Issue metadata
Sign in to add a comment
|
Incorrect rendering of spell-check underlines |
||||||||||||||||||||||||
Issue descriptionGoogle Chrome 72.0.3626.15 (Official Build) dev (64-bit) Revision 22ece8dd03ba589d4dd022d8762d00932df48be1-refs/branch-heads/3626@{#272} Platform 11316.18.0 (Official Build) dev-channel eve What steps will reproduce the problem? (1) Edit bugs on https://bugs.chromium.org (2) Add email addresses to the cc: list What is the expected result? Sometimes, if Blink's spellcheck underlines email addresses in the cc: list, then the red underlines will be incorrectly rendered. Here's a recording where several small dots are drawn in sync with the caret: https://photos.app.goo.gl/PgLXtXvS8wTTd2uKA Have seen this while composing Gmail messages too. I'm marking this as P1 because the artifacts are *very* similar to some which we've seen on the GPU bots in Issue 917040. At this point we think that those are actually bad hardware - but I don't know at what level these spellcheck underlines are handled in Chrome, and if there's some sort of awful compositing bug where they're sometimes accidentally persisting from page to page, it's crucial that we get to the bottom of that right away.
,
Jan 10
Should be a paint bug. Not familiar with that part.
,
Jan 10
Painting of these markers is conceptually painting of a wavy line (A Skia bezier path). If there's a bug it's downstream of Blink, either in raster or the underlying hardware rendering. This is the code: https://cs.chromium.org/chromium/src/third_party/blink/renderer/core/paint/applied_decoration_painter.cc?dr=CSs&g=0&l=108
,
Jan 10
,
Jan 11
It looks to me like a partial invalidation bug (maybe the squiggly lines extended further, we backspaced a bit, and left previous content at the bottom in one of the partial raster frames. pdr@, do you know if anything about damage rects has changed in this area of rendering? I guess it's possible that Skia is drawing outside of expected bounds as well - so if this doesn't feel like paint to you, let's follow-up w/ bsalomon@.
,
Jan 14
I am able to reproduce this on linux with the following steps: 1) zoom to 125% 2) Visit https://bugs.chromium.org/p/chromium/issues/detail?id=917264#c5 3) type cc: buffalobuffalobuffalo buffalobuffalobuffalo buffalobuffalobuffalo 4) Wait for spellcheck underlines 5) double-click second "word" to select it, then delete it. 6) Notice parts of the red underlines are left over. I think this is https://crbug.com/856982 which goes way back (at least M69). @kbr, because this is not a recent issue, does that exonerate this issue from being the root cause of 917040, or should this still be treated as P1?
,
Jan 14
,
Jan 14
Attaching a fairly simple repro on linux. Just type in the box to see invalidation issues.
,
Jan 14
I'm attaching a fairly simple repro case of the bug that doesn't require an input box. This is probably one solid day of work to fix for the paint team. Assigning to kbr@chromium.org for comment #6: should this be P1? If so, please re-assign back to me.
,
Jan 15
At some point I started fixing it but the way it's set up makes it painful. I should just restart that work. I don't think it's P1 because, while very annoying, it doesn't affect functionality in a significant way.
,
Jan 15
I thought this might be related to somewhat random pixel test failures seen in Issue 917040, but am pretty sure at this point that those were caused by failing hardware. Downgrading to P2 and reassigning to pdr@.
,
Jan 15
Strange coincidence, Xianzhu just sent me a patch for the dupe of this bug. I'll go ahead and merge. |
|||||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||||
Comment 1 by yosin@chromium.org
, Jan 10Status: Assigned (was: Untriaged)