1 - Remove the warning suppression in the ACCESS_MEM macro and fix the
4 - It probably makes sense to move the more strange X region API
5 into pixman as well, but guarded with PIXMAN_XORG_COMPATIBILITY
7 - Go through things marked FIXME
9 - Test if pseudo color still works. It does, but it also shows that
10 copying a pixman_indexed_t on every composite operation is not
11 going to fly. So, for now set_indexed() does not copy the
14 Also just the malloc() to allocate a pixman image shows up pretty
19 - Make all the setters not copy their arguments
21 - Possibly combined with going back to the stack allocated
22 approach that we already use for regions.
24 - Keep a cached pixman_image_t around for every picture. It would
25 have to be kept uptodate every time something changes about the
28 - Break the X server ABI and simply have the relevant parameter
29 stored in the pixman image. This would have the additional benefits
32 - We can get rid of the annoying repeat field which is duplicated
35 - We can use pixman_color_t and pixman_gradient_stop_t
36 etc. instead of the types that are defined in
39 - Reinstate the FbBits conditional typedef? At the moment we don't
40 even have the FbBits type; we just use uint32_t everywhere.
42 Keith says in bug 2335:
44 The 64-bit code in fb (pixman) is probably broken; it hasn't been
45 used in quite some time as PCI (and AGP) is 32-bits wide, so
46 doing things 64-bits at a time is a net loss. To quickly fix
47 this, I suggest just using 32-bit datatypes by setting
48 IC_SHIFT to 5 for all machines.
50 - Consider whether calling regions region16 is really such a great idea