[cairo] Road to 1.10

[M Joonas Pihlaja]

On Thu, 12 Feb 2009, Chris Wilson wrote:

> This is what I've found from diffing the headers since 1.8...

[snip]

> New experimental backends:
> 
>    Simple DirectMedia Layer
>    CairoScript

Regarding the SDL backend, could we drop it before it gets any more 
use?  While it certainly makes cairo/SDL integration a bit easier, 
it's not really doing all that much, but does have its own set of 
problems in turn:

- Makes assumptions about SDL_Surface locking semantics that aren't 
necessarily true; namely that SDL_Surface->pixels is valid and stays 
constant.

- Is impossible to use correctly with 32 bit SDL_Surfaces which use 
per-pixel alpha.  SDL's pixel format is unpremultiplied vs. cairo's 
premultiplied format.  The pixels end up in cairo's format in the 
SDL_Surface and later when blitting using SDL things look wrong.

There are also some minor issues like allowing any pixman supported 
format to be used with cairo instead of the few CAIRO_FORMAT_* ones. 
While we could fix most of the problems in the implementation, we'd 
still be left in a situation where we're shipping a surface which 
doesn't bring much value to the table.  The main thing that it does 
provide that the user cannot at present do themselves is damage 
tracking.  Even there, it's notifying SDL_Surfaces immediately after 
rendering, but that's not necessarily what the user wants to happen!

Since SDL integration mainly boils down to pretty much just "make an 
image surface, unpremultiply sometimes", I don't think it needs its 
own surface.  For the unpremultiplication part, I've put together a 
set of helper functions that make that nearly transparent, so that all 
the user needs to do is indicate that they'd like the contents of the 
image surface to be flushed to the SDL_Surface now.  It's available 
(with demos) at

http://cgit.freedesktop.org/~joonas/cairosdl/

So if there aren't strong objections, I'd like to go ahead and take 
the cairo_sdl_surface_t out.

Cheers,

Joonas