flypig.co.uk

We took a bit of an interlude from my usual plan yesterday to consider some of the many useful suggestions that I've received over the last couple of weeks. I generated quite a few log files but didn't find any obvious discrepancies in them. That's not to say we've seen the end of those ideas and I'm still interested to know if anyone else can spot anything that looks worth following up. In the meantime I'm dropping back into my usual cadence with a plan to test out changes to the source code.

The problem I'm trying to solve is the seizing up of the app. I first noticed this after fixing a bug in the SharedSurface_EGLImage::Create() method. Over the next couple of days I'm planning to work through this method disabling various parts of it to try to pin down exactly which parts are causing the problem.

The start of the method looks like this:

/*static*/
UniquePtr<SurfaceFactory_EGLImage> SurfaceFactory_EGLImage::Create(
    GLContext* prodGL, const SurfaceCaps& caps,
    const RefPtr<layers::LayersIPCChannel>& allocator,
    const layers::TextureFlags& flags) {
  const auto& gle = GLContextEGL::Cast(prodGL);
  //const auto& egl = gle->mEgl;
  const auto& context = gle->mContext;

  typedef SurfaceFactory_EGLImage ptrT;
  UniquePtr<ptrT> ret;

  if (HasEglImageExtensions(*gle)) {
    ret.reset(new ptrT({prodGL, SharedSurfaceType::Basic, layers::TextureType::
    Unknown, true}, caps, allocator, flags, context));
  }

  return ret;
}

The original problem was that the HasExtensions() condition was causing the method to be exited early. In fact, really before the method had done anything at all. So I'm forcing an early return to simulate the same behaviour; like this:

[...]
  typedef SurfaceFactory_EGLImage ptrT;
  UniquePtr<ptrT> ret;

  return ret;
[...]

I've rebuilt it using the standard partial-build process, copied the resulting libxul.so over to my phone and installed it.

The rendering was already broken and this change will only break it further, but the question I want to know the answer to is: "will this stop the app from seizing up".

The answer is: "yes". This one small change means I can now leave the app running for long periods, swiping between the lipstick home screen and the app, swiping up and down on the page (which no obvious effect, but the touch input is still going through). The app remains responsive, my phone's watchdog doesn't bite and the OS doesn't reboot.

I'm glad about this. If it hadn't been the case it would have meant I'd misunderstood where the problem was coming from. Now reassured I can continue on to partition the code up and try to narrow down the error.

But that will be for tomorrow. For today, it's good to be armed with this backstop of knowledge about where the error is emanating from.

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