acid2 has regressed

[metajack]

Current master 7541b57 has a few different problems, around the top of the smiley.

[larsbergstrom]

People are seeing problems with the eyes:
http://imgur.com/GMDntyj

[SimonSapin]

I can reproduce this bug on 83aabe3 but not on 7ece5f9. In between these, there’s only the Rust upgrade #2041.

[june0cho]

Still reproduced.
If the first line of face(https://github.com/mozilla/servo/blob/master/src/test/html/acid2.html#L128) is commented out, then it looks good.
This line has the following style:

.picture p { position: fixed; margin: 0; padding: 0; border: 0; top: 9em; left: 11em; width: 140%; max-width: 4em; height: 8px; min-height: 1em; max-height: 2mm;
 background: black; border-bottom: 0.5em yellow solid; }

[mrobinson]

The issue here seems is that the missing bits (eyes and top of the forehead) are layerized, but those layers are not placed in the correct position. They seem to be layerized because they have siblings with position:fixed. The layer origins are 0, 36 and 36, 60. I've attached a screenshot of this with layer borders turned on, the two invisible layers are not position:fixed so are rendered nowhere near the face which is much further down the page. I think that they are invisible because there is no layer content at that position on the page.

[mrobinson]

After applying the following patch applied in block.rs

         // If we got here, then we need a new layer.
         let layer_rect = self.base.position.union(&self.base.overflow);
         let size = Size2D(layer_rect.size.inline.to_nearest_px() as uint,
                           layer_rect.size.block.to_nearest_px() as uint);
-        let origin = Point2D(layer_rect.start.i.to_nearest_px() as uint,
-                             layer_rect.start.b.to_nearest_px() as uint);
+        //let origin = Point2D(layer_rect.start.i.to_nearest_px() as uint,
+        //                     layer_rect.start.b.to_nearest_px() as uint);
+        let origin = Point2D(self.base.abs_position.x.to_nearest_px() as uint,
+                             self.base.abs_position.y.to_nearest_px() as uint);

The rendering still isn't correct, but I think that since layers are always added as children of the root layer, using the relative position is incorrect. I'm not sure exactly why the forehead is still incorrect, but I'll continue to investigate.

[mrobinson]

At first I was confused as to why the Acid2 reftest is passing. The Acid2 reftest is a special non-scrolling version of Acid2. The misplaced layers are still incorrectly placed, but because tiles are 512x512, the layer's top left tile covers the face completely. This means that the contents are rendered properly, just at an unexpected place in the tile. Failures appear when activating fitted tiles from rust-layers HEAD.

Here's a screenshot, where orange is layer boundaries and blue is tile boudaries: