make sure masters for global/glyph/layout metrics are rounded before computing deltas #1082
Conversation
to be able to use impl OtRound<f64> for <f64> googlefonts/fontations#1209
…computing deltas Fixes #1043
| // a VF at the masters' location is expected to be equivalent to building | ||
| // individual masters as static fonts. fontmake does the same, see | ||
| // https://github.com/googlefonts/fontc/issues/1043 | ||
| let value: f32 = value.into_inner().ot_round(); |
There was a problem hiding this comment.
should we ot_round to f64 and avoid f32 then f64? - I recall we need to add that to otround, maybe we should do that first?
There was a problem hiding this comment.
I already added OtRound<f64> for f64 and OtRound<f32> for f32 here:
googlefonts/fontations#1209 (unreleased yet, which is why this PR is fetching fontations from github)
Are you saying we should also add OtRound<f64> for f32? Then why not any other combinations of primitive number types? It feels to me a bit out of scope for a rounding trait that we are casting f32 as f64 because of some other API which works best with those.
yes, I mentioned Teachers-Italic.glyphs in #1043 (comment), which contains a composite with a scaled component that gets its anchors propagated as float coordinates and produces different deltas when these get rounded beforehand. I can add a subset of that one the fontc/lib.rs integration tests. |
the deltas for Teachers-Italic.glyphs' ordfeminine 'top' anchor are off-by-one because the anchor's master values are un-rounded floats. #1082 will fix this and similar others.
|
I made a test in a separate branch to show it failing on |
the deltas for Teachers-Italic.glyphs' ordfeminine 'top' anchor are off-by-one because the anchor's master values are un-rounded floats. #1082 will fix this and similar others.
Fixes #1043