New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 914097 link

Starred by 1 user

Issue metadata

Status: Assigned
Owner:
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Chrome
Pri: 1
Type: Bug



Sign in to add a comment

Bookmark menu fails to paint if it has more than 8 items that requires eliding.

Project Member Reported by osh...@chromium.org, Dec 11

Issue description

Original bug:  crbug.com/911272 

This is regression when GPU rasterization is enabled.

This happens when you add more than 8 items that is long enough to
require shader based eliding.

It paints correctly if I remove the shader here:

https://cs.chromium.org/chromium/src/ui/gfx/render_text.cc?rcl=b847f9c1759f900bc7183a4fe2084f3c68f53ab2&l=1309


 
Forgot to mention. This is only seen on veyron_micky/veyron_fievel so far.
Cc: jamescook@chromium.org
This stopped reproducing on 73. Enabling SingleProcessMash (http://crrev.com/badd8a0) seems to make the issue go away.

James, do you have any idea what could be different that changes how we render text items in the bookmark menu?
Cc: msw@chromium.org
I don't think we're doing anything different with text raster in SingleProcessMash. Bookmark menus open a new views::Widget, which will now have 2 aura::Windows instead of 1 due to use of DesktopNativeWidgetAura instead of NativeWidgetAura, but I don't think that should change text labels.

msw, any ideas here?

I don't know enough about our GPU integration to offer any worthwhile ideas.
Perhaps SingleProcessMash changes something about GPU raster?
The shader usage likely also triggers a GPU operation.

Can someone post a picture of the defect? The original bug doesn't seem to have a working screen cap.
 
MenuTextMissingCrop.jpg
130 KB View Download
Some entries (above "m71.47 - Google Docs") are transparent.
The first entry above "m71.47 - Google Docs" seems to be partially transparent, there is some text in bold ("- Point bugs logged/updated").
Hovering with the cursor on top of them will fix the issue.
This happened only after we enabled GPU rasterization.
Skia unit tests pass on the device.  No crashes were observed while drawing gms either.  Here are the output of those gms.
https://drive.google.com/file/d/1Y_3hn3i6PM8brYYN1KrMYzac4CQXh_HM/view?usp=sharing
I compared the gm output to the images generated by my linux box. I don't see any egregious differences that might be related to the bug.
Cc: hcm@google.com
AFAICT this is the mali t624, midgard 2nd gen...but we're having trouble finding any lower level GPU rendering issues thus far.

Sign in to add a comment