Bug 292754

Created attachment 475172 [details] Test case Consider this pseudo-HTML (see attached HTML for a fully working test case) <div anchor-name: --a /> // A <div position-anchor: --a position-area: top left> // B <div anchor-name: --c ></div> // C </div> <div position-anchor: --c position-area: top right /> // D A is an anchor. B is anchor-positioned, and its child, C, is an anchor. D is anchor-positioned. Style/layout interleaving timeline: * First style resolution: A is discovered as an anchor. B is discovered as anchor-positioned to A. D is discovered as anchor-positioned to C. Crucially, there's an optimization where if an anchor-positioned element hasn't been resolved yet, we skip descending into its children until the next style resolution, when the anchor-positioned element has been resolved. So here, we don't discover C as an anchor (because its parent, B, is anchor-positioned and not resolved yet) * First layout: A is laid out. Post processing task links B to A, but doesn't link D to anything (because D depends on C, but the code hasn't seen C yet) * Second style resolution: B, D are marked as resolved. Since B is resolved, the code descends into B and discovers C as an anchor. BUT, since D are already marked as resolved, we don't go back and fix that C actually exists. * Second layout: B is laid out Result: the code treats C as if it doesn't exist.