Trying to create a reduction of this page, I noticed that deleting the two iframes in devtools causes the page to render correctly. Both are supposed to be 1x1 pixel frames, absolutely positioned at -100px in y. I'm guessing that the filters for those frames are computing the wrong destination buffer size and covering the page.

Note that the above CL is not revertible at this point, so it's a little more difficult to verify that it's to blame.

Turning on the bat-signal for paint folks (chrishtr, pdr).

Any clues here? It seems like iframes are getting the wrong destination filter size, perhaps due to my change above (https://codereview.chromium.org/1459953002).

Re-reviewing your patch, I can't see anything clearly wrong. If you manually revert the
parts in ReferenceFilterBuilder.cpp does it fix the bug?

Ping on comment 3. Also, were you able to make a reduction?

No, I haven't been actively working on this.

Ok. This is currently marked as a P1 regression. We also have a goal starting this quarter to not allow paint regressions to reach the stable channel.

Re: #6: given that it broke in 49, I think that ship has sailed. Unless we find this breaking the web more widely, I think we should punt to a later release.