Bug 89502 – Overflow RenderLayers are not allocated in SVG leading to multiple crashers

• Status: NEW
• Component: Layout and Rendering

[Julien Chaffraix]

Created attachment 148399 [details]
Example test case - beware that it will crash

Following bug 75568, we added a lot of NULL-check for overflow clip layers. Incidentally this solved a lot of NULL crashers due to us NULL-checking the layer() instead of calling it directly. Bug 75568 was rolled out in bug 83954 so we are back to crashing, except that people started depending on this to work and consider this to be a regression.

The core issue is that even if requiresLayer() returns true, we are not guaranteed to have a RenderLayer as it's also determined by RenderObject::layerCreationAllowedForSubtree() (which can return false in SVG documents). One way would be to make RenderLayer work in SVG but it's likely not a good idea as RenderLayer is very CSS specific.

My view on the issue is to re-add the NULL-checks while we determine a better solution.

[Julien Chaffraix]

Created attachment 148415 [details]
Proposed change 1: open to comment if you see a better short-term fix.

[Julien Chaffraix]

Comment on attachment 148415 [details]
Proposed change 1: open to comment if you see a better short-term fix.

There is some work that would enable RenderLayer in SVG and would prevent the need for this work-around. See bug 86022 and bug 90738 for example.

This change was a work-around so it's probably better to work towards the long terms goal directly.

[Eric Seidel (no email)]

I think we have a central bug for teh fact the RenderLayer is disabled for SVG subtrees.  bug 93358 is yet another example of this.