- check rounding

- remove cairo-surface-fallback

- simpleops-ize:
  - atomics, mutexes (partly done, see wip/simpleops)
  - ctor/dtor
  - investigate tls for caches
  - wideint?

- gradients:
  - mesh
  - along stroked path??? (did someone at least try to define what should happen?)
  - filtering (supersampling? analitical?)

- testsuite:
  - include fonts in package
  - specify freetype,fontconfig versions that pass the testsuite
  - separate font backends refimages
  - make sure that the buildbot passes the testsuite

- snapshots:
  - improve definition&interaction with surfaces (subsurface-parent-modify failure; see wip/references)
  - cow snapshots for recording
  - multithreading???

- damage? (reopening the thread would be a good start)

- path:
  - double? (investigate performance loss)
  - arc-to
  - stroke-to-path (easier if paths are double and support arc-to)
    - would allow arbitrary caps/joins/cusps
  - boolean ops (check if b-o is appropriate, investigate balaban)
  - "reusable paths" (cache stroked path & backend-specific paths, see http://lists.freedesktop.org/archives/cairo/2010-April/019711.html)

- gstate:
  - move most optimizations from backends to gstate
  - check that optimization does as much as possible (never pass DEST/completely clipped etc to a backend)

- font:
  - see testsuite!
  - size 0 fonts... should they get to the backends?
  - improve scale matrix handling (translation in gstate, linear transformation in font backend)
  - text-glyph-range, toy-font-face (failing bad with @cairo:)

- quartz-font:
  - unicode support (might require coretext... different backend?)
  - subsetting (done?)
  - _load_truetype_table (done?), _index_to_ucs4 (is it possible? see unicode support)

- user-font:
  - test me! last time there were many failures
  - halo, halo-transform
  - text-transform
  - user-font-rescale

- matrix:
  - projective? (would breaks lots of assumptions in cairo, but it is already supported by pixman)

- win32, svg
  - test, maintain

- xcb
  - what would it take to make it supported?





- cairo2
  - remove set_device_offset (provide the same functionality through a wrapper surface)
  - no current point line_to->move_to+line_to
  - a lot of subtle chanes tha will break most exising applications