VideoCapture: add support for GpuMemoryBuffer allocation and lifetime mgmt in VideoCaptureBufferPool

content/browser/media/capture/content_video_capture_device_core.cc

Index: content/browser/media/capture/content_video_capture_device_core.cc
diff --git a/content/browser/media/capture/content_video_capture_device_core.cc b/content/browser/media/capture/content_video_capture_device_core.cc
index 3c507c6f46b6fcbe5811067de64ac6bd0fb641fb..b208ce82010e051969774d60a741e657a55eb3b0 100644
--- a/content/browser/media/capture/content_video_capture_device_core.cc
+++ b/content/browser/media/capture/content_video_capture_device_core.cc
@@ -118,12 +118,14 @@ bool ThreadSafeCaptureOracle::ObserveEventAndDecideCapture(
   // NATIVE_TEXTURE frames wrap a texture mailbox, which we don't have at the
   // moment.  We do not construct those frames.
   if (params_.requested_format.pixel_format != media::PIXEL_FORMAT_TEXTURE) {
+    const scoped_ptr<media::DataHandle> scoped_data_handle(

[miu, 2015/04/15 19:12:40]

Why is the DataHandle allowed to be destroyed while VideoFrame is still
referencing the data pointer?  Seems like the DataHandle should remain alive
(i.e., data is not yet unmapped) until all references to the data pointer are
gone.

[mcasas, 2015/04/16 03:11:21]

Since I'm taking in your suggestion below, this
code is changed and this lifetime bug is no more.

+        output_buffer->GetDataHandle());
     *storage = media::VideoFrame::WrapExternalPackedMemory(
         media::VideoFrame::I420,
         coded_size,
         gfx::Rect(visible_size),
         visible_size,
-        static_cast<uint8*>(output_buffer->data()),
+        static_cast<uint8*>(scoped_data_handle->data()),
         output_buffer->size(),
         base::SharedMemory::NULLHandle(),
         0,