Won't provide awesome performance, but it shouldn't take much time to do and it will validate the extension's semantics, which is a good thing before working on #653657.
This doesn't seem like a Developer Tools bug?
I think this is, I thought one of us at DevTools (Victor?) would implement this.
Not very efficient but this does not require native implementation, which will take longer to write/review/land.
We will then switch to the efficient native implementation (which is definitely a canvas.webgl bug) as soon it lands with no change.
yup. I agree with Cedric that this is something that we can and should do. We are the primary consumers of this shim.
This is a placeholder for the native/optimized implementation of the extension, reusing underlying textures (or FBOs?) when possible.
First we need to all agree on a sane design for #653656, and it does not seem necessary for now as Tilt's framerate seems quite acceptable as it does not do interactive refresh yet afaik (the native implementation will be probably needed for interactive refresh rates - especially on pages using canvas or video heavily).
Additionally it might not make sense to implement this natively before Azure is ready, depending whether or not access to underlying gfx objects is going to significantly change. Benoit?
Please ignore comment #4. It was intended for #653657.
Created attachment 547059 [details]
First version. Completely untested.
Need to update according to comment #26 over at #653656.
- add readWindowTexCoords, always [0.0,0.0, <stride>, 0.0,1.0, <stride>, 1.0,1.0, <stride>, 1.0,0.0] in the shim
- rename MAX_WINDOW_REGION to MAX_WINDOW_REGIONS (+1!)
We're getting closer to making our tools update themselves. This would be a great start for Tilt, maybe we can start playing with this again? See also bug 703910.
Cedric, what do you think?
(In reply to Victor Porof from comment #9)
> Or we could skip this entirely and focus on the actual bug 653656 and bug
> 653657. I would somewhat prefer the latter
> Cedric, what do you think?
I think that's reasonable since the proposed API in bug 653656 seems to be in a good shape now (there's still a few issues to answer/validate tho).
Ok, I'll mark bug 653656 as P2.