Issue metadata
Sign in to add a comment
|
Font atlas leak to background-image
Reported by
teodorma...@gmail.com,
Dec 11 2016
|
||||||||||||||||||||||
Issue descriptionUserAgent: Mozilla/5.0 (Windows NT 6.0; WOW64) AppleWebKit/537.27 (KHTML, like Gecko) Chrome/54.0.2733.32 Safari/537.27 Example URL: https://thexeos.github.io/cfal/index.html Steps to reproduce the problem: Page's font atlas leaks onto the first background-image, which has dimensions equal to the screen dimensions and is displayed on the element with the height at least 1px taller than the image itself, when the font size is at least 2px different from default. For example: - screen size 1920x1080 - image size (JPG) 1920x1080 - body has background-image set - body has height of 1081px (1080px + 1px) - default font size is 16px - document contains at least some text nodes - text has font size of at least 18px (16px + 2px) What is the expected behavior? What went wrong? Depending on the parameters (font size, height of the element with background-image, screen size, image size, text contents) it is possible to make the font atlas toggle (disappear and then reappear) on each repaint, for example by highlighting text letter by letter, or by switching between two tabs. Does it occur on multiple sites: Yes Is it a problem with a plugin? No Did this work before? N/A Does this work in other browsers? Yes Chrome version: 54.0.2733.32 Channel: n/a OS Version: 6.0 (Windows Vista, Windows Server 2008) Flash Version: Shockwave Flash 23.0 r0 The graphics card is Intel HD Graphics 520, driver version 21.20.16.4526. This has been happening for a few months now. I thought new GPU drivers would fix this, however several updates later the problem is still there.
,
Dec 17 2016
Yes, it is happening with the newly created profile and in incognito mode. I guess I didn't explicitly mention the fact that GPU Rasterization is used (hence the font atlas). I have enabled it with a flag as my GPU (Intel) doesn't seem to be affected by Issue 527565 (slow path rendering), and it gives me an FPS boost on pages with heavy animations/transitions. It could be the video card driver issue, but conditions needed to reproduce the issue are very dependent on the displayed web page. I have only encountered this issue on one live webpage in the past few months. Normally, at least one of the conditions is not met on other webpages. I think if it was a bug in the driver, then more applications and webpages would have been affected by it. I can provide about:tracing, but last time I ran it, I couldn't trace font atlas generation/uploading to determine if a new version of it "lands" on the background image or if the zero-copy mapping links the atlas to the position of the background-image (0,0) and when compositing the layer the atlas gets copied in it's entirety to the layer. P.S. I thought my email address was not to be displayed in full according to the settings page.
,
Dec 20 2016
Thanks for more detailed feedback. Routing GPU team to triage this issue. P.S. sorry for exposing your address. I'll delete the comment#2. The comment#2 was just asking more info, so deleting should not lose any important information here.
,
Dec 20 2016
,
Jan 10 2017
bsalomon@, sounds like we're failing to clear an intermediate? not sure...
,
Jan 10 2017
I'm guessing that this is a repro of 665673 (which was denied a merge request to M55). OP, can you verify that this doesn't reproduce on the Canary channel of Chrome?
,
Jan 23 2017
I can confirm that this is fixed in Canary.
,
Jan 26 2017
This fix is in M56 stable, so closing. |
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by ajha@chromium.org
, Dec 12 2016