Re: [AD] OpenGL unlock fix and discussion

[Elias Pschernig <elias.pschernig@xxxxxxxxxx>]

On Wed, 2010-02-17 at 10:53 -0700, Trent Gamblin wrote:
> It's debatable whether or not set_current_display
> should set the target or not, but it seems natural to me so
> I'm in favor of keeping it the way it is, unless someone
> has a good reason to change it.
> 

Yes. We should add a note about this to both al_set_target_bitmap and
al_set_current_display as it's not obvious at all.

Basically, our state system looks like this, with the extra hack that
the target bitmap is set to the backbuffer whenever the current display
changes:

global
    active system driver
        current config
per thread
    current display
        current transformation
        deferred drawing
    current target bitmap
        current clipping rectangle
        bitmap locking
    current blending
    new bitmap params
    new display params
    active file interface
    errno

This reveals another serious problem - the current transformation is
per-display. So it's impossible right now to change the transformation
when you draw into a memory bitmap but never create a display. I can see
three solutions:

1. An easy solution would be to put the current transformation into TLS,
just like blending.

2. Another solution would be to always require an active display for all
drawing operations. It would be conceptually cleanest (and we would have
a lot less TLS state since blending and target bitmap would fall away),
but I think we decided against it in the past as it would require to
always create a display before using any graphics operations. (We could
just provide a function which creates a dummy display, so it would not
be that hard to do.)

3. A third solution would be to leave things like they are, that is
require an active display for using transformations and deferred
drawing, but allow other graphics operations and things like clipping or
blending without a display. It wouldn't be very well designed though so
I don't think we want it.


-- 
Elias Pschernig <elias.pschernig@xxxxxxxxxx>