Signals, Cairo, Events + General Cleanup

[mjakeman]

This restores our old support for signals now that delegates work as expected.

This is technically done, but needs some further testing.

It also includes:

• Basic implementation of Cairo
• Support for GdkEvent
• A complex sample demonstrating custom drawing code
• Fixes to union and bitfield generation

---

TODO (before merge):

• Value.Extract must work for arbitrary records
• Fix bitfield size (long vs uint vs int?)
• Fix issue where DrawingArea occasionally stops drawing on low memory (GC?)
• Handle unions (e.g. GdkEvent / GdkEventKey)
  • Special case Gdk.Event / Gst.MiniObject / etc (MiniObject is for another PR)
  • (funnily enough, Gst.MiniObject is the example of what not to do when writing bindable APIs)
• We need a solution for cross-platform cairo (handwritten bindings)
• Tests (Next PR: Unit Test all of GLib/GObject)
• Port a subset of text editor as a sample project
  • Either fix deletion or remove it from the text editor demo (currently poorly implemented, prone to crashing)

[badcel]

We need to check the documentation generation as the old union generation behavior is restored which could break the documentation again.

[mjakeman]

Had this error when doing key input after the drawing area stopped drawing:

(dotnet.exe:17948): GLib-GObject-CRITICAL **: 18:25:20.068: g_object_remove_toggle_ref: assertion 'G_IS_OBJECT (object)' failed

So maybe it isn't signal related and in fact a sign some underlying object is being freed.

Edit: Closure is being freed too:

(dotnet.exe:16444): GLib-GObject-CRITICAL **: 18:32:16.389: g_closure_unref: assertion 'closure->ref_count > 0' failed

Edit 2: Another error:

(dotnet.exe:5284): GLib-GObject-CRITICAL **: 18:56:59.389: g_object_remove_toggle_ref: assertion 'G_IS_OBJECT (object)' failed
Fatal error. System.AccessViolationException: Attempted to read or write protected memory. This is often an indication that other memory is corrupt.
Repeat 2 times:
--------------------------------
   at GObject.Native.Object+Instance+Methods.RemoveToggleRef(IntPtr, GObject.Native.ToggleNotifyCallback, IntPtr)
--------------------------------
   at GObject.Native.ObjectMapper+ToggleRef.Dispose()
   at GObject.Native.ObjectMapper.Unmap(IntPtr)
   at GObject.Native.ObjectHandle.ReleaseHandle()
   at System.Runtime.InteropServices.SafeHandle.InternalRelease(Boolean)
   at System.Runtime.InteropServices.SafeHandle.Dispose(Boolean)
   at System.Runtime.InteropServices.SafeHandle.Finalize()

Edit 3: Some more findings, this is very bizarre.

I'm getting an output of "<NULL-class>" when checking the TypeNameFromInstance of the object that is having its toggle ref removed.

From glib's gtype.c:

const gchar *
g_type_name_from_instance (GTypeInstance *instance)
{
  if (!instance)
    return "<NULL-instance>";
  else
    return g_type_name_from_class (instance->g_class);
}

const gchar *
g_type_name_from_class (GTypeClass *g_class)
{
  if (!g_class)
    return "<NULL-class>";
  else
    return g_type_name (g_class->g_type);
}

This means the type class (g_class) is somehow NULL. Which means either the object is half-finalized or very badly corrupted.

Edit 4: The rest of the fields are all NULL as well. Looks like a simple case of use-after-free. Probably means the ToggleRef implementation is broken.