485846 - Clipping of gradients

Status: RESOLVED WORKSFORME

(Core :: SVG, defect)

Description

[Jonathan Watt [:jwatt]]

See http://www.w3.org/Graphics/SVG/Test/20061213/htmlObjectHarness/full-pservers-grad-14-b.html

When you make the window wider and narrower you will see that some gradients get clipped. When you scroll the window up and down slowly you will also see some weird clipping.

Comment 1

[Robert Longson [:longsonr]]

I'm not seeing any of this on Windows at all.

Comment 2

[Robert Longson [:longsonr]]

I'd start looking around 2009-03-20-04 when the latest cairo upgrade landed if I was looking for a regression window.

Comment 3

[benjamin.shockley+bugzilla]

Has the same problem on 3.0.8 Mac platform:

Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:1.9.0.8) Gecko/2009032608 Firefox/3.0.8

Comment 4

[Robert Longson [:longsonr]]

Is this a regression at all then Jonathan?

Comment 5

[Jonathan Watt [:jwatt]]

Apparently this regressed between Firefox 2.0 and 3.0. It works in Firefox 2.0.20, but is broken in Firefox 3.0.6. I'm surprised nobody on Mac has reported this already, but if it's not a regression from Firefox 3.0 then I guess this isn't a strict blocker. Would be good to get it in though.

Comment 6

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Let's take it off the blocker list then.

If we can come up with a simple canvas testcase then we can probably convince Vlad it's a cairo/gfx bug...

Comment 7

[Jonathan Watt [:jwatt]]

Canvas doesn't support an equivalent of 'extendMethod' and it effectively defaults to 'pad'. Since this bug only appears with the values 'reflect' and 'repeat', we can't reproduce this using canvas.

Comment 8

[Jonathan Watt [:jwatt]]

Attachment: testcase - scroll rect in and out of view

This testcase seems somewhat related. Scrolling the blue rectangle in and out of view produces really bad rendering artifacts. Again, it only happens for 'repeat' and 'mirror' gradients.

Comment 9

[Jonathan Watt [:jwatt]]

Attachment: testcase - svg only (rects are offset 500px across, 600px down)

When I scroll this testcase in and out of view vertically and I see entire bands of the gradient not being painted. Interestingly, the stroke around the <rect>s that are filled with the gradients is always painted file, even though the gradient contents are not, as is the <line> across the <rect>s.

Horizontal scrolling is always repainted correctly. So if I've paused vertical scrolling at a point where I'm seeing lots of bands, I can horizontally scroll parts of the <rect>s in and out of view, and each part I bring horizontally back into view is then cleaned up.

Interestingly, it's only on Minefield that I see banding happening when strolling both up and down. In Firefox 3.1 beta 3 the banding only occurs when the <rect>s are scrolled into view from the bottom. When scrolled into view from the top, everything paints fine.

Comment 10

[Jonathan Watt [:jwatt]]

Attachment: testcase - svg only (rects are offset 500px across, 600px down)

Comment 11

[Jonathan Watt [:jwatt]]

Hardcoding nsSVGOuterSVGFrame::mEnableBitmapFallback to true, in other words enabling the PushGroup/PopGroupToSource calls around the nsSVGUtils::PaintFrameWithEffects call in nsSVGOuterSVGFrame::Paint, fixes this bug. I'm not sure exactly why pushing a group would cause a *larger* area to be painted.

Comment 12

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Perhaps because it creates a different kind of Quartz surface, perhaps working around a Quartz bug?

Comment 13

[Jeff Muizelaar [:jrmuizel]]

Is this still broken?

Comment 14

[Jonathan Watt [:jwatt]]

No, this works now in both trunk and 1.9.2.

Comment 15

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Probably fixed by bug 507939 and friends.

Comment 16

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Not worth checking for a regression range on an already fixed bug.