623972 - border joint broken in conjunction with rotate and radius

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[Adam]

User-Agent:       Mozilla/5.0 (Windows NT 5.1; rv:2.0b8) Gecko/20100101 Firefox/4.0b8
Build Identifier: FF4b8 20101214170338

In certain cases, the border has a slight gap between the border and the box that is very noticeable.  This occurs when the box is rotated, and more pronounced when border-radius is applied.  It is present regardless of border thickness.

It seems related to this:
https://bugzilla.mozilla.org/show_bug.cgi?id=455308

But I believe this is a more general case.



Reproducible: Always

Steps to Reproduce:
Will provide sample code
Actual Results:  
Gap in border

Expected Results:  
No gap in border

Comment 1

[Adam]

Attachment: this shows border in multiple cases

Comment 2

[Adam]

Comment on attachment 502045 [details]
this shows border in multiple cases

It is important to note that the outer div with the padding IS required.

It requires any parent, grand-parent, etc to have padding.

Comment 3

[Adam]

At different zoom levels you can notice even the correct case, without transform, exhibits the gaps.

Comment 4

[Adam]

Oh, also, when used with hover, or if the rotate value is changed, it shows up fine fir the first 100ms or so, then the gaps show up.

Comment 5

[Bas Schouten (:bas.schouten)]

Do you also see this in 3.6? I think I do? Or is that because of some border usage which isn't supported in 3.6. It looks good on Windows 7 with D2D as far as I can see.

Comment 6

[Adam]

The problem also occurs in 3.6.7, WinXP, though not as bad.  Instead of the tick-tack-toe looking box, it's simply a square around the inside box.

Comment 7

[Adam]

Attachment: screenshot of the html demo in FF4b8 WinXP

Image of how it looks in FF4b8 WinXP.

Comment 8

[Adam]

On this site it is very noticeable on the "last 10" text that is in the central images.  If you also mouse over them, you'll notice the issue pops back after ~400ms.
http://unicaworld.com/

Comment 9

[Bas Schouten (:bas.schouten)]

I just realized I was testing this with beta 7 and not beta 8. This is broken in Beta 8 on windows 7 too. And is almost certainly a regression from my border rendering code changes. I should analyze this and fix it, it shouldn't be terribly complex.

Comment 10

[Bas Schouten (:bas.schouten)]

Hrm, actually, Beta 7 already had a square, just not the tick-tack-toe behavior, much like 3.6.7. So my border code did affect this, but it wasn't responsible for the initial seam, anyone else have any ideas? I'm not sure what happens when padding is applied and such.

Comment 11

[Bas Schouten (:bas.schouten)]

Hmm, I think I might actually have been wrong about this having to do with the border code at all, it might very well have been changes in layers. I'll keep owning this though until I can find out if this has anything to do with borders and who else should look into it. I'm not yet sure this should block, but the behavior seems to have gotten worse with relation to beta 7 and 3.6 so perhaps.

Comment 12

[Timothy Nikkel (:tnikkel)]

Attachment: testcase

We only unprefixed border-radius after Firefox 3.6, so to accurately compare we need to add the prefixed version of border-radius to the testcase.

Comment 13

[Timothy Nikkel (:tnikkel)]

The reason the padding is important is because it is 1.6em, which makes things not line up exactly at pixel boundaries.

The regression range for the box artifact becoming a grid is http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=baa5ae44f0ba&tochange=0ff6d5984287 which includes Bas' border changes.

Comment 14

[Joe Drew (not getting mail)]

This doesn't seem bad enough to hold the release.

Comment 15

[Adam]

It does make transform rotate, or just border-radius unusable in many cases.

Comment 16

[Hugh Nougher [:Hughman]]

Attachment: Problem visible in Aurora 18.0

Definitely related to anti aliasing since the gaps disappear when something is animated inside it.

And I am going to release this page with this problem and if anyone complains to me I'll be forwarding them to this bug.