_helper_unittest.cc
index 1a6b592120da4b8425361e606a89b28cb334c1b9..8807f97c66343c5963c818e5e81f46f14a9b00b1 100644
--- a/media/base/audio_discard_helper_unittest.cc
+++ b/media/base/audio_discard_helper_unittest.cc
@@ -4,6 +4,7 @@
 
 #include "base/memory/scoped_ptr.h"
 #include "media/base/audio_buffer.h"
+#include "media/base/audio_bus.h"
 #include "media/base/audio_discard_helper.h"
 #include "media/base/buffers.h"
 #include "media/base/decoder_buffer.h"
@@ -12,6 +13,7 @@
 
 namespace media {
 
+static const float kDataStep = 0.01f;
 static const int kSampleRate = 48000;
 
 static scoped_refptr<DecoderBuffer> CreateEncodedBuffer(
@@ -29,14 +31,23 @@ static scoped_refptr<AudioBuffer> CreateDecodedBuffer(int frames) {
                          1,
                          kSampleRate,
                          0.0f,
-                         0.01f,
+                         kDataStep,
                          frames,
                          kNoTimestamp(),
                          kNoTimestamp());
 }
 
+static float ExtractDecodedData(const scoped_refptr<AudioBuffer>& buffer,
+                                int index) {
+  // This is really inefficient, but we can't access the raw AudioBuffer if any
+  // start trimming has been applied.
+  scoped_ptr<AudioBus> temp_bus = AudioBus::Create(buffer->channel_count(), 1);
+  buffer->ReadFrames(1, index, 0, temp_bus.get());
+  return temp_bus->channel(0)[0];
+}
+
 TEST(AudioDiscardHelperTest, TimeDeltaToFrames) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
 
   EXPECT_EQ(0, discard_helper.TimeDeltaToFrames(base::TimeDelta()));
   EXPECT_EQ(
@@ -53,7 +64,7 @@ TEST(AudioDiscardHelperTest, TimeDeltaToFrames) {
 }
 
 TEST(AudioDiscardHelperTest, BasicProcessBuffers) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
@@ -85,7 +96,7 @@ TEST(AudioDiscardHelperTest, BasicProcessBuffers) {
 }
 
 TEST(AudioDiscardHelperTest, ProcessBuffersWithInitialDiscard) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
@@ -108,7 +119,7 @@ TEST(AudioDiscardHelperTest, ProcessBuffersWithInitialDiscard) {
 }
 
 TEST(AudioDiscardHelperTest, ProcessBuffersWithLargeInitialDiscard) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
@@ -138,7 +149,7 @@ TEST(AudioDiscardHelperTest, ProcessBuffersWithLargeInitialDiscard) {
 }
 
 TEST(AudioDiscardHelperTest, AllowNonMonotonicTimestamps) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
@@ -163,8 +174,8 @@ TEST(AudioDiscardHelperTest, AllowNonMonotonicTimestamps) {
   EXPECT_EQ(kTestFrames, decoded_buffer->frame_count());
 }
 
-TEST(AudioDiscardHelperTest, DiscardPadding) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+TEST(AudioDiscardHelperTest, DiscardEndPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
@@ -176,7 +187,55 @@ TEST(AudioDiscardHelperTest, DiscardPadding) {
   scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
 
   // Set a discard padding equivalent to half the buffer.
-  encoded_buffer->set_discard_padding(kDuration / 2);
+  encoded_buffer->set_discard_padding(
+      std::make_pair(base::TimeDelta(), kDuration / 2));
+
+  ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  ASSERT_TRUE(discard_helper.initialized());
+  EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
+  EXPECT_EQ(kDuration / 2, decoded_buffer->duration());
+  EXPECT_EQ(kTestFrames / 2, decoded_buffer->frame_count());
+}
+
+TEST(AudioDiscardHelperTest, BadDiscardEndPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
+  ASSERT_FALSE(discard_helper.initialized());
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Set a discard padding equivalent to double the buffer size.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(base::TimeDelta(), kDuration * 2));
+
+  // Verify the end discard padding is rejected.
+  ASSERT_FALSE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  ASSERT_TRUE(discard_helper.initialized());
+}
+
+TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardEndPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
+  ASSERT_FALSE(discard_helper.initialized());
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Set a discard padding equivalent to a quarter of the buffer.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(base::TimeDelta(), kDuration / 4));
+
+  // Set an initial discard of a quarter of the buffer.
+  discard_helper.Reset(kTestFrames / 4);
 
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   ASSERT_TRUE(discard_helper.initialized());
@@ -186,7 +245,7 @@ TEST(AudioDiscardHelperTest, DiscardPadding) {
 }
 
 TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardPadding) {
-  AudioDiscardHelper discard_helper(kSampleRate);
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
   ASSERT_FALSE(discard_helper.initialized());
 
   const base::TimeDelta kTimestamp = base::TimeDelta();
@@ -198,7 +257,8 @@ TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardPadding) {
   scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
 
   // Set a discard padding equivalent to a quarter of the buffer.
-  encoded_buffer->set_discard_padding(kDuration / 4);
+  encoded_buffer->set_discard_padding(
+      std::make_pair(kDuration / 4, kDuration / 4));
 
   // Set an initial discard of a quarter of the buffer.
   discard_helper.Reset(kTestFrames / 4);
@@ -206,9 +266,82 @@ TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardPadding) {
   ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
   ASSERT_TRUE(discard_helper.initialized());
   EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
+  EXPECT_EQ(kDuration / 4, decoded_buffer->duration());
+  EXPECT_EQ(kTestFrames / 4, decoded_buffer->frame_count());
+}
+
+TEST(AudioDiscardHelperTest, InitialDiscardAndDiscardPaddingAndCodecDelay) {
+  // Use a codec delay of 5ms.
+  const int kCodecDelay = kSampleRate / 100 / 2;
+  AudioDiscardHelper discard_helper(kSampleRate, kCodecDelay);
+  ASSERT_FALSE(discard_helper.initialized());
+  discard_helper.Reset(kCodecDelay);
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Set a discard padding equivalent to half of the buffer.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(kDuration / 2, base::TimeDelta()));
+
+  // All of the first buffer should be discarded.
+  ASSERT_FALSE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  ASSERT_TRUE(discard_helper.initialized());
+
+  // Processing another buffer (with the same discard padding) should discard
+  // the back half of the buffer since kCodecDelay is half a buffer.
+  encoded_buffer->set_timestamp(kTimestamp + kDuration);
+  decoded_buffer = CreateDecodedBuffer(kTestFrames);
+  ASSERT_FLOAT_EQ(0.0f, ExtractDecodedData(decoded_buffer, 0));
+  ASSERT_NEAR(kCodecDelay * kDataStep,
+              ExtractDecodedData(decoded_buffer, kCodecDelay),
+              kDataStep * 1000);
+  ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
   EXPECT_EQ(kDuration / 2, decoded_buffer->duration());
   EXPECT_EQ(kTestFrames / 2, decoded_buffer->frame_count());
+
+  // Verify it was actually the latter half of the buffer that was removed.
+  ASSERT_FLOAT_EQ(0.0f, ExtractDecodedData(decoded_buffer, 0));
 }
 
+TEST(AudioDiscardHelperTest, DelayedDiscardInitialDiscardAndDiscardPadding) {
+  AudioDiscardHelper discard_helper(kSampleRate, 0);
+  ASSERT_FALSE(discard_helper.initialized());
+
+  const base::TimeDelta kTimestamp = base::TimeDelta();
+  const base::TimeDelta kDuration = base::TimeDelta::FromMilliseconds(10);
+  const int kTestFrames = discard_helper.TimeDeltaToFrames(kDuration);
+
+  scoped_refptr<DecoderBuffer> encoded_buffer =
+      CreateEncodedBuffer(kTimestamp, kDuration);
+
+  // Set a discard padding equivalent to a quarter of the buffer.
+  encoded_buffer->set_discard_padding(
+      std::make_pair(kDuration / 4, kDuration / 4));
+
+  // Set an initial discard of a quarter of the buffer.
+  discard_helper.Reset(kTestFrames / 4);
+
+  // Verify nothing is output for the first buffer, yet initialized is true.
+  ASSERT_FALSE(discard_helper.ProcessBuffers(encoded_buffer, NULL));
+  ASSERT_TRUE(discard_helper.initialized());
+
+  // Create an encoded buffer with no discard padding.
+  encoded_buffer = CreateEncodedBuffer(kTimestamp + kDuration, kDuration);
+  scoped_refptr<AudioBuffer> decoded_buffer = CreateDecodedBuffer(kTestFrames);
+
+  // Verify that when the decoded buffer is consumed, the discards from the
+  // previous encoded buffer are applied.
+  ASSERT_TRUE(discard_helper.ProcessBuffers(encoded_buffer, decoded_buffer));
+  EXPECT_EQ(kTimestamp, decoded_buffer->timestamp());
+  EXPECT_EQ(kDuration / 4, decoded_buffer->duration());
+  EXPECT_EQ(kTestFrames / 4, decoded_buffer->frame_count());
+}
 
 }  // namespace media