Mobile/Layers

See also Gecko:CrossProcessLayers.

Pre-Fennec work

• "shadow" layer trees (parent's reference to a child's tree) hooked into frame or widget code; currently they're only hooked into testing code. We need to associate TabChilds' shadow layer trees with their frames/widgets in the browser process.
• BasicThebesLayer pixel data shared efficiently with its BasicShadowThebesLayer (needed for realistic performance testing)

Fennec-specific work

2.0 b1 TODO

• (blassey?) band-aid for popup widgets in content processes
• (cjones) retained-content-only layer transactions
• clear transparent surfaces correctly
• invalidation of AJAX-y sites: search "foo" on google.com, click next; nothing happens. Zoom-in/zoom-out shows what should have been there.
• fix seaming, at least for simple cases (v. google.com @ 2.0 or 0.5 scale-zoom)
• shadow layer perf work
  • PROFILE MEASURE PROFILE
  • bug 570625 make ThebesLayer paint directly into shareable surfaces, remove third buffer
  • bug 523172 non-sucky "basic" shmem backend for android, flush existing down the toilet
  • (no bug) track updated region, only blit that from back --> front
    • ... unless measurements says we mostly just update the entire surface, in which case swapping and always repainting would be better
  • (no bug) smarter allocation of ThebesLayer backing buffers
  • drawing scale-zoomed layers should not be slow
• We should be using 16bpp for all surfaces on mobile platforms.
  • 16bpp is faster and doesn't take as much memory
• daringfireball, pmo, and twitter all have some white regions at the top of the page

2.0 b1 nice-to-have

• hacky slow fallback path for scrolling subframes (e.g. mobile gmail), only on pages with scrollable subframes
• (no bug) draw a "checkerboard" or other indicator over waiting-to-be-filled regions
• (no bug) make CanvasLayer paint directly into shareable surfaces, remove third buffer
• don't draw the caret in the content process

2.0 musts

• compositing Basic*Layer tree published by content, using GL/EGL in the browser process
• HW-accelerated video that bypasses content-process main thread
• fast scrolling of subframes
• better support for position:fixed elements
• WebGL? proxied?
• fast plugin drawing? that bypasses content-process main thread?

Questions that come with using browser element

• Where does panning and kinetic panning live?
  • Ben: hopefully as javascript in browser binding
• Where will handling of zoom metadata live?
• Where will the tap => element code live (including better link clicking and transforming coordinates for zoom level)?
• Where does double-tap to zoom live?
• Where does double-click simulation live? Nokia devices didn't support a native double click event. Maybe we still need to do some form of event re-dispatching to browser elements.
• Where does tap highlighting live? I think webkit has a tap-highlight-color CSS property.

Other questions

• What is the proper behavior for position: fixed elements when there is an additional displayport on top of the CSS viewport?
  • what do other browsers with displayport-type things do?
    • MicroB fixes wrt displaport, hides on pan, shows when pan finishes
    • mobilesafari appears to basically ignore position:fixed wrt displayport

Building

hg clone http://hg.mozilla.org/projects/cedar
cd cedar
hg clone http://hg.mozilla.org/users/bstover_mozilla.com/mobile2 mobile

From there, a normal opt or debug build will work.