fix readRenderTargetPixels() and adapt to implementation format

[nraynaud]

the renderTarget.texture was forgotten and I enjoyed the edit to integrate that too:

mrdoob#7226

(I can't find an editor were I can comfortably and safely 3-way hand merge a 4kloc file, so I just copy-paste)

[nraynaud]

I think that was a bad reflex, sorry.

[nraynaud]

Well, actually, the current dev branch uses console.error instead of THREE.error(), that was a copy/paste mixup.

[kintel]

Since my branch has diverged quite a bit from the upstream dev branch - I'd rather merge properly rather than patch this branch. I'll give you a note once it's pushed.

[kintel]

@nraynaud Hey, could you give the current depthtextures branch a spin? I think I merged everything correctly, but I don't have any tests for the code you contributed...

[nraynaud]

Yes, I will give it a try in the next few hours.

[nraynaud]

I think your changes work, I tried my main depthbuffer code, and it works, and the rest of my codebase seems fine after all the API change done.

The CSS renderer is broken, but I need more time to check for API changes or actual regression problems from the branch you pulled. I'm pretty sure it's not your code.

Thanks for your work.

[kintel]

@nraynaud Cool - is this related to your gcode stuff or are you working on some other fun stuff?

[nraynaud]

yep, it's for my CAM system.

I am projecting a mesh of the part I want to mill into the depth buffer (that gives me a height field), and then doing a minkowski sum of the part and the shape of the tool, that gives me a map of the lowest position the tool is allowed to be without destroying the part shape. Then I add some voodoo to actually determine a toolpath.

Currently I'm working on even more perversion of webgl: I'm extracting the medial axis of a polygon, entirely in the fragment buffer: http://nraynaud.github.io/webgcode/various_tests/test_voronoi.html
(it's still buggy for now)

[kintel]

was merged manually