After further investigation, captureStream() might not be the culprit
here: if we just do <video1>.captureStream() and do not plug it into 
the <video2> tag, the source <video1> _does not get affected_. E.g. 
if we plug the captured MediaStream into a MediaRecorder, the source
video tag is not corrupted.

This seems to point to some bug in MediaStreamVideoRendererSink and/or
WebMediaPlayer_MS.

Codepen illustrating the bug:
http://codepen.io/miguelao/pen/OpXVVK?editors=1010#0


Codepen with MediaRecorder, showing no bug:
http://codepen.io/miguelao/pen/oZzQLO?editors=1000

Adding some peeps.

dcastagna@ do you know if we're (re)using some resource in 
WebMediaPlayerMS [1] that corrupts a "normal" WebMediaPlayerMS
(or its internal SkCanvasVideoRenderer...) ?


[1] https://cs.chromium.org/chromium/src/content/renderer/media/webmediaplayer_ms.h?q=WebMediaPlayer_MS&dr=CSs

Is this a regression? Is this happening only on Mac?

We save a cache of the last converted VideoFrame in SkCanvasVideoRenderer::last_image_, do you think it could be related to this issue?

ccameron@ has been working on video color conversion on Mac lately, any chance this could be related to  crbug.com/676054 ?

1) this is a regression: although this feature is not shipped,
it was announced and the demo worked back then (~EOQ216).

It's positively only happening on Mac, at least as far as I
have tested, Linux and Win are OK.

+ccameron@ for more inputs

Can you quickly check with --disable-features=video-color-management (and --enable-features as well)?

I'm running 

./out/gn/Chromium.app/Contents/MacOS/Chromium \ -vmodule="*html_*_element_capturer*=2,*HTMLMediaElementCapture*=1,*video_track_recorder*=3"  \ --enable-blink-features=MediaCaptureFromVideo \ https://codepen.io/miguelao/full/OpXVVK --disable-features=video-color-management

but still the original <video> goes bananas colour-wise.

Can you bisect if it's a regression?

I'm also taking a look as soon as my local build completes...

I have a local repro too.

Bisect isn't running any recent builds, complaining that

dlopen /private/var/folders/00/191h0000h01000cxqpysvccm005460/T/bisect_tmpSXkMIn/chrome-mac/Chromium.app/Contents/MacOS/../Versions/58.0.3027.0/Chromium Framework.framework/Chromium Framework: dlopen(/private/var/folders/00/191h0000h01000cxqpysvccm005460/T/bisect_tmpSXkMIn/chrome-mac/Chromium.app/Contents/MacOS/../Versions/58.0.3027.0/Chromium Framework.framework/Chromium Framework, 261): no suitable image found.  Did find:
	/private/var/folders/00/191h0000h01000cxqpysvccm005460/T/bisect_tmpSXkMIn/chrome-mac/Chromium.app/Contents/MacOS/../Versions/58.0.3027.0/Chromium Framework.framework/Chromium Framework: file too short

Is anyone having luck bisecting this?

Fix ready.

Just tried r455269 which includes the fix in
https://bugs.chromium.org/p/chromium/issues/detail?id=699123#c7
and I cannot repro, all looks normal, thanks ccameron@ !