regression: Specific wikipedia page: Semi-donut chart shows lines on the bottom half where there shouldn't be

[staktrace]

Filing for the "first regression" reported at https://bugzilla.mozilla.org/show_bug.cgi?id=1430063 where the chart goes from looking correct to having streaky lines in the bottom half.

Narrowed down the WR regression range to this:

* c16b37dd Auto merge of #2228 - glennw:overlap-border, r=kvark
* e3d7353a Fix overlapping border radii handling.

So definitely from #2228.

[glennw]

I did some investigation of this - that bisect is not quite correct. It occurred in #2242 which merged just after that.

I can see what the problem is, but I'm not sure of the correct fix. @mrobinson could you take a look at this?

A reduced test case:

---
root:
  items:
      -
        type: "stacking-context"
        items:
          -
            bounds: [1488, 111, 356, 197]
            "clip-rect": [1488, 111, 356, 197]
            type: clip
            id: 5
            "content-size": [356, 197]
          -
            "clip-and-scroll": 5
            type: "stacking-context"
            transform: [0.98883086, 0.14904226, 0, 0, -0.14904226, 0.98883086, 0, 0, 0, 0, 1, 0, 1533.3134, 109.21605, 0, 1]
            items:
              -
                "clip-and-scroll": 5
                type: "stacking-context"
                transform: [0.7818315, 0.6234898, 0, 0, -0.6234898, 0.7818315, 0, 0, 0, 0, 1, 0, 132.98201, -64.04077, 0, 1]
                items:
                  -
                    "clip-and-scroll": 5
                    type: "stacking-context"
                    transform: [0.93087375, 0.36534107, 0, 0, -0.36534107, 0.93087375, 0, 0, 0, 0, 1, 0, 68.64584, -46.80194, 0, 1]
                    items:
                      -
                        "clip-and-scroll": 5
                        type: "stacking-context"
                        transform: [0.8262389, 0.56332004, 0, 0, -0.56332004, 0.8262389, 0, 0, 0, 0, 1, 0, 116.458824, -61.550323, 0, 1]
                        items:
                          -
                            "clip-and-scroll": 5
                            type: "stacking-context"
                            transform: [0.90096885, 0.43388373, 0, 0, -0.43388373, 0.90096885, 0, 0, 0, 0, 1, 0, 84.200554, -52.906708, 0, 1]
                            items:
                              -
                                "clip-and-scroll": 5
                                type: "stacking-context"
                                transform: [0.98883086, 0.14904226, 0, 0, -0.14904226, 0.98883086, 0, 0, 0, 0, 1, 0, 25.3134, -21.78395, 0, 1]
                                items:
                                  -
                                    "clip-and-scroll": 5
                                    type: "stacking-context"
                                    transform: [0.73305184, 0.68017274, 0, 0, -0.68017274, 0.73305184, 0, 0, 0, 0, 1, 0, 149.64511, -65.28949, 0, 1]
                                    items:
                                      -
                                        bounds: [0, 0, 316, 158]
                                        "clip-rect": [0, 0, 316, 158]
                                        "clip-and-scroll": 5
                                        type: rect
                                        color: red

After that PR, this displays as a fairly large rectangle. Before that PR, this will correctly have a clip mask applied and show as a very thin rectangle.

With current master it doesn't get any clip mask applied at all. What happens is that the code here

webrender/webrender/src/prim_store.rs

Line 1984 in c8babed

if node.screen_inner_rect.contains_rect(&combined_outer_rect) {

drops the clip source in question. It says that this clip source is irrelevant, because the clip chain outer result intersected with the prim rect is completely inside the inner rect of that clip.

But since they are incompatible coordinate systems, there's no local clip rect that applies in the vertex shader, and since this clip gets filtered out, there is no clip mask generated either.

Thoughts @mrobinson ?

[staktrace]

I did some investigation of this - that bisect is not quite correct. It occurred in #2242 which
merged just after that.

It depends on the STR you use - there are two changes in behavior reported in bug 1430063 comment 1. I filed this issue and #2293 for them. #2242 is in the regression range for #2293, so that might be what you're seeing. That being said maybe I shouldn't have filed two issues since that's probably just confusing the matter further.

[glennw]

Fixed by #2324.