flypig.co.uk

I'm back on track today, returning to the question of why the WebView offscreen renderer isn't rendering. At least, not on screen. Before I got sidetracked my plan was to look at what's happening between the completion of rendering the page to the texture (which is working) and rendering of that texture to the screen (which is not).

Previously my plan was to look between DeclarativeWebContainer::clearWindowSurface() and CompositorOGL::EndFrame() and try to find the problem somewhere between the two. The clearWindowSurface() even had a handy glClear() call which I could use to change the background colour of the unrendered page.

But I subsequently discovered that clearWindowSurface() is used by the browser, but not the WebView. And that led me on my path to having to fix the browser.

It's good I went down that path, because now the browser is working again, but it was nevertheless a diversion from my path of fixing the WebView. It's good to be able to pop that topic off the stack and return to this.

So the question I'm interested to know is whether there's an equivalent of clearWindowSurface() for the WebView. But I have a suspicion that this is all performed behind the scenes directly by Qt. If so, that's going to complicate things.

So I've spent this evening digging through the code, staring at the sailfish-components-webview code and RawWebView in particular, moving rapidly on to qtmozembed where I looked at QOpenGLWebPage (not used), QuickMozView (used), QMozViewPrivate (where the actual action happens) and then ending up looking back in the Gecko code again.

And the really interesting bit seems to be EmbedLiteCompositorBridgeParent. For example, there's this PresentOffscreenSurface() method which is called frequently and includes a call to screen->PublishFrame():

void
EmbedLiteCompositorBridgeParent::PresentOffscreenSurface()
{
  const CompositorBridgeParent::LayerTreeState* state = CompositorBridgeParent::
    GetIndirectShadowTree(RootLayerTreeId());
  NS_ENSURE_TRUE(state && state->mLayerManager, );

  GLContext* context = static_cast<CompositorOGL*>(
    state->mLayerManager->GetCompositor())->gl();
  NS_ENSURE_TRUE(context, );
  NS_ENSURE_TRUE(context->IsOffscreen(), );

  // RenderGL is called always from Gecko compositor thread.
  // GLScreenBuffer::PublishFrame does swap buffers and that
  // cannot happen while reading previous frame on 
    EmbedLiteCompositorBridgeParent::GetPlatformImage
  // (potentially from another thread).
  MutexAutoLock lock(mRenderMutex);

  // TODO: The switch from GLSCreenBuffer to SwapChain needs completing
  // See: https://phabricator.services.mozilla.com/D75055
  GLScreenBuffer* screen = context->Screen();
  MOZ_ASSERT(screen);

  if (screen->Size().IsEmpty() || !screen->PublishFrame(screen->Size())) {
    NS_ERROR(&quot;Failed to publish context frame&quot;);
  }
}

The PublishFrame() method ends up just calling the Swap() method:

  bool PublishFrame(const gfx::IntSize& size) { return Swap(size); }

But the Swap() method is more interesting: it deals with the surfaces and is part of the GLScreenBuffer class. So this looks like a good place to investigate further.

It's late already though, so that investigation will have to wait until the morning.

If you'd like to read any of my other gecko diary entries, they're all available on my Gecko-dev Diary page.