svg stroke-dasharray rendering deformity on small dash size when on a rotated circle
Reported by
jonbur...@gmail.com,
Jun 21 2017
|
|||||||||||||
Issue descriptionUserAgent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_12_5) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.36 Steps to reproduce the problem: 1. Set small stroke-dasharray size on a circle 2. Wrap circle in <g> 3. Set <g> transform to rotate(-90, X, X); What is the expected behavior? The stroke dash should not become deformed What went wrong? The stroke dash became deformed Did this work before? N/A Does this work in other browsers? Yes Chrome version: 58.0.3029.110 Channel: stable OS Version: OS X 10.12.5 Flash Version: https://codepen.io/jon/pen/eRWXgo?editors=1100 The rendering problem is tricky in that it manifests itself at different sizes, and screen sizes - however I cannot reproduce the glitch unless there is the rotate(-90, X, X) or rotate(270, X, X) or anything close to those rotations - so I'm assuming it a problem between transforms and rendering size/position and may occur in other situations also that I have not come across. I have been unable to reproduce this deformity in Firefox or Safari - however it is still present in Chrome Canary at this time.
,
Jun 21 2017
My guess would be that this is a GPU rasterization bug - could you provide/attach a copy of the chrome://gpu page? (May contain useful information in case this bug does not reproduce generally.)
,
Jun 21 2017
,
Jun 22 2017
So far I've been unable to repro on my Mac. Reporter, does this go away if you disable GPU rasterization in about://flags?
,
Jun 22 2017
The bug is still present after disabling GPU rasterization and relaunching :( Also, if it helps at all, I'm including exact details of my Mac. Finally, this bug is also present in Opera, but not Safari - so it's something which I guess is a glitch in Blink rather than something old from Webkit
,
Jun 23 2017
I can repro this on Windows/Linux/Mac with or without GPU raster. Note that it helps to get a tall/narrow window and then play around with the window height. As you adjust the height, you should see the issue. Putting code-pen into "full page" view makes this easier to repro (but requires a free account). If you show tile boundaries (see attached image) you can see that we are failing to draw the content into the top tile. It's likely that an interaction between the rotation and our boudning box calculations is being done incorrectly, and we're deciding that the content doesn't overlap the top tile (even when it does). Over to pdr@ for blink investigation.
,
Jun 23 2017
Testers, can you please follow these instructions: 1) Load https://codepen.io/jon/pen/eRWXgo?editors=1100 2) Resize the middle slider, notice that the purple area can get clipped.
,
Jun 27 2017
I bisected this and it has existed for a long time so not a regression after all. Here's a minimized repro: http://output.jsbin.com/jimihah/quiet I think we're getting the wrong bounds from Skia, possibly due to numeric precision of the stroke. Florin, would you be able to triage this further?
,
Jun 27 2017
,
Jun 27 2017
Filed https://bugs.chromium.org/p/skia/issues/detail?id=6806.
,
Jun 27 2017
|
|||||||||||||
►
Sign in to add a comment |
|||||||||||||
Comment 1 by jonbur...@gmail.com
, Jun 21 201739.5 KB
39.5 KB View Download