flypig.co.uk

I've been trying to get the swap chain to initialise correctly over the last few days. This is part of the code that I made large changes to early on in this process, before the build would fully compile. I'm now having to simultaneously unravel the changes I made, while at the same time finally figuring out what they're supposed to be doing. It's quite a relief to finally get the chance to fix the mistakes I made in the past.

But the task right now is a little more prosaic. I'm just trying to get the thing to run without crashing. Getting the actual rendering working will be stage two of this process.

So I'm still trying to get the back buffer to be initialised before it's accessed. Sounds simple, but the code is a bit of web. We have a call to Resize() which is crashing and a call to PrepareOffscreen() which creates the swap chain. We need to create the swap chain and initialise the back buffer before the Resize() happens.

If we follow the backtraces back, the ordering problem seems to end up here:

PLayerTransactionParent*
EmbedLiteCompositorBridgeParent::AllocPLayerTransactionParent
    (const nsTArray<LayersBackend>& aBackendHints, const LayersId& aId)
{
  PLayerTransactionParent* p =
    CompositorBridgeParent::AllocPLayerTransactionParent(aBackendHints, aId);

  EmbedLiteWindowParent *parentWindow = EmbedLiteWindowParent::From(mWindowId);
  if (parentWindow) {
    parentWindow->GetListener()->CompositorCreated();
  }

  if (!StaticPrefs::embedlite_compositor_external_gl_context()) {
    // Prepare Offscreen rendering context
    PrepareOffscreen();
  }
  return p;
}

That's because the call stack for CreateContext(), which is where the SwapChain::Resize() gets called, includes CompositorBridgeParent::AllocPLayerTransactionParent(), whereas the SwapChain object is created in PrepareOffscreen(). As we can see, these happen in the wrong order.

One thing that seems worth trying is configuring the back buffer immediately after creating the swap chain. So I've given it a go by adding the call to ResizeScreenBuffer() in directly after the SwapChain constructor is called, like this:

    SwapChain* swapChain = context->GetSwapChain();
    if (swapChain == nullptr) {
      swapChain = new SwapChain();
      new SwapChainPresenter(*swapChain);
      context->mSwapChain.reset(swapChain);

      bool success = context->ResizeScreenBuffer(mEGLSurfaceSize);
      if (!success) {
          NS_WARNING("Failed to create SwapChain back buffer");
      }
    }

When I execute the updated code, this call to resize the screen buffer now triggers a crash.

$ gdb harbour-webview
GNU gdb (GDB) Mer (8.2.1+git9)
[...]
=============== Preparing offscreen rendering context ===============

Thread 36 "Compositor" received signal SIGSEGV, Segmentation fault.
[Switching to LWP 16991]
mozilla::gl::SwapChain::Resize (this=0x7ed81ce090, size=...)
    at gfx/gl/GLScreenBuffer.cpp:134
134           mFactory->CreateShared(size);
(gdb) bt
#0  mozilla::gl::SwapChain::Resize (this=0x7ed81ce090, size=...)
    at gfx/gl/GLScreenBuffer.cpp:134
#1  0x0000007ff110dc14 in mozilla::gl::GLContext::ResizeScreenBuffer
    (this=this@entry=0x7ed819ee40, size=...)
    at ${PROJECT}/obj-build-mer-qt-xr/dist/include/mozilla/UniquePtr.h:290
#2  0x0000007ff366824c in mozilla::embedlite::EmbedLiteCompositorBridgeParent::
    PrepareOffscreen (this=this@entry=0x7fc4bef570)
    at mobile/sailfishos/embedthread/EmbedLiteCompositorBridgeParent.cpp:132
#3  0x0000007ff36682b8 in mozilla::embedlite::EmbedLiteCompositorBridgeParent::
    AllocPLayerTransactionParent (this=0x7fc4bef570, aBackendHints=..., aId=...)
    at mobile/sailfishos/embedthread/EmbedLiteCompositorBridgeParent.cpp:90
#4  0x0000007ff0c65ad0 in mozilla::layers::PCompositorBridgeParent::
    OnMessageReceived (this=0x7fc4bef570, msg__=...)
    at PCompositorBridgeParent.cpp:1285
[...]
#19 0x0000007ff6a0489c in ?? () from /lib64/libc.so.6
(gdb) p size.width
$2 = 1080
(gdb) p size.height
$3 = 2520
(gdb) p mFactory.mTuple.mFirstA
$5 = (mozilla::gl::SurfaceFactory *) 0x0
(gdb) 

As we can see from the above value of mFactory.mTuple.mFirstA and the code below, the reason for the crash is that the SurfaceFactory needed to generate the surface hasn't yet been initialised. As before, it's all about the sequencing.

bool SwapChain::Resize(const gfx::IntSize& size) {
  UniquePtr<SharedSurface> newBack =
      mFactory->CreateShared(size);
  if (!newBack) return false;

  if (mPresenter->mBackBuffer) mPresenter->mBackBuffer->ProducerRelease();

  mPresenter->mBackBuffer.reset(newBack.release());

  mPresenter->mBackBuffer->ProducerAcquire();

  return true;
}

It turns out, the factory is created before, but isn't set until afterwards:

  if (context->IsOffscreen()) {
    UniquePtr<SurfaceFactory> factory;
    if (context->GetContextType() == GLContextType::EGL) {
      // [Basic/OGL Layers, OMTC] WebGL layer init.
      factory = SurfaceFactory_EGLImage::Create(*context);
    } else {
      // [Basic Layers, OMTC] WebGL layer init.
      // Well, this *should* work...
      factory = MakeUnique<SurfaceFactory_Basic>(*context);
    }

    SwapChain* swapChain = context->GetSwapChain();
    if (swapChain == nullptr) {
      swapChain = new SwapChain();
      new SwapChainPresenter(*swapChain);
      context->mSwapChain.reset(swapChain);
    }

    if (factory) {
      swapChain->Morph(std::move(factory));
    }
  }

So I've rejigged things. Crucially though, although the factory should be reset independently of whether we're creating a new swap chain or not, we don't want the resize to happen except when it's a new swap chain. I've therefore had to create a new initSwapChain Boolean to capture whether this is a new swap chain or not. If it is, we can then perform the resize after the factory code has executed.

  bool initSwapChain = false;
  // TODO: The switch from GLSCreenBuffer to SwapChain needs completing
  // See: https://phabricator.services.mozilla.com/D75055
  if (context->IsOffscreen()) {
[...]
    SwapChain* swapChain = context->GetSwapChain();
    if (swapChain == nullptr) {
      initSwapChain = true;
      swapChain = new SwapChain();
      new SwapChainPresenter(*swapChain);
      context->mSwapChain.reset(swapChain);
    }

    if (factory) {
      swapChain->Morph(std::move(factory));
    }

    if (initSwapChain) {
      bool success = context->ResizeScreenBuffer(mEGLSurfaceSize);
      if (!success) {
          NS_WARNING("Failed to create SwapChain back buffer");
      }
    }
  }

This seems worth a try, so I've set the build off running again and we'll see how it pans out when it's done.

As always the build is taking a very long, so we'll have to wait until the morning to find out how this has gone.

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