594921 - Have cairo_xlib_surface support "simple self-copies" for Pixmap drawables that it doesn't own, and allow clients to "take" an owned Pixmap from a cairo_xlib_surface

Status: RESOLVED INCOMPLETE

(Core :: Graphics, defect)

Description

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

My use case is the following

 - have a "content process" that paints web content into X surfaces (through Gecko's layers system)
 - have a "compositor process" that composites X surfaces (layers) painted by the content process and renders to screen
 - the content process paints into "back surfaces", and the compositor process composites "front surfaces".  (Each layer has a front and back surface.)
 - when the content process pushes a back-->front surface update to the compositor process, one of the following may happen in the compositor process
   (1) back/front surfaces are swapped
   (2) updated region is copied from back-->front

Re: the first part of this bug, for scrolling performance, our layers system relies on the content process being able to do simple self-copies of its back surface.  According to http://hg.mozilla.org/mozilla-central/annotate/9c5a7e114cda/gfx/thebes/gfxXlibSurface.cpp#l200 (bug 579736), which I admittedly don't understand well, cairo_xlib_surface will only do this for surfaces it creates itself (that it knows are Pixmaps).

Re: the second part of this bug, our IPC system maintains a strong single-owner model for "shared" surfaces.  Only one or the other process, not both, can read/write/destroy a surface at any given time.  We use OS-level mechanisms to make this work and enforce the ownership model for shared-memory image surfaces (POSIX and SysV shm).  This model allows us to transfer ownership of surfaces back and forth between content/compositor process.  We would like to fit Xlib surfaces into this model as well.  This entails being able to destroy an Xlib surface in a different process than the one that created it.

There are various workarounds and hacks we can employ in Gecko, but ideally I think we'd want new cairo APIs like

  /* cairo_xlib_surface can treat |pixmap| as if it were created with cairo_surface_create_similar() */
  cairo_surface_t* cairo_xlib_surface_create_for_pixmap(Display *dpy,
                                                        Pixmap pixmap,
                                                        Screen *screen,
                                                        XRenderPictFormat* format,
                                                        int width,
                                                        int height);
  /* True if the Drawable underlying this cairo_xlib_surface will be destroyed along with the cairo_xlib_surface. */
  cairo_bool_t cairo_xlib_surface_owns_drawable(void);
  /* Precondition: cairo_xlib_surface_owns_drawable() is True.  Transfers ownership of the underlying Drawable to the caller. */
  Drawable cairo_xlib_surface_take_drawable(void);

Thoughts?

Comment 1

[Joe Drew (not getting mail)]

Chris, I don't think we still want this, so I'm going to resolve incomplete. Please reopen if I'm wrong.

Comment 2

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

We still want this for mulitprocess/x11/cairo/no-GL configurations, but the odds of gecko supporting that in an important product are slim and approaching none.