Bug 15724 – REGRESSION: svg/custom/foreign-object-skew.svg is broken

• Component: SVG
• Status: RESOLVED CONFIGURATION CHANGED

[Eric Seidel (no email)]

REGRESSION: svg/custom/foreign-object-skew.svg is broken on TOT

I'm not sure if it's related to leopard/tiger, or if it might have broken with Hyatt's recent transform changes... not sure.  but it's broken.

Putting this in SVG for now, even if the underlying cause may not be SVG.  Would be nice if someone would run bisect-builds on this.

[David Kilzer (:ddkilzer)]

bisect-builds reports:
Works: r27021  Fails: r27031

[David Kilzer (:ddkilzer)]

(In reply to comment #1)
> bisect-builds reports:
> Works: r27021  Fails: r27031

Nearly all of the changes are related to JavaScriptCore in this range.  Highly suspicious.

[David Kilzer (:ddkilzer)]

<rdar://problem/5619349>

[Sam Weinig]

I am not seeing this failure anymore. Eric, do you still see it?

[David Kilzer (:ddkilzer)]

(In reply to comment #1)
> bisect-builds reports:
> Works: r27021  Fails: r27031

WebKit nightly r27021 and r27031 and r28505 all look the same to me on Leopard 10.5.1 with Safari 3.0.4.

They're still all broken compared to the current svg/custom/foreign-object-skew-expected.png, though.

I think this is still an issue.

[David Kilzer (:ddkilzer)]

(In reply to comment #2)
> (In reply to comment #1)
> > bisect-builds reports:
> > Works: r27021  Fails: r27031
> 
> Nearly all of the changes are related to JavaScriptCore in this range.  Highly
> suspicious.

Probably not a JS issue--this was just speculation based on the range of changes.

Also, since r27021 fails to render properly on Leopard, the regression range may be incorrect as well.

[David Kilzer (:ddkilzer)]

WebKit nightly r28505 renders correctly on Tiger 10.4.11 (8S165) with Safari 3.0.4 (523.12).  (The size of the SVG is a bit bigger in ToT than in the expected pixel test results, however.  I believe it's actually more correct now--closer to 580x380.)

WebKit nightly r28505 renders INCORRECTLY on Leopard 10.5.1 (9B18) with Safari 3.0.4 (5523.10).

Looks like this is a Leopard-only issue.  (Radar is currently down so I can't update that bug.)

[David Kilzer (:ddkilzer)]

(In reply to comment #7)
> WebKit nightly r28505 renders correctly on Tiger 10.4.11 (8S165) with Safari
> 3.0.4 (523.12).  (The size of the SVG is a bit bigger in ToT than in the
> expected pixel test results, however.  I believe it's actually more correct
> now--closer to 580x380.)

I think the image may be showing an 800x600 screen, not just the SVG image, so ignore the size issue.

> WebKit nightly r28505 renders INCORRECTLY on Leopard 10.5.1 (9B18) with Safari
> 3.0.4 (5523.10).
> 
> Looks like this is a Leopard-only issue.  (Radar is currently down so I can't
> update that bug.)

Works on WebKit nightly r28503 with Safari 3.0.4 (523.12.9) on Windows XP SP2 as well.

[David Kilzer (:ddkilzer)]

Created attachment 20781 [details]
Expected results from r27021 at LayoutTests/platform/mac/svg/custom/foreign-object-skew-expected.png

This is the expected result.  Compare to the current actual results to see the regression.

[Eric Seidel (no email)]

Interesting.  Chromium/Skia does not have this problem.  Seems like it's related to the CG code path in some way.

[Nikolas Zimmermann]

*** Bug 22062 has been marked as a duplicate of this bug. ***

[Nikolas Zimmermann]

If I zoom in a few times the problem comes apparent again even on Lion with trunk :-(
Does anyone have some time for this bug? CC'ing the usual suspects.

[Stephen Chenney]

Confirmed that the issue appears in Safari on Lion. Doesn't appear in Chromium.

My guess is that this is outside of SVG rendering, as the button is a native html element. I presume that SVG foreign object is the only way to skew a html button, and that skew breaks the button drawing results.

[Stephen Chenney]

Hang on, we do see it on Mac DRT. So presumably it's Mac specific but not port specific.

[Philip Rogers]

I suggest we expand the scope of this bug, as it doesn't appear in just SVG. This also occurs in HTML and CSS transforms:

<html>
<head>
<style type="text/css"> 
div{
width:100px;
height:75px;
border:1px solid black;
}
div#div2 {
-webkit-transform:skew(30deg, 30deg); /* Safari and Chrome */
}
</style>
</head>
<body>
<div id="div2">Hello. This is a DIV element.<input type="button" value="Halo" /></div>
</body>
</html>

[Elliott Sprehn]

Philip, Your example looks exactly the same in Firefox, Safari (GC) and Chrome (Skia). What are you seeing that's unexpected?

[Philip Rogers]

(In reply to comment #16)
> Philip, Your example looks exactly the same in Firefox, Safari (GC) and Chrome (Skia). What are you seeing that's unexpected?

Just checked and we're still hitting it on ToT. Zoom in a bunch and notice how the button is longer than it should be, then click on it and it'll re-render but with the correct size.

[Philip Rogers]

Created attachment 149435 [details]
Example of the HTML version of the bug

[Philip Rogers]

(In reply to comment #18)
> Created an attachment (id=149435) [details]
> Example of the HTML version of the bug

Just a note about what this is showing: I changed the code a bit to show 3 skewed input elements then pressed on the middle one and managed to take a screenshot.

[Karl Dubost]

Created attachment 478887 [details]
testcases

The bug seems to be fixed all the tests renders the same in all browsers and as expected with regards to the description.