Issues:

1. Should all operations be async? We can sync on read/write turnarounds.
   Allows for higher throughput using hw accel, at the cost of a few check per
   operation. put_pixel will be adversely affected, but that's not the
   performence case.


2. al_blit() cannot cross-blit from one video bitmap family to another.
   Should we lift this restriction? Render to Texture makes this
   implementable on GL/D3D. Allegro could force a memory bitmap copy
   on systems that don't support this functionlity (are there any?).
   What about blitting from the frame buffer in one context to another?
   We'll pry use a memory copy for that.


3. Source clip rectangles. Are they really that useful? FOr their limited
   uses, could we leave this functionality to the user / add-ons?
   Computing the intersections/unions of 3 sets of clip rectangles
   is not trivial. Not even DDraw exposes this functionlity, so it's not
   hw accelerated.
   Dropping this would greatly simplify the implementation of some functions
   (al_blit_region3()).


4. Multiple sets of clip rectangles. How useful is this? Would blit with
   a mask bitmap be more useful? DDraw has lots of restrictions on this
   (cannot use on a window, colorkeying must be disabled, etc).


5. Support for floating-point frame buffers / bitmaps. Allegro shouldn't have
   too much trouble supporting them.


6. Monochrome bitmap support. We could drop it in favor of paletted bitmaps
   (For fonts) or luminence bitmaps. Arguably, those are far more useful.
   They can also be hardware accelerated.


7. Sub-bitmap memory alignment. ModeX requires an alignment to 4 pixel boundaries.
   Can Allegro cope without this restriction?


8. ModeX support. Should we still support it? It just makes our lives twice as hard,
   and doesn't help much.
   Elias proposed a double-buffering scheme for supporting mode-x at a performence loss.
   Basically, only the mode-x driver has access to the screen.
   Double buffering apps wouldn't see any perf loss. Page flip isn't supported.
   Only direct screen writes will run at half speed.
   Is this acceptable?