Most DRT results for SVG are wrong and not reliable. This is because we're not dumping meaningful information like absoluteClippedOverflowRect(), but instead use the absoluteTransform() legacy code, to map the repaintRectInLocalCoordinates() through that wrong and not CSS-aware rectangle. absoluteTransform() needs to die, and it's only used by DRT & SVGPaintServerGradient (for Cg gradient-on-text support). It turns out to be easy fixable, all we need is proper overflow clipping for RenderSVGRoot -> SVG has _special_ rules on initial clipping paths that _differ_ from CSS overflow handling rules. If we respect them properly, and do the right transformations between SVG-world (RenderSVGRoot children) and CSS Box-world (RenderSVGRoot parent objects, if any in a non-standalone SVG document) the absoluteClippedOverflowRect() for all objects is magically right, and finally repaint rects seem fixed across all boundaries. Thanks to Erics early ground-breaking integration work :-) As side effect, it's a matter of 2 LOC to fix the looong standing: SVG does not display scrollbars problem, which will be done in a follow-up patch.
Created attachment 48329 [details] Initial patch Not including LayoutTests changes, will upload them compressed, as it's over 5mb.
Created attachment 48330 [details] LayoutTests: expected.txt changes (compressed)
Created attachment 48331 [details] LayoutTests: expected.png changes
Comment on attachment 48329 [details] Initial patch Greak work Niko. I hope you checked every single test ;-)
Comment on attachment 48330 [details] LayoutTests: expected.txt changes (compressed) lgtm
Comment on attachment 48331 [details] LayoutTests: expected.png changes lgtm
Landed in r54483. Waiting for bot results...
Landed the last fix in r54492 - baselines are properly updated for all platforms.