media: add basic MediaRenderer mojom and TypeConverters

media/mojo/services/media_type_converters.cc

Index: media/mojo/services/media_type_converters.cc
diff --git a/media/mojo/services/media_type_converters.cc b/media/mojo/services/media_type_converters.cc
new file mode 100644
index 0000000000000000000000000000000000000000..0841f1308efb8f078d3b05a807d70c526aa1ebda
--- /dev/null
+++ b/media/mojo/services/media_type_converters.cc
@@ -0,0 +1,100 @@
+// Copyright 2014 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/mojo/services/media_type_converters.h"
+
+#include "base/macros.h"
+#include "media/base/buffering_state.h"
+#include "media/base/decoder_buffer.h"
+#include "mojo/public/cpp/system/data_pipe.h"
+
+namespace mojo {
+
+#define ASSERT_ENUM_VALUES_EQUAL(value)                              \
+  COMPILE_ASSERT(media::BUFFERING_##value ==                         \
+      static_cast<media::BufferingState>(BUFFERING_STATE_##value),   \
+          value##_enum_value_matches)
+
+ASSERT_ENUM_VALUES_EQUAL(HAVE_NOTHING);
+ASSERT_ENUM_VALUES_EQUAL(HAVE_ENOUGH);
+
+MediaDecoderBufferPtr TypeConverter<MediaDecoderBufferPtr,
+    scoped_refptr<media::DecoderBuffer> >::ConvertFrom(
+        const scoped_refptr<media::DecoderBuffer> input) {
+  MediaDecoderBufferPtr mojo_buffer(MediaDecoderBuffer::New());
+  mojo_buffer->timestamp = input->timestamp().ToInternalValue();

[xhwang, 2014/09/03 00:09:08]

Here and below, s/ToInternalValue/InMicroseconds/ to avoid relying on TimeDelta
implementation detail?

[tim (not reviewing), 2014/09/03 00:43:50]

Ah, good call.  Done.

+  mojo_buffer->duration = input->duration().ToInternalValue();
+  mojo_buffer->data_size = input->data_size();
+  mojo_buffer->side_data_size = input->side_data_size();
+  mojo_buffer->front_discard_usec =
+      input->discard_padding().first.ToInternalValue();
+  mojo_buffer->back_discard_usec =
+      input->discard_padding().second.ToInternalValue();
+  mojo_buffer->splice_timestamp = input->splice_timestamp().ToInternalValue();
+
+  // TODO(tim): Assuming this is small so allowing extra copies.
+  std::vector<uint8> side_data(input->side_data(),
+                               input->side_data() + input->side_data_size());
+  mojo_buffer->side_data.Swap(&side_data);
+
+  MojoCreateDataPipeOptions options;
+  options.struct_size = sizeof(MojoCreateDataPipeOptions);
+  options.flags = MOJO_CREATE_DATA_PIPE_OPTIONS_FLAG_NONE;
+  options.element_num_bytes = 1;
+  options.capacity_num_bytes = input->data_size();
+  DataPipe data_pipe(options);
+  mojo_buffer->data = data_pipe.consumer_handle.Pass();
+
+  uint32_t num_bytes = input->data_size();
+  // TODO(tim): ALL_OR_NONE isn't really appropriate. Check success?
+  // If fails, we'd still return the buffer, but we'd need to HandleWatch
+  // to fill the pipe at a later time, which means the de-marshalling code
+  // needs to wait for a readable pipe (which it currently doesn't).
+  WriteDataRaw(data_pipe.producer_handle.get(),
+               input->data(),
+               &num_bytes,
+               MOJO_WRITE_DATA_FLAG_ALL_OR_NONE);
+  return mojo_buffer.Pass();
+}
+
+scoped_refptr<media::DecoderBuffer>  TypeConverter<MediaDecoderBufferPtr,
+    scoped_refptr<media::DecoderBuffer> >::ConvertTo(
+        const MediaDecoderBufferPtr& input) {
+  uint32_t num_bytes  = 0;
+  // TODO(tim): We're assuming that because we always write to the pipe above
+  // before sending the MediaDecoderBuffer that the pipe is readable when
+  // we get here.
+  ReadDataRaw(input->data.get(), NULL, &num_bytes, MOJO_READ_DATA_FLAG_QUERY);
+  CHECK_EQ(num_bytes, input->data_size) << "Pipe error converting buffer";
+
+  scoped_ptr<uint8[]> data(new uint8[num_bytes]);  // Uninitialized.
+  ReadDataRaw(input->data.get(), data.get(), &num_bytes,
+              MOJO_READ_DATA_FLAG_ALL_OR_NONE);
+  CHECK_EQ(num_bytes, input->data_size) << "Pipe error converting buffer";
+
+  // TODO(tim): We can't create a media::DecoderBuffer that has side_data
+  // without copying data because it wants to ensure alignment. Could we
+  // read directly into a pre-padded DecoderBuffer?
+  scoped_refptr<media::DecoderBuffer> buffer;
+  if (input->side_data_size) {
+    buffer = media::DecoderBuffer::CopyFrom(data.get(),
+                                            num_bytes,
+                                            input->side_data.storage().data(),
+                                            input->side_data_size);
+  } else {
+    buffer = media::DecoderBuffer::CopyFrom(data.get(), num_bytes);
+  }
+
+  buffer->set_timestamp(base::TimeDelta::FromInternalValue(input->timestamp));

[xhwang, 2014/09/03 00:09:08]

ditto: FromMicroseconds()

[tim (not reviewing), 2014/09/03 00:43:50]

Done.

+  buffer->set_duration(base::TimeDelta::FromInternalValue(input->duration));
+  media::DecoderBuffer::DiscardPadding discard_padding(
+      base::TimeDelta::FromInternalValue(input->front_discard_usec),
+      base::TimeDelta::FromInternalValue(input->back_discard_usec));
+  buffer->set_discard_padding(discard_padding);
+  buffer->set_splice_timestamp(
+      base::TimeDelta::FromInternalValue(input->splice_timestamp));
+  return buffer;
+}
+
+}  // namespace mojo