flypig.co.uk

This morning I started looking at why the sailfish-browser has been broken by my changes to the offscreen rendering process. The approach I'm using is to look at the intersection of:

1. Parts of the code I've changed (as shown by the diff generated by git).
2. Parts of the code that are touched by sailfish-browser (as indicated by gdb backtraces).

The first place I started looking was the CompositorOGL::CreateContext() method. This is the last method that gets called by sailfish-browser before the behaviour diverges from what I'd expect it to (as we saw yesterday). Plus I made very specific changes to this, as you can see from this portion of the diff:

$ git diff -- gfx/layers/opengl/CompositorOGL.cpp
diff --git a/gfx/layers/opengl/CompositorOGL.cpp b/gfx/layers/opengl/
    CompositorOGL.cpp
index 8a423b840dd5..11105c77c43b 100644
--- a/gfx/layers/opengl/CompositorOGL.cpp
+++ b/gfx/layers/opengl/CompositorOGL.cpp
@@ -246,12 +246,14 @@ already_AddRefed<mozilla::gl::GLContext> CompositorOGL::
    CreateContext() {
 
   // Allow to create offscreen GL context for main Layer Manager
   if (!context && gfxEnv::LayersPreferOffscreen()) {
+    SurfaceCaps caps = SurfaceCaps::ForRGB();
+    caps.preserve = false;
+    caps.bpp16 = gfxVars::OffscreenFormat() == SurfaceFormat::R5G6B5_UINT16;
+
     nsCString discardFailureId;
-    context = GLContextProvider::CreateHeadless(
-        {CreateContextFlags::REQUIRE_COMPAT_PROFILE}, &discardFailureId);
-    if (!context->CreateOffscreenDefaultFb(mSurfaceSize)) {
-      context = nullptr;
-    }
+    context = GLContextProvider::CreateOffscreen(
+        mSurfaceSize, caps, CreateContextFlags::REQUIRE_COMPAT_PROFILE,
+        &discardFailureId);
   }
[...]

I've reversed this change and performed a partial build.

Now when I transfer it over to my phone the browser crashes with a segmentation fault as soon as the page is loaded. I feel like I've been here before! Because it takes so long to transfer the libxul.so output from the partial build over to my phone I stripped it of debugging symbols before the transfer. This makes a huge difference to its size of the library (stripping out around 3 GiB of data to leave just 100 MiB of code). Unfortunately that means I can't now find out where or why the crash is occurring.

Even worse, performing another partial rebuild doesn't seem to restore the debug symbols. So the only thing for me to do is a full rebuild, which is an overnight job.

I'll do that, but in the meantime I'll continue browsing through the diff in case a reason jumps out at me, or there's something obvious that needs changing.

Things will get a bit messy the more I change, but the beauty of these diaries is that I'll be keeping a full record. So it should all be clear what gets changed and why.

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