- Go through things marked FIXME

  - Add calls to prepare and finish access where necessary.  grep for
    ACCESS_MEM, and make sure they are correctly wrapped in prepare
    and finish.

  - restore READ/WRITE in the fbcompose combiners since they sometimes
    store directly to destination drawables.

  - It probably makes sense to move the more strange X region API
    into pixman as well, but guarded with PIXMAN_XORG_COMPATIBILITY

  - Reinstate the FbBits typedef? At the moment we don't
    even have the FbBits type; we just use uint32_t everywhere.

    Keith says in bug 2335:

        The 64-bit code in fb (pixman) is probably broken; it hasn't been
        used in quite some time as PCI (and AGP) is 32-bits wide, so
        doing things 64-bits at a time is a net loss.  To quickly fix
        this, I suggest just using 32-bit datatypes by setting
        IC_SHIFT to 5 for all machines.

  - Consider whether calling regions region16 is really such a great
    idea Vlad wants 32 bit regions for Cairo. This will break X server
    ABI, but should otherwise be mostly harmless, though a
    pixman_region_get_boxes16() may be useful.

  - Make source clipping optional.
      - done: source clipping happens through an indirection.
	still needs to make the indirection settable. (And call it
        from X)

  - Consider optimizing the 8/16 bit solid fills in pixman-util.c by
    storing more than one value at a time.

  - Add an image cache to prevent excessive malloc/free. Note that pixman
    needs to be thread safe when used from cairo.

  - Review the pixman_format_code_t enum to make sure it will support
    future formats. Some formats we will probably need:

    	   ARGB/ABGR with 16/32/64 bit integer/floating channels
	   YUV2,
	   YV12

    Also we may need the ability to distinguish between PICT_c8 and
    PICT_x4c4. (This could be done by interpreting the A channel as
    the depth for TYPE_COLOR and TYPE_GRAY formats).

    A possibility may be to reserve the two top bits and make them
    encode "number of places to shift the channel widths given" Since
    these bits are 00 at the moment everything will continue to work,
    but these additional widths will be allowed:

    	     All even widths between 18-32
	     All multiples of four widths between 33 and 64
	     All multiples of eight between 64 and 128

    This means things like r21g22b21 won't work - is that worth
    worrying about? I don't think so. And of course the bpp field
    can't handle a depth of over 256, so > 64 bit channels arent'
    really all that useful.

    We could reserve one extra bit to indicate floating point, but
    we may also just add 

       	   PIXMAN_TYPE_ARGB_FLOAT
	   PIXMAN_TYPE_BGRA_FLOAT
	   PIXMAN_TYPE_A_FLOAT
    
    image types. With five bits we can support up to 32 different
    format types, which should be enough for everybody, even if we
    decide to support all the various video formats here:

    	        http://www.fourcc.org/yuv.php

    It may make sense to have a PIXMAN_TYPE_YUV, and then use the
    channel bits to specify the exact subtype.

    What about color spaces such a linear vs. srGB etc.?


done:

- Run cairo test suite; fix bugs
	- one bug in source-scale-clip

 - Remove the warning suppression in the ACCESS_MEM macro and fix the
    warnings that are real
	- irrelevant now.

- make the wrapper functions global instead of image specific
	- this won't work since pixman is linked to both fb and wfb

- Add non-mmx solid fill

- Make sure the endian-ness macros are defined correctly.

- The rectangles in a region probably shouldn't be returned const as
  the X server will be changing them.

- Right now we _always_ have a clip region, which is empty by default.
  Why does this work at all? It probably doesn't. The server
  distinguishes two cases, one where nothing is clipped (CT_NONE), and
  one where there is a clip region (CT_REGION).

- Default clip region should be the full image

  - Test if pseudo color still works. It does, but it also shows that
    copying a pixman_indexed_t on every composite operation is not
    going to fly. So, for now set_indexed() does not copy the 
    indexed table. 

    Also just the malloc() to allocate a pixman image shows up pretty
    high.

    Options include

      - Make all the setters not copy their arguments

      - Possibly combined with going back to the stack allocated 
        approach that we already use for regions.

      - Keep a cached pixman_image_t around for every picture. It would
        have to be kept uptodate every time something changes about the
        picture.

      - Break the X server ABI and simply have the relevant parameter
        stored in the pixman image. This would have the additional benefits
        that:

          - We can get rid of the annoying repeat field which is duplicated
            elsewhere.

          - We can use pixman_color_t and pixman_gradient_stop_t
            etc. instead of the types that are defined in
            renderproto.h