REGRESSION(r27173): Web Inspector freezes beneath Image::drawPattern()

[Alexey Proskuryakov]

Steps to reproduce:
1. Open fast/xsl/xslt-mismatched-tags-in-xslt.xml.
2. Open Web Inspector.
3. Click on a stylesheet at the left.

r27173 was an Inspector fix that has apparently uncovered a pre-existing WebCore bug.

Created attachment 16937 [details]
Reduction (will hang)

The tile width gets rounded down to 0...

*** Bug 15752 has been marked as a duplicate of this bug. ***

Created attachment 16940 [details]
Clamp background tile dimensions up to 1

[Adam Roben (:aroben)]

Comment on attachment 16940 [details]
Clamp background tile dimensions up to 1

+    if (!patternTransform.isInvertible()) {
+        ASSERT_NOT_REACHED();
+        // Avoid a hang under CGContextDrawTiledImage on release builds.
+        return;
+    }

I think it would be slightly better to do:

ASSERT(!patternTransform.isInvertible());
if (!patternTransform.isInvertible)
    return;

This way the failed condition will appear on the console, instead of just "NOT REACHED".

Should the test be text-only?

r=me

(In reply to comment #4)
> I think it would be slightly better to do:
> 
> ASSERT(!patternTransform.isInvertible());
> if (!patternTransform.isInvertible)
>     return;

Actually, that would be the wrong thing to assert :-)
I think the ASSERT_NOT_REACHED style is more readable, but I'll change to the explicit (but correct) assertion because of console output.

> Should the test be text-only?

Since it's a rendering test, I feel that the pixel result gives a little better regression protection than just "didn't crash".

Landed in <http://trac.webkit.org/projects/webkit/changeset/27251>.