shaka-packager/packager/media/chunking/cue_alignment_handler_unitt...

625 lines
22 KiB
C++

// Copyright 2018 Google Inc. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file or at
// https://developers.google.com/open-source/licenses/bsd
#include "packager/media/chunking/chunking_handler.h"
#include <gmock/gmock.h>
#include <gtest/gtest.h>
#include "packager/media/base/media_handler_test_base.h"
#include "packager/media/chunking/cue_alignment_handler.h"
#include "packager/media/public/ad_cue_generator_params.h"
#include "packager/status_macros.h"
#include "packager/status_test_util.h"
using ::testing::_;
using ::testing::ElementsAre;
using ::testing::IsEmpty;
namespace shaka {
namespace media {
namespace {
const size_t kOneInput = 1;
const size_t kOneOutput = 1;
const size_t kThreeInputs = 3;
const size_t kThreeOutputs = 3;
const bool kKeyFrame = true;
const uint32_t kMsTimeScale = 1000;
const size_t kStreamIndex = 0;
} // namespace
class CueAlignmentHandlerTest : public MediaHandlerTestBase {
protected:
std::unique_ptr<SyncPointQueue> CreateSyncPoints(
std::initializer_list<double> cues) {
AdCueGeneratorParams params;
for (double cue_time : cues) {
Cuepoint cue;
cue.start_time_in_seconds = cue_time;
params.cue_points.push_back(cue);
}
return std::unique_ptr<SyncPointQueue>(new SyncPointQueue(params));
}
Status DispatchAudioInfo(size_t input_index) {
auto info = GetAudioStreamInfo(kMsTimeScale);
auto data = StreamData::FromStreamInfo(kStreamIndex, std::move(info));
return Input(input_index)->Dispatch(std::move(data));
}
Status DispatchTextInfo(size_t input_index) {
auto info = GetTextStreamInfo(kMsTimeScale);
auto data = StreamData::FromStreamInfo(kStreamIndex, std::move(info));
return Input(input_index)->Dispatch(std::move(data));
}
Status DispatchVideoInfo(size_t input_index) {
auto info = GetVideoStreamInfo(kMsTimeScale);
auto data = StreamData::FromStreamInfo(kStreamIndex, std::move(info));
return Input(input_index)->Dispatch(std::move(data));
}
Status DispatchMediaSample(size_t input_index,
int64_t start_time,
int64_t duration,
bool keyframe) {
auto sample = GetMediaSample(start_time, duration, keyframe);
auto data = StreamData::FromMediaSample(kStreamIndex, std::move(sample));
return Input(input_index)->Dispatch(std::move(data));
}
Status DispatchTextSample(size_t input_index,
int64_t start_time,
int64_t end_time) {
const char* kNoId = "";
const char* kNoPayload = "";
auto sample = GetTextSample(kNoId, start_time, end_time, kNoPayload);
auto data = StreamData::FromTextSample(kStreamIndex, std::move(sample));
return Input(input_index)->Dispatch(std::move(data));
}
Status FlushAll(std::initializer_list<size_t> inputs) {
for (auto& input : inputs) {
RETURN_IF_ERROR(Input(input)->FlushAllDownstreams());
}
return Status::OK;
}
};
TEST_F(CueAlignmentHandlerTest, VideoInputWithNoCues) {
const size_t kVideoStream = 0;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample1Start = kSample0Start + kSampleDuration;
const int64_t kSample2Start = kSample1Start + kSampleDuration;
auto sync_points = CreateSyncPoints({});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kOneInput, kOneOutput));
{
testing::InSequence s;
EXPECT_CALL(*Output(kVideoStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kVideoStream), OnFlush(_));
}
ASSERT_OK(DispatchVideoInfo(kVideoStream));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample0Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample1Start, kSampleDuration,
!kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(FlushAll({kVideoStream}));
}
TEST_F(CueAlignmentHandlerTest, AudioInputWithNoCues) {
const size_t kAudioStream = 0;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample1Start = kSample0Start + kSampleDuration;
const int64_t kSample2Start = kSample1Start + kSampleDuration;
auto sync_points = CreateSyncPoints({});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kOneInput, kOneOutput));
{
testing::InSequence s;
EXPECT_CALL(*Output(kAudioStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kAudioStream), OnFlush(_));
}
ASSERT_OK(DispatchAudioInfo(kAudioStream));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample0Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample1Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(FlushAll({kAudioStream}));
}
TEST_F(CueAlignmentHandlerTest, TextInputWithNoCues) {
const size_t kTextStream = 0;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample0End = kSample0Start + kSampleDuration;
const int64_t kSample1Start = kSample0End;
const int64_t kSample1End = kSample1Start + kSampleDuration;
const int64_t kSample2Start = kSample1End;
const int64_t kSample2End = kSample2Start + kSampleDuration;
auto sync_points = CreateSyncPoints({});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kOneInput, kOneOutput));
{
testing::InSequence s;
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample0Start, kSample0End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample1Start, kSample1End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample2Start, kSample2End, _, _)));
EXPECT_CALL(*Output(kTextStream), OnFlush(_));
}
ASSERT_OK(DispatchTextInfo(kTextStream));
ASSERT_OK(DispatchTextSample(kTextStream, kSample0Start, kSample0End));
ASSERT_OK(DispatchTextSample(kTextStream, kSample1Start, kSample1End));
ASSERT_OK(DispatchTextSample(kTextStream, kSample2Start, kSample2End));
ASSERT_OK(FlushAll({kTextStream}));
}
TEST_F(CueAlignmentHandlerTest, TextAudioVideoInputWithNoCues) {
const size_t kTextStream = 0;
const size_t kAudioStream = 1;
const size_t kVideoStream = 2;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample0End = kSample0Start + kSampleDuration;
