When a gfxWindowsSurface is created around an HDC it currently always believes it's opaque. For the glass window surface this isn't true. It should be told properly if it's wrapping a transparent DC, this allows BasicLayers to later determine if it's dealing with a transparent surface so it knows if it needs to clear it when reusing it.

Comment on attachment 456229 [details] [diff] [review] Allow HDC surfaces to contain alpha The new function needs documentation describing what it is there for and what it does. I think one problem, is that the function doesn't really make it clear that difference between cairo_win32_surface_create and cairo_win32_surface_create_transparent is not really in the surface that's created but in the way that we interpret the data associated with the HDC. A better name would help, though I'm not exactly sure what that name should be.

roc and I came up with cairo_win32_surface_create_with_alpha which I think is better.

With Both of the patches shown here I was seeing bug 552114 with OOPP on.

Comment on attachment 456603 [details] [diff] [review] Part 1 v2 >- * Creates a cairo surface that targets the given DC. The DC will be >- * queried for its initial clip extents, and this >+ * Creates a cairo surface that targets the given DC. The DC will be >+ * queried for its initial clip extents, and this will be used as the >+ * size of the cairo surface. The resulting surface will always be of >+ * format %CAIRO_FORMAT_ARGB32 (as supported for transparent windows with >+ * Windows Vista and up. Fix up the comment and a close paren.

http://hg.mozilla.org/mozilla-central/rev/c48562866467 http://hg.mozilla.org/mozilla-central/rev/8c3a6563be2d