Very bad visual artifacts when zooming in on Polymon |
||||||||||||||
Issue descriptionGoogle Chrome 53.0.2785.154 (Official Build) (64-bit) Revision 0 Platform 8530.96.0 (Official Build) stable-channel samus ARC 3322987 What steps will reproduce the problem? (1) Open https://polymon.polymer-project.org/ (2) Login with a google account. (3) Tap Polydex (4) Pinch zoom the window in and out. Also (3) Zoom in on the mute icon. (4) Notice the mute icon vanishes after zooming in a bit. There's a bunch of visual artifact bugs here: - I see super bad visual artifacts on my Android phone, sometimes the entire screen blinks black, sometimes I see what looks like half of each layer at the different scales. You need to zoom in and out repeatedly to see it get super bad, it seems to get worse and worse until it blinks black. - On my Android phone I see what looks like the mute icon is painted over itself repeatedly, if you zoom in far the whole icon vanishes. - On my Pixel Chromebook I see bad visual artifacts if I pinch zoom where I seem to see various versions of the tiles at different scales appear all at once, like a mix of tiles on different layers at different scales. This might be related to the bug where tapping a polymon in this view blinks the buttons. That seems to be right after a scale animation on the layers. I tried with with and without Ganesh on my Android phone and it reproduces either way.
,
Oct 26 2016
The mute icon looks like a separate issue with Emoji. https://bugs.chromium.org/p/chromium/issues/detail?id=659463 Here's another screenshot of the visual artifacts when zooming too.
,
Oct 26 2016
Is the majority of the corruption you are seeing momentary? If the visual issues seem to "flicker", I think this is just a case where we are checkerboarding as you pinch-zoom. Because the site has a large number of layers stacked on top of each other, you see through to the underlying layers while some of the upper layers checkerboard. Because all layers in the site are kept visible, we have to re-raster them at each scale (all the way back to the map layer at the very bottom). This ends up creating a lot of extra raster work, which increases checkerboarding. I'm not sure if we could auto-detect occlusion better in some of these cases, but the site could reduce Chrome's workload significantly by hiding layers that are not currently visible. This would probably alleviate the majority of these issues. This would also reduce memory consumption - currently it takes ~130mb to keep all the visible layers in memory - it appears that there are round ~9 full overlapped screens of tiles being kept alive. There may still be some issues in Chrome - the other flickering bug (without zooming) is a bit concerning, will continue to investigate.
,
Nov 16 2016
,
Nov 23 2016
Ping? This issue has been untriaged for almost a month.
,
Nov 24 2016
As per #4, a large number of issues here is the fact that the site is creating too many layers and starves the renderer of tile memory. This results in activations to checkerboards which looks like missing content. I'm not sure about layer positioning and whether that's a layout bug (+chrishtr), but it's somewhat hard to understand what is going on given the number of layers here. On n5x, on one frame, I see the active tree taking up 212mb of tile memory (out of a 256mb cap), with a pending tree being there as well. It's unlikely that the invalidations on the pending tree would fit in the remaining 44mb of tile memory, so it most certainly results in the next frame checkerboarding.
,
Dec 29 2016
cdata@, can you reproduce with less memory than mentioned in comment 7? If there is a layout or paint bug, it should reproduce with a smaller testcase. Otherwise I think this bug is WontFix.
,
Jan 10 2017
|
||||||||||||||
►
Sign in to add a comment |
||||||||||||||
Comment 1 by esprehn@chromium.org
, Oct 26 2016