♻️ Share duplicated CJK layout helpers across both axes

Several CJK passes had a copy per writing direction. Merge them without
changing their results:

- allocate_bounded_expansion and bounded_expansion divide justify and
  distribute slack for both axes. Each axis still collects its own gaps;
  the horizontal one checks no-break controls over the source range and
  the vertical one at cell edges, as before.
- PolicyBoundaries holds the break and emergency rules at each grapheme
  of a paragraph. The horizontal overlong-word and atomic-overflow passes
  and the vertical pairwise column planner read it. relax_overlong_words
  moves next to retain_atomic_overflow and shares annotation_ranges.
- TextLayoutPolicy::pair_spacing_em gives the compression and script gap
  that both axes apply with their own caps.
- cjk_chars keeps the Han, kana, hangul and punctuation ranges. Ranges
  that differ on purpose stay separate there, with a note on why.
- OverhangClaims tracks the ruby overhang an earlier reading takes.
- horizontal_zhuyin_glyphs places Zhuyin for position data and painting.
- paint_horizontal_annotations paints warichu, emphasis and ruby; the
  emoji overlay keeps its place between the marks and the ruby.

AI-assisted-by: claude-opus-5-5

render-wasm/src/render/fonts.rs

@@ -2,6 +2,7 @@ use skia_safe::{self as skia, textlayout, Font, FontMgr};
 use std::collections::{BTreeSet, HashMap, HashSet};
 
 use crate::error::{Error, Result};
+use crate::shapes::cjk_chars::{is_hangul, is_kana};
 use crate::shapes::text_policy::LanguageTag;
 use crate::shapes::{FontFamily, FontStyle};
 use crate::uuid::Uuid;
@@ -71,14 +72,6 @@ impl CjkFace {
     }
 }
 
-fn is_kana(c: char) -> bool {
-    matches!(c, '\u{3040}'..='\u{30FF}' | '\u{31F0}'..='\u{31FF}' | '\u{FF66}'..='\u{FF9F}')
-}
-
-fn is_hangul(c: char) -> bool {
-    matches!(c, '\u{1100}'..='\u{11FF}' | '\u{3130}'..='\u{318F}' | '\u{AC00}'..='\u{D7AF}')
-}
-
 /// Families that take part in character fallback, with the CJK region of
 /// those that draw Han.
 #[derive(Debug, Default)]

render-wasm/src/render/text.rs

@@ -575,6 +575,34 @@ pub fn try_paint_from_layout_cache(
     Ok(false)
 }
 
+/// Paint the warichu, emphasis marks and ruby of the laid-out horizontal
+/// paragraph `index` at `origin`, with the paint of this pass.
+fn paint_horizontal_annotations(
+    canvas: &Canvas,
+    text_content: &TextContent,
+    index: usize,
+    laid_out: &LayoutParagraph,
+    origin: (f32, f32),
+) {
+    paint_horizontal_notes_and_marks(canvas, text_content, index, laid_out, origin);
+    crate::shapes::paint_horizontal_ruby(canvas, text_content, index, laid_out, origin.0, origin.1);
+}
+
+/// The warichu and emphasis marks of `paint_horizontal_annotations`.
+fn paint_horizontal_notes_and_marks(
+    canvas: &Canvas,
+    text_content: &TextContent,
+    index: usize,
+    laid_out: &LayoutParagraph,
+    (x, y): (f32, f32),
+) {
+    let plans = text_content.horizontal_plans();
+    if let (Some(paragraph), Some(plan)) = (text_content.paragraphs().get(index), plans.get(index))
+    {
+        crate::shapes::paint_horizontal_warichu(canvas, paragraph, plan, laid_out, x, y);
+        crate::shapes::paint_horizontal_emphasis(canvas, paragraph, plan, laid_out, x, y);
+    }
+}
 fn paint_from_cached_layout(canvas: &Canvas, shape: &Shape, text_content: &TextContent) {
     let selrect = shape.selrect();
     // Absolute image/gradient shaders were baked at `layout_paint_origin`. Paint
@@ -612,33 +640,10 @@ fn paint_from_cached_layout(canvas: &Canvas, shape: &Shape, text_content: &TextC
             .zip(text_content.paragraphs().get(index))
             .map_or(0.0, |(plan, model)| plan.x_offset(model, selrect.width()));
         paragraph.paint(canvas, (x, y_accum));
+        paint_horizontal_annotations(canvas, text_content, index, paragraph, (x, y_accum));
         if let (Some(text_paragraph), Some(plan)) =
             (text_content.paragraphs().get(index), plans.get(index))
         {
-            crate::shapes::paint_horizontal_warichu(
-                canvas,
-                text_paragraph,
-                plan,
-                paragraph,
-                x,
-                y_accum,
-            );
-            crate::shapes::paint_horizontal_emphasis(
-                canvas,
-                text_paragraph,
-                plan,
-                paragraph,
-                x,
-                y_accum,
-            );
-            crate::shapes::paint_horizontal_ruby(
-                canvas,
-                text_content,
-                index,
-                paragraph,
-                x,
-                y_accum,
-            );
             if draw_decorations {
                 for deco in
                     decoration_segments(paragraph, text_paragraph, &plan.offsets, x, y_accum)
@@ -791,45 +796,23 @@ fn paint_text_with_emoji_overlay(
 
     let mut layout_info =
         calculate_text_layout_data(shape, text_content, paragraph_builder_groups, true);
-    let plans = text_content.horizontal_plans();
 
     for para in &mut layout_info.paragraphs {
         para.paragraph.paint(canvas, (para.x, para.y));
-
-        if let (Some(source_paragraph), Some(plan)) = (
-            text_content.paragraphs().get(para.source_paragraph),
-            plans.get(para.source_paragraph),
-        ) {
-            crate::shapes::paint_horizontal_warichu(
-                canvas,
-                source_paragraph,
-                plan,
-                &para.paragraph,
-                para.x,
-                para.y,
-            );
-            crate::shapes::paint_horizontal_emphasis(
-                canvas,
-                source_paragraph,
-                plan,
-                &para.paragraph,
-                para.x,
-                para.y,
-            );
-        }
-
+        let (index, origin) = (para.source_paragraph, (para.x, para.y));
+        paint_horizontal_notes_and_marks(canvas, text_content, index, &para.paragraph, origin);
+        // The emoji overlay goes over the marks and under the ruby, which
+        // takes the paint of this pass like warichu and emphasis.
         if overlay_emoji {
             paint_emoji_overlay(canvas, para);
         }
-
-        // Like warichu and emphasis, ruby takes the paint of this pass.
         crate::shapes::paint_horizontal_ruby(
             canvas,
             text_content,
-            para.source_paragraph,
+            index,
             &para.paragraph,
-            para.x,
-            para.y,
+            origin.0,
+            origin.1,
         );
 
         for deco in &para.decorations {

render-wasm/src/shapes.rs

@@ -11,6 +11,7 @@ use std::iter::once;
 mod blend;
 mod blurs;
 mod bools;
+pub mod cjk_chars;
 mod corners;
 mod fills;
 mod fonts;

render-wasm/src/shapes/cjk_chars.rs

@@ -0,0 +1,64 @@
+//! Character ranges and punctuation sets of CJK text, shared by the layout
+//! policies, the JLREQ classes, Zhuyin and font fallback. Where two users
+//! need different ranges on purpose, each keeps its own function here with
+//! a note on why. The JLREQ class tables (`jlreq`) and the vertical upright
+//! table (`text_shaping::is_upright_char`) follow their own references and
+//! stay with their users.
+
+/// Han ideographs of the BMP unified blocks: Extension A and the URO.
+pub fn is_bmp_unified_han(c: char) -> bool {
+    matches!(u32::from(c), 0x3400..=0x4DBF | 0x4E00..=0x9FFF)
+}
+
+/// Han ideographs for the Chinese and Korean presets: the BMP unified
+/// blocks, the compatibility ideographs and Extensions B to H.
+pub fn is_han(c: char) -> bool {
+    is_bmp_unified_han(c) || matches!(u32::from(c), 0xF900..=0xFAFF | 0x20000..=0x323AF)
+}
+
+/// A base a Zhuyin reading may annotate. Besides `is_han`, it takes the
+/// Yijing hexagram symbols (U+4DC0..U+4DFF) between Extension A and the URO,
+/// which the Zhuyin check has always accepted, so saved readings keep their
+/// mode.
+pub fn is_zhuyin_base(c: char) -> bool {
+    is_han(c) || matches!(u32::from(c), 0x4DC0..=0x4DFF)
+}
+
+/// Assigned hiragana, including the voiced marks and the iteration marks.
+pub fn is_hiragana(c: char) -> bool {
+    matches!(u32::from(c), 0x3041..=0x309F)
+}
+
+/// Katakana, its phonetic extensions and half-width katakana.
+pub fn is_katakana(c: char) -> bool {
+    matches!(u32::from(c), 0x30A0..=0x30FF | 0x31F0..=0x31FF | 0xFF66..=0xFF9F)
+}
+
+/// Kana that mark untagged text as Japanese for font choice: the whole
+/// Hiragana block (with its unassigned U+3040) and `is_katakana`.
+pub fn is_kana(c: char) -> bool {
+    u32::from(c) == 0x3040 || is_hiragana(c) || is_katakana(c)
+}
+
+/// Hangul syllables, conjoining jamo and compatibility jamo.
+pub fn is_hangul(c: char) -> bool {
+    matches!(u32::from(c), 0x1100..=0x11FF | 0x3130..=0x318F | 0xAC00..=0xD7AF)
+}
+
+/// Full-width opening brackets with a removable leading half-em.
+pub const FULLWIDTH_OPENING_BRACKETS: &str = "（［｛〈《「『【〔〖〘〚";
+
+/// Full-width closing brackets with a removable trailing half-em.
+pub const FULLWIDTH_CLOSING_BRACKETS: &str = "）］｝〉》」』】〕〗〙〛";
+
+/// Full-width commas and full stops.
+pub const COMMAS_AND_FULL_STOPS: &str = "，。、．";
+
+/// CJK and ASCII punctuation, brackets and quotes. A comma or full stop next
+/// to one of these does not hang alone.
+pub fn is_punctuation(c: char) -> bool {
+    COMMAS_AND_FULL_STOPS.contains(c)
+        || FULLWIDTH_OPENING_BRACKETS.contains(c)
+        || FULLWIDTH_CLOSING_BRACKETS.contains(c)
+        || "！？：；…‥—―‘’“”.,;!?:()[]{}".contains(c)
+}

render-wasm/src/shapes/jlreq.rs

@@ -6,6 +6,8 @@
 //! characters, and callers assign the virtual classes when building
 //! reference marks, ruby, grouped numerals, warichu, or tate-chu-yoko.
 
+use super::cjk_chars::{is_bmp_unified_han, is_hiragana, is_katakana};
+
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 #[repr(u8)]
 pub enum JlreqClass {
@@ -386,6 +388,8 @@ pub const fn pair_rule(before: JlreqClass, after: JlreqClass) -> PairRule {
     PAIR_RULES[before.index()][after.index()]
 }
 
+// JLREQ Appendix A class tables. They are wider than the `cjk_chars`
+// punctuation sets the pairwise policies use.
 const OPENING_BRACKETS: &str = "（〔［｛〈《「『【〖〘〚‘“〝«⦅｟｢";
 const CLOSING_BRACKETS: &str = "）〕］｝〉》」』】〗〙〛’”〟〞»⦆｠｣";
 const HYPHENS: &str = "‐゠–〜～";
@@ -408,7 +412,7 @@ const MATH_OPERATORS: &str = "＋+－−-÷×±∓∗∙√∫∬∭∑∏";
 pub fn classify(c: char) -> JlreqClass {
     // Common letters skip the table scans.
     match c {
-        '\u{4E00}'..='\u{9FFF}' | '\u{3400}'..='\u{4DBF}' => JlreqClass::Ideographic,
+        c if is_bmp_unified_han(c) => JlreqClass::Ideographic,
         'a'..='z' | 'A'..='Z' => JlreqClass::Western,
         _ => classify_by_tables(c),
     }
@@ -517,14 +521,8 @@ pub fn is_cjk_text_char(c: char) -> bool {
         || matches!(u32::from(c), 0x3000..=0x30FF | 0x31F0..=0x31FF | 0xFF00..=0xFFEF)
 }
 
-fn is_hiragana(c: char) -> bool {
-    matches!(u32::from(c), 0x3041..=0x309F)
-}
-
-fn is_katakana(c: char) -> bool {
-    matches!(u32::from(c), 0x30A0..=0x30FF | 0x31F0..=0x31FF | 0xFF66..=0xFF9F)
-}
-
+/// JLREQ cl-19 ideographs. Wider than `cjk_chars::is_han` (radicals, strokes,
+/// the compatibility supplement and a few marks), as JLREQ Appendix A lists.
 fn is_ideographic(c: char) -> bool {
     matches!(u32::from(c),
         0x2E80..=0x2FDF

render-wasm/src/shapes/line_breaks.rs

@@ -42,6 +42,70 @@ use super::jlreq::{classify, pair_rule, JlreqClass};
 use super::text_policy::TextLayoutPolicy;
 use unicode_segmentation::UnicodeSegmentation;
 
+/// One grapheme of a paragraph's text with the policy's rules for the
+/// boundary before it. The first grapheme allows both kinds of break.
+#[derive(Debug, Clone, Copy, PartialEq)]
+pub struct PolicyBoundary {
+    /// UTF-16 offset of the grapheme in the text.
+    pub utf16: usize,
+    /// First character of the grapheme, which the policy rules read.
+    pub first: char,
+    /// `TextLayoutPolicy::break_allowed` before this grapheme.
+    pub allowed: bool,
+    /// `TextLayoutPolicy::emergency_break_allowed` before this grapheme.
+    pub emergency: bool,
+}
+
+/// Break boundaries of a whole paragraph text (all spans, transformed), so
+/// font and style runs cannot split a grapheme or create a boundary.
+#[derive(Debug, Clone, Default, PartialEq)]
+pub struct PolicyBoundaries {
+    pub graphemes: Vec<PolicyBoundary>,
+    /// UTF-16 length of the text.
+    pub length: usize,
+}
+
+impl PolicyBoundaries {
+    pub fn new(text: &str, policy: TextLayoutPolicy) -> Self {
+        let mut length = 0;
+        let mut previous: Option<char> = None;
+        let graphemes = text
+            .graphemes(true)
+            .map(|grapheme| {
+                let first = grapheme.chars().next().unwrap();
+                let boundary = PolicyBoundary {
+                    utf16: length,
+                    first,
+                    allowed: previous.is_none_or(|before| policy.break_allowed(before, first)),
+                    emergency: previous
+                        .is_none_or(|before| policy.emergency_break_allowed(before, first)),
+                };
+                previous = Some(first);
+                length += grapheme.encode_utf16().count();
+                boundary
+            })
+            .collect();
+        Self { graphemes, length }
+    }
+
+    /// (allowed, emergency) at a UTF-16 offset: a grapheme start, or the
+    /// text end, which allows both. `None` inside a grapheme.
+    pub fn at(&self, utf16: usize) -> Option<(bool, bool)> {
+        if utf16 == self.length {
+            return Some((true, true));
+        }
+        self.graphemes
+            .binary_search_by_key(&utf16, |g| g.utf16)
+            .ok()
+            .map(|index| {
+                (
+                    self.graphemes[index].allowed,
+                    self.graphemes[index].emergency,
+                )
+            })
+    }
+}
+
 /// Translates between original UTF-16 offsets (the source-of-truth span text)
 /// and layout-text UTF-16 offsets (the text handed to skparagraph).
 #[derive(Debug, Clone, Default, PartialEq)]

render-wasm/src/shapes/text_allocation.rs

@@ -1,6 +1,8 @@
 //! Equal division of an adjustment over capped opportunities, shared by the
 //! horizontal and vertical justification and squeezing passes.
 
+use crate::shapes::text_policy::TextLayoutPolicy;
+
 /// Amounts below this count as zero when dividing adjustments.
 const ALLOCATION_EPSILON: f32 = 0.0001;
 
@@ -38,3 +40,45 @@ pub(crate) fn capped_equal_allocations(
     }
     allocations
 }
+
+/// Priority and pixel cap of the justification gap between `before` and
+/// `after`, at the smaller adjacent font size `em`, or `None` when the gap
+/// takes no expansion. Distribution (`distribute_budget`, the line or column
+/// budget) fills every gap at one priority, capped only by that budget;
+/// otherwise the policy bounds the gap (`TextLayoutPolicy::expansion_em`).
+pub(crate) fn bounded_expansion(
+    policy: TextLayoutPolicy,
+    before: char,
+    after: char,
+    em: f32,
+    distribute_budget: Option<f32>,
+) -> Option<(u8, f32)> {
+    let (priority, capacity) = match distribute_budget {
+        Some(budget) => (1, budget / em.max(0.01)),
+        None => policy.expansion_em(before, after),
+    };
+    (priority > 0).then_some((priority, capacity * em))
+}
+
+/// Divide `slack` over bounded justification `opportunities` (index,
+/// priority, cap in px): priority 1 first, then priority 2 with what is
+/// left, equally within a priority. Returns (index, amount) per allocation.
+pub(crate) fn allocate_bounded_expansion(
+    opportunities: &[(usize, u8, f32)],
+    slack: f32,
+) -> Vec<(usize, f32)> {
+    let mut remaining = slack;
+    let mut allocations = Vec::new();
+    for priority in 1..=2 {
+        let caps: Vec<_> = opportunities
+            .iter()
+            .filter(|(_, p, _)| *p == priority)
+            .map(|(index, _, cap)| (*index, *cap))
+            .collect();
+        for (index, amount) in capped_equal_allocations(&caps, remaining) {
+            allocations.push((index, amount));
+            remaining -= amount;
+        }
+    }
+    allocations
+}

render-wasm/src/shapes/text_horizontal/justify.rs

@@ -5,7 +5,9 @@
 // character. The other policies justify over their pairwise expansion
 // opportunities with bounded amounts.
 
-use crate::shapes::text_allocation::capped_equal_allocations;
+use crate::shapes::text_allocation::{
+    allocate_bounded_expansion, bounded_expansion, capped_equal_allocations,
+};
 use crate::shapes::text_paragraph::LayoutParagraph;
 
 use super::*;
@@ -118,7 +120,7 @@ pub(crate) fn horizontal_pairwise_justify_spacing(
         lines.len().saturating_sub(1)
     };
     for line in lines.iter().take(count) {
-        let mut remaining = (width - line.width as f32 - JUSTIFY_EPSILON).max(0.0);
+        let remaining = (width - line.width as f32 - JUSTIFY_EPSILON).max(0.0);
         let mut opportunities = Vec::new();
         let start = chars.partition_point(|c| c.utf16 < line.start_index);
         let end = chars.partition_point(|c| c.utf16 < line.end_excluding_whitespaces);
@@ -148,30 +150,16 @@ pub(crate) fn horizontal_pairwise_justify_spacing(
             {
                 continue;
             }
-            let (priority, capacity) = if distribute {
-                (1, width / before.font_size.min(after.font_size).max(0.01))
-            } else {
-                policy.expansion_em(before.ch, after.ch)
-            };
-            if priority > 0 {
-                opportunities.push((
-                    index,
-                    priority,
-                    capacity * before.font_size.min(after.font_size),
-                ));
+            let em = before.font_size.min(after.font_size);
+            if let Some((priority, cap)) =
+                bounded_expansion(policy, before.ch, after.ch, em, distribute.then_some(width))
+            {
+                opportunities.push((index, priority, cap));
             }
         }
-        for priority in 1..=2 {
-            let caps: Vec<_> = opportunities
-                .iter()
-                .filter(|(_, p, _)| *p == priority)
-                .map(|(i, _, cap)| (*i, *cap))
-                .collect();
-            for (index, amount) in capped_equal_allocations(&caps, remaining) {
-                let ch = &chars[index];
-                result[ch.span].push((ch.index, amount));
-                remaining -= amount;
-            }
+        for (index, amount) in allocate_bounded_expansion(&opportunities, remaining) {
+            let ch = &chars[index];
+            result[ch.span].push((ch.index, amount));
         }
     }
     result

render-wasm/src/shapes/text_horizontal/mod.rs

@@ -148,14 +148,16 @@ fn horizontal_pairwise_spacing(paragraph: &Paragraph, texts: &[String]) -> Vec<V
                 let b = &paragraph.children()[span];
                 if !a.has_ruby() && !b.has_ruby() && !a.is_warichu() && !b.is_warichu() {
                     let em = a.font_size.min(b.font_size);
+                    let spacing = policy.pair_spacing_em(before, ch, false);
+                    // `palt` already sets punctuation proportionally.
                     let compression = if a.font_features == FontFeatures::Palt
                         || b.font_features == FontFeatures::Palt
                     {
                         0.0
                     } else {
-                        policy.compression_em(before, ch, false)
+                        spacing.compression
                     };
-                    let amount = (compression - policy.script_spacing_em(before, ch)) * em;
+                    let amount = (compression - spacing.script_gap) * em;
                     if amount != 0.0 {
                         result[before_span].push((before_index, amount));
                     }
@@ -354,113 +356,15 @@ impl HorizontalParagraphPlan {
         Self::with_emergency_breaks(paragraph, &[])
     }
 
-    /// SkParagraph's emergency fallback can ignore punctuation joins inside
-    /// an overlong word. Measure whole units once, then reopen only safe
-    /// grapheme boundaries of units that cannot fit an empty line. The
-    /// replacement markers have identical UTF-8/UTF-16 lengths, so every
-    /// source/builder offset remains identical to the unrelaxed plan.
-    pub(crate) fn relax_overlong_words(
-        &mut self,
-        paragraph: &Paragraph,
-        width: f32,
-        measure: impl FnOnce(&Self) -> LayoutParagraph,
-    ) {
-        use unicode_segmentation::UnicodeSegmentation;
-        let policy = paragraph.layout_policy();
-        if !policy.reopens_overlong_words()
-            || !width.is_finite()
-            || width <= 0.0
-            || width == f32::MAX
-        {
-            return;
-        }
-        let source: String = paragraph
-            .children()
-            .iter()
-            .map(TextSpan::apply_text_transform)
-            .collect();
-        let mut utf16 = 0;
-        let graphemes: Vec<_> = source
-            .graphemes(true)
-            .map(|g| {
-                let at = utf16;
-                utf16 += g.encode_utf16().count();
-                (at, g.chars().next().unwrap())
-            })
-            .collect();
-        let mut offset = 0;
-        let protected: