[cairo] resize image surface

[chris nuernberger]

I have a situation where I am rendering text using pango to cairo surfaces
and uploading them to openGL as textures.

I render lots of small, usually one line but dynamically changing pieces of
text.

I am going to attempt to add an image resize function so I can resize an
image context efficiently so that it will only allocate new memory if the
original data capacity of the image context is insufficient to hold width *
stride * height.  I would like to resize in the average case without any
calls to malloc or free.

I was wondering if there was a way to force cairo to treat a surface as if
it had a smaller width than it was originally allocated with instead when
cairo is rendering?  This would solve the problem.

I need to take the surface post rendering and upload the surface data to
openGL; uploading data, especially texture data to opengl is pretty
sensitive to the size of the data.  Other options would include manually
copying the subrect into a smaller image or something like that but this
isn't as ideal as simple setting the image rect to be appropriately sized
before rendering.

Since there is no image_surface_resize function there seem to be two
options.  First would be to write one which is the most straightforward way
to a naive user like myself.  Second would be to somehow tell cairo that
this surface is really less wide than it may seem at first glance.  Any
other option seems to require either another memcpy post render somewhere
or allocating memory for every text object every frame.

I am moving forward with attempting to resize the image surface, I would
like to at least know what possible gotchas there are and how I could
ensure they don't happen.  One possible gotcha I found after looking at the
image surface struct was the cairo_damage could now have regions that are
outside the dimensions of the image.  This won't cause crashing but it
might cause invalid rendering.  Are there any other failures I should be
aware of?  Any interactions with pango that might make this pathway
infeasable?  I can always recreate the pango layer although I think that is
kind of silly and most likely unnecessary.

Another perhaps more conservative path forward would be to drill down into
every possible malloc and free that occurs during image surface creation
and replace those with possibly user-overloaded functions and then ensure
on the client side that I overload those functions to hit a special memory
pool.  This, combined with the image surface creation system that takes a
pre-allocated piece of memory would eliminate at least the memory traffic
during rendering although it still does seemingly more work than it needs
to.  It may be the only real safe way to resize the image surface depending
on how much context cairo stores on the image surface and how difficult it
is to clear this context.

Any help would be appreciated.

Chris

-- 
A foolish consistency is the hobgoblin of little minds - Emerson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cairographics.org/archives/cairo/attachments/20120929/6564bd86/attachment.html>