Created attachment 206272 [details] reftest for Roman writing mode If a polygon is defined to have a zero-width segment - such as polygon(0% 0%, 0% 51%, 100% 100%, 0 100%) - then content that intersects with that segment wraps around the bounding box of the shape instead of up against the shape segment.
Created attachment 206284 [details] additional reftest The latest nightly appears to ignore empty shape segments, so the original reftest now passes. I've attached a new reftest that moves the empty segment to the midpoint of the float. Here I'd expect the inline content to start at that midline.
<rdar://problem/15958849>
What's the latest on this?
We're discussing whether this is a bug in the implementation or a bug in the spec. In either case, I think it's very low-priority - you have to create a fairly degenerate polygon in order to reproduce the problem.
As of the patch for https://bugs.webkit.org/show_bug.cgi?id=130975 we've addressed this issue. More about the zero-width segment case and the semantics of self-intersecting polygonal shapes can be found here: http://hansmuller-webkit.blogspot.com/2014/03/a-simpler-algorithm-for-css-shapes.html