const int64_t kSample1Start = kSample0Start + kSampleDuration;
const int64_t kSample1End = kSample1Start + kSampleDuration;
const int64_t kSample2Start = kSample1Start + kSampleDuration;
const int64_t kSample2End = kSample2Start + kSampleDuration;
auto sync_points = CreateSyncPoints({});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kThreeInputs, kThreeOutputs));
{
testing::InSequence s;
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample0Start, kSample0End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample1Start, kSample1End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample2Start, kSample2End, _, _)));
EXPECT_CALL(*Output(kTextStream), OnFlush(_));
}
{
testing::InSequence s;
EXPECT_CALL(*Output(kAudioStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kAudioStream), OnFlush(_));
}
{
testing::InSequence s;
EXPECT_CALL(*Output(kVideoStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kVideoStream), OnFlush(_));
}
// Dispatch Stream Info
ASSERT_OK(DispatchTextInfo(kTextStream));
ASSERT_OK(DispatchAudioInfo(kAudioStream));
ASSERT_OK(DispatchVideoInfo(kVideoStream));
// Dispatch Sample 0
ASSERT_OK(DispatchTextSample(kTextStream, kSample0Start, kSample0End));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample0Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample0Start, kSampleDuration,
kKeyFrame));
// Dispatch Sample 1
ASSERT_OK(DispatchTextSample(kTextStream, kSample1Start, kSample1End));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample1Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample1Start, kSampleDuration,
!kKeyFrame));
// Dispatch Sample 2
ASSERT_OK(DispatchTextSample(kTextStream, kSample2Start, kSample2End));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(FlushAll({kTextStream, kAudioStream, kVideoStream}));
}
TEST_F(CueAlignmentHandlerTest, VideoInputWithCues) {
const size_t kVideoStream = 0;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample1Start = kSample0Start + kSampleDuration;
const int64_t kSample2Start = kSample1Start + kSampleDuration;
const double kSample2StartInSeconds =
static_cast<double>(kSample2Start) / kMsTimeScale;
// Put the cue between two key frames.
auto sync_points =
CreateSyncPoints({static_cast<double>(kSample1Start) / kMsTimeScale});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kOneInput, kOneOutput));
{
testing::InSequence s;
EXPECT_CALL(*Output(kVideoStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kVideoStream),
OnProcess(IsCueEvent(_, kSample2StartInSeconds)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kVideoStream), OnFlush(_));
}
ASSERT_OK(DispatchVideoInfo(kVideoStream));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample0Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample1Start, kSampleDuration,
!kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(FlushAll({kVideoStream}));
}
TEST_F(CueAlignmentHandlerTest, AudioInputWithCues) {
const size_t kAudioStream = 0;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample1Start = kSample0Start + kSampleDuration;
const int64_t kSample2Start = kSample1Start + kSampleDuration;
const double kSample1StartInSeconds =
static_cast<double>(kSample1Start) / kMsTimeScale;
auto sync_points =
CreateSyncPoints({static_cast<double>(kSample1Start) / kMsTimeScale});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kOneInput, kOneOutput));
{
testing::InSequence s;
EXPECT_CALL(*Output(kAudioStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kAudioStream),
OnProcess(IsCueEvent(_, kSample1StartInSeconds)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kAudioStream), OnFlush(_));
}
ASSERT_OK(DispatchAudioInfo(kAudioStream));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample0Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample1Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(FlushAll({kAudioStream}));
}
TEST_F(CueAlignmentHandlerTest, TextInputWithCues) {
const size_t kTextStream = 0;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample0End = kSample0Start + kSampleDuration;
const int64_t kSample1Start = kSample0End;
const int64_t kSample1End = kSample1Start + kSampleDuration;
const int64_t kSample2Start = kSample1End;
const int64_t kSample2End = kSample2Start + kSampleDuration;
const double kSample1StartInSeconds =
static_cast<double>(kSample1Start) / kMsTimeScale;
auto sync_points =
CreateSyncPoints({static_cast<double>(kSample1Start) / kMsTimeScale});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kOneInput, kOneOutput));
{
testing::InSequence s;
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample0Start, kSample0End, _, _)));
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsCueEvent(_, kSample1StartInSeconds)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample1Start, kSample1End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample2Start, kSample2End, _, _)));
EXPECT_CALL(*Output(kTextStream), OnFlush(_));
}
ASSERT_OK(DispatchTextInfo(kTextStream));
ASSERT_OK(DispatchTextSample(kTextStream, kSample0Start, kSample0End));
ASSERT_OK(DispatchTextSample(kTextStream, kSample1Start, kSample1End));
ASSERT_OK(DispatchTextSample(kTextStream, kSample2Start, kSample2End));
ASSERT_OK(FlushAll({kTextStream}));
}
TEST_F(CueAlignmentHandlerTest, TextInputWithCueAfterLastStart) {
const size_t kTextStream = 0;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample0End = kSample0Start + kSampleDuration;
const int64_t kSample1Start = kSample0End;
const int64_t kSample1End = kSample1Start + kSampleDuration;
const int64_t kSample2Start = kSample1End;
const int64_t kSample2End = kSample2Start + kSampleDuration;
// Put the cue between the start and end of the last sample.
const double kCueTimeInSeconds =
static_cast<double>(kSample2Start + kSample2End) / kMsTimeScale;
auto sync_points = CreateSyncPoints({kCueTimeInSeconds});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kOneInput, kOneOutput));
{
testing::InSequence s;
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample0Start, kSample0End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample1Start, kSample1End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample2Start, kSample2End, _, _)));
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsCueEvent(_, kCueTimeInSeconds)));
EXPECT_CALL(*Output(kTextStream), OnFlush(_));
}
ASSERT_OK(DispatchTextInfo(kTextStream));
ASSERT_OK(DispatchTextSample(kTextStream, kSample0Start, kSample0End));
ASSERT_OK(DispatchTextSample(kTextStream, kSample1Start, kSample1End));
ASSERT_OK(DispatchTextSample(kTextStream, kSample2Start, kSample2End));
ASSERT_OK(FlushAll({kTextStream}));
}
TEST_F(CueAlignmentHandlerTest, TextAudioVideoInputWithCues) {
const size_t kTextStream = 0;
const size_t kAudioStream = 1;
const size_t kVideoStream = 2;
const int64_t kSampleDuration = 1000;
const int64_t kSample0Start = 0;
const int64_t kSample0End = kSample0Start + kSampleDuration;
const int64_t kSample1Start = kSample0End;
const int64_t kSample1End = kSample1Start + kSampleDuration;
const int64_t kSample2Start = kSample1End;
const int64_t kSample2End = kSample2Start + kSampleDuration;
const double kSample2StartInSeconds =
static_cast<double>(kSample2Start) / kMsTimeScale;
// Put the cue between two key frames.
auto sync_points =
CreateSyncPoints({static_cast<double>(kSample1Start) / kMsTimeScale});
auto handler = std::make_shared<CueAlignmentHandler>(sync_points.get());
ASSERT_OK(SetUpAndInitializeGraph(handler, kThreeInputs, kThreeOutputs));
{
testing::InSequence s;
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample0Start, kSample0End, _, _)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample1Start, kSample1End, _, _)));
EXPECT_CALL(*Output(kTextStream),
OnProcess(IsCueEvent(_, kSample2StartInSeconds)));
EXPECT_CALL(
*Output(kTextStream),
OnProcess(IsTextSample(_, _, kSample2Start, kSample2End, _, _)));
EXPECT_CALL(*Output(kTextStream), OnFlush(_));
}
{
testing::InSequence s;
EXPECT_CALL(*Output(kAudioStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kAudioStream),
OnProcess(IsCueEvent(_, kSample2StartInSeconds)));
EXPECT_CALL(
*Output(kAudioStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kAudioStream), OnFlush(_));
}
{
testing::InSequence s;
EXPECT_CALL(*Output(kVideoStream),
OnProcess(IsStreamInfo(_, kMsTimeScale, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample0Start, kSampleDuration, _, _)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample1Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kVideoStream),
OnProcess(IsCueEvent(_, kSample2StartInSeconds)));
EXPECT_CALL(
*Output(kVideoStream),
OnProcess(IsMediaSample(_, kSample2Start, kSampleDuration, _, _)));
EXPECT_CALL(*Output(kVideoStream), OnFlush(_));
}
// Dispatch Stream Info
ASSERT_OK(DispatchTextInfo(kTextStream));
ASSERT_OK(DispatchAudioInfo(kAudioStream));
ASSERT_OK(DispatchVideoInfo(kVideoStream));
// Dispatch Sample 0
ASSERT_OK(DispatchTextSample(kTextStream, kSample0Start, kSample0End));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample0Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample0Start, kSampleDuration,
kKeyFrame));
// Dispatch Sample 1
ASSERT_OK(DispatchTextSample(kTextStream, kSample1Start, kSample1End));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample1Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample1Start, kSampleDuration,
!kKeyFrame));
// Dispatch Sample 2
ASSERT_OK(DispatchTextSample(kTextStream, kSample2Start, kSample2End));
ASSERT_OK(DispatchMediaSample(kAudioStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(DispatchMediaSample(kVideoStream, kSample2Start, kSampleDuration,
kKeyFrame));
ASSERT_OK(FlushAll({kTextStream, kAudioStream, kVideoStream}));
}
// TODO(kqyang): Add more tests, in particular, multi-thread tests.
} // namespace media
} // namespace shaka