WebGL (three.js) erratic working

[xunfin]

Using a three.js program using webGL launched from Brackets I am erratically (80% of the time) getting an error:

"Error creating WebGL context."

from within the three.js THREE.WebGLRenderer constructor..

Happens with all programs I have tried including the trivial rotating cube example at http://threejs.org/docs/#Manual/Introduction/Creating_a_scene

Notes:
Programs tested work correctly when launched in (standard) chrome.
Drag and drop to a new tab in the bracket's chrome window (live preview) works.
If I launch the bracket's chrome window (live preview), add some extra tabs, and close the launched program tab, then launch live preview (ie on the chrome live preview window) - it works.

Summary:
This issue only occurs when live preview causes a new chrome window to appear.

Using Chrome 33.0.1750.117 m WinXP 32 bit. This wasn't a problem 1 month ago

Quick test seems to confirm that there is no issue on Brackets 35.

(Is Sprint 36 starting chrome "too quickly" before all the add-ins can initiallise ?) Thanks.

[njx]

Very weird. I do suspect that there might have been some change in 35 that changed our live preview startup timing. OTOH, it could also be exacerbated by a Chrome change, since we've seen some more instability on that side (e.g. #6275, which also seems to be sensitive to the timing of when we communicate with the page during load).

To @redmunds to look at as related to the other live preview issues. Low pri for now since it seems like a fairly specific case.

[redmunds]

@xunfin Since it works when Chrome browser with Live Preview profile is already running, but usually not if starting browser from scratch, I suspect that it's related to the "Wait time for live development to start is too short" issue. This will hopefully be fixed by pull request #6997 (already been merged into master) or pull request #6889 (still in review). Let us know if you are able to test your scenario against either of those branches and what the results are.

[njx]

@redmunds - just out of curiosity, since I haven't been fully following the discussion around #6997, do we think that something changed between 35 and 36 that might have made the timing issue worse?

[redmunds]

@njx Oops, I missed the "no issue on Brackets 35" comment for this issue. I was under the impression that Live Preview changes were made in Sprint 36, but I just looked through the github history and don't see any significant changes.

@jasonsanjose Can you think of any changes that may have slowed down the timing of Live Preview startup between 35 and 36?

The next thing that may have caused it was the "launch chrome in a separate profile" change, but that (finally) made it in to Sprint 35 and is Mac-only.

[jasonsanjose]

Can't think of anything

[redmunds]

@xunfin We put some effort into looking at Live Preview bugs this sprint, and can't find any change in Brackets that we think caused this. It works for me every time on Win7. It could be a Chrome update.

We made some improvements to Live Preview this sprint, so please try out Release 37 when it comes out (or fetch latest master from Brackets repo) and let us know how it goes.