[skrifa] autohint: CJK edges #1134
Conversation
This was similar enough to latin that changes were folded into the current code with some branches.
skrifa/src/outline/autohint/axis.rs
Outdated
| @@ -184,6 +184,10 @@ impl Segment { | |||
| edges.get(self.edge_ix.map(|ix| ix as usize)?) | |||
| } | |||
|
|
|||
| pub fn edge_next<'a>(&self, segments: &'a [Segment]) -> Option<&'a Segment> { | |||
There was a problem hiding this comment.
is there a reason this isn't next_edge? Does it mean something different?
There was a problem hiding this comment.
This returns the next segment in the parent edge. This name was copied from FT but I agree that it's ambiguous. Renamed to next_in_edge for clarity.
| ) { | ||
| axis.edges.clear(); | ||
| let scale = metrics.scale; | ||
| let top_to_bottom_hinting = if axis.dim == Axis::HORIZONTAL { | ||
| let top_to_bottom_hinting = if axis.dim == Axis::HORIZONTAL || group != ScriptGroup::Default { |
There was a problem hiding this comment.
horizontal implies not top-to-bottom makes sense. The group limitation somewhat less so. I suggest a comment.
There was a problem hiding this comment.
I'm probably being overly defensive here but the CJK code always passes constant 0 where this value is expected. I added a comment with a link to FT code
| 'segments: for segment_ix in 0..axis.segments.len() { | ||
| let segment = &axis.segments[segment_ix]; | ||
| if segment.dir != Direction::None { | ||
| continue; | ||
| } |
There was a problem hiding this comment.
Why not for (ix, segment) in axis.segments.iter().enumerate().filter(no direction)? - particularly the .iter().enumerate()
There was a problem hiding this comment.
Would that create unpleasently overlapping borrows? - if so suggest a comment noting as much
There was a problem hiding this comment.
Yes, we make mutable calls into axis below which trips up the borrow checker. Added a comment
| if segment.dir != Direction::None { | ||
| continue; | ||
| } | ||
| // Find a matching edge |
There was a problem hiding this comment.
Suggest // Find the first matching edge
There was a problem hiding this comment.
Changed to "Try to find an edge that coincides with this segment within the threshold" which better explains what's happening here
| axis.append_segment_to_edge(segment_ix, edge_ix); | ||
| // Move to next segment | ||
| continue 'segments; | ||
| } |
There was a problem hiding this comment.
At a glance this also looks possible to do via iterators? - I didn't try so perhaps not; if not perhaps add a comment explaining why.
if let Some(edge_ix) = axis.edges.iter().enumerate().filter_map(if dist < threshold retain index).find() type of deal?
| ) { | ||
| if axis.dim != Axis::VERTICAL { | ||
| // For the default script group, we only handle vertical blues |
There was a problem hiding this comment.
... FreeType does this. I added a link to the relevant code.
In reality, FT doesn't use any alignment zones in the horizontal direction but the infra exists for it. I believe this is an attempt at better hinting for vertical text but the code is permanently disabled with an #undef. After this all lands, I think it's worth considering simplifying our code and removing this altogether. Filed #1140
| // from the edge | ||
| // See <https://gitlab.freedesktop.org/freetype/freetype/-/blob/57617782464411201ce7bbc93b086c1b4d7d84a5/src/autofit/afcjk.c#L1356> | ||
| if (edge.fpos as i32 - unscaled_blue.position).abs() | ||
| > (edge.fpos as i32 - unscaled_blue.overshoot).abs() |
There was a problem hiding this comment.
I suggest adding some methods to things like ScaledBlue and Edge. For example, instead of let is_top = blue.flags & (blue_flags::TOP | blue_flags::LATIN_SUB_TOP) != 0; we could have blue.is_top(), instead of (edge.fpos as i32 - unscaled_blue.position).abs() we could have edge.distance_to(unscaled_blue) (...or would it need to be specifically unscaled_distance_to?)
This could be deferred to an issue if we don't want to address here. I feel like autohinting would be quite a bit easier to read if widely deployed.
There was a problem hiding this comment.
There's a lot that can be done to clean up the whole autohint module but I prioritized (mostly) matching the FT code in the first pass. I agree that we can do much better but I would like to defer until after we land. Filed #1141
| scale.y_scale | ||
| }; | ||
| // Initial threshold | ||
| let initial_best_dest = fixed_mul(scale.units_per_em / 40, axis_scale).min(64 / 2); | ||
| for edge in &mut axis.edges { | ||
| let mut best_blue = None; | ||
| let mut best_is_neutral = false; |
There was a problem hiding this comment.
Optional. I don't have time to try it but finding the best values for an edge smells a lot like reduce. Is it possible it would make sense to write this in terms of reduce? - if it does but we don't want to do it now feel free to defer to an issue.
There was a problem hiding this comment.
Added to #1141. This seems possible but the interior code would need to be factored out into a function to make the change worthwhile.
This was similar enough to latin that changes were folded into the current code with some branches.