1033375 - Jagged borders on triangles made using borders in css

Status: RESOLVED DUPLICATE of bug 1185636

(Core :: Graphics, defect)

Description

[Nicolas Silva [:nical]]

Attachment: Screenshot from 2014-07-02 15:46:23.png

Test case: http://css-tricks.com/snippets/css/css-triangle/

I can reproduce this on linux (nightly and stable).
The border of the triangles is jagged while it should look like a proper triangle (see attachment). Looks good in chrome.
I have seen this issue in other websites out there but this one seems like the most straightforward test case.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

I feel like this might be a regression

Comment 2

[Jonathan Kew [:jfkthame]]

This sounds familiar..... ah yes... is this the same as bug 805393 or bug 965966, maybe?

Comment 3

[Jonathan Kew [:jfkthame]]

Or perhaps not; those bugs sounds more like they're just about antialiasing (or lack thereof), whereas here there are distinct periodic glitches.

Comment 4

[Saurabh Anand [:sawrubh]]

Attachment: Arrow in Chrome

I'm not sure if this is a related issue but I see similar issues with the downward arrow in the YouTube menu. In Chrome, it looks fine while not so in Firefox. Attaching screenshots.

Comment 5

[Saurabh Anand [:sawrubh]]

Attachment: Arrow in Nightly

This is how it looks in Nightly.

Comment 6

[Saurabh Anand [:sawrubh]]

This seems to have been fixed somehow. Will close if someone can confirm this being fixed.

Comment 7

[Saurabh Anand [:sawrubh]]

Never mind, I was confused. I didn't remember this was an issue on Linux. It was working for me on Windows :P

Comment 8

[Norman Paschke]

Not working on Windows if hardware acceleration is enabled.

Comment 9

[Lee Salzman [:lsalzman]]

Attachment: nudge simple linear gradients with hard stops to half-pixel gradient

Apparently this is down to how Cairo backends that rely on the pixman fallback for gradients work. If this fallback gets used, then you get the pixel drop-outs observed.

The CSS border rendering code in the triangle border test case was basically doing that - linear gradient with 2 stops set with offset 0.5.

This patch works around this by noticing in the Cairo draw target is such a gradient was requesting, and nudging the offsets to occupy half a pixel to more or less anti-alias the gradient

Comment 10

[Lee Salzman [:lsalzman]]

Attachment: nudge simple linear gradients with hard stops to half-pixel gradient

Added some comments to document how the fix works and point back at the bug for posterity's sake (otherwise changed no code in patch)

Comment 11

[Lee Salzman [:lsalzman]]

Try results: https://treeherder.mozilla.org/#/jobs?repo=try&revision=a096bde6da55

Comment 12

[Lee Salzman [:lsalzman]]

Oops, some reftest failures on zero length gradients. Working on fix...

Comment 13

[Lee Salzman [:lsalzman]]

Attachment: nudge simple linear gradients with hard stops to half-pixel gradient

Fix zero-length gradient reftests

Try: https://treeherder.mozilla.org/#/jobs?repo=try&revision=06a689e23285

Comment 14

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/9e8d2b0dd0ce

Comment 15

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/9e8d2b0dd0ce

Comment 16

[Lee Salzman [:lsalzman]]

Intentionally reverted patch as part of bug 1185636, due to the regression noticed in bug 1183506. Once the new border rendering in bug 1185636 lands, this will be addressed.