bug2067555.yaml - mozsearch

Enable keyboard shortcuts

# bug 2067555: a text run's clip must be quantised the way its glyphs are

# quantised, per axis. The glyph pen is rounded to a whole device pixel on the

# non-sub-pixel axis, so a clip left exact on that axis disagrees with the grid

# the glyphs landed on - and the shader's clamp is a hard pixel-centre test, so

# the disagreement costs a whole device row of ink.

# Gecko produces exactly this: it sizes a line box from the font's metrics and

# puts the overflow clip on the content-box edge, so a descender's last row of

# ink ends flush with the clip, and a box at a fractional device position puts

# that clip edge mid-row.

# The Ahem block is solid, so the ink boundary is unambiguous: it covers device

# rows 17..70. The ancestor clip ends at 40.5 and must round to 41.0, the same

# grid the glyph pen rounds to, keeping row 40. Left exact at 40.5 the clamp

# resolves at row 40's centre - exactly 40.5 - and drops the row entirely.

# The clip edge has to sit on the .5 tie: exact and nearest differ ONLY there,

# which is why this bug reproduces at one sub-pixel phase in twenty (and why

# chrome UI, whose boxes land on .0 and .5 at DPR 1, hits it). Do not "tidy" the

# 40.5 to another value - the test stops discriminating.

# The clip bounds x to 0..50 so the covered span is exactly x 10..50 and does not

# depend on the rasterized glyph width, which for Ahem 55 at size 40 is wider

# than the em box.

root:

  items:

    - type: rect

      bounds: [0, 0, 200, 120]

      color: white

    - type: clip

      id: 2

      bounds: [0, 0, 50, 40.5]

    - type: clip-chain

      id: 3

      clips: [2]

    - bounds: [0, 0, 120, 120]

      glyphs: [55]

      offsets: [10, 60]

      size: 40

      color: [0, 0, 0, 1]

      font: "Ahem.ttf"

      clip-chain: 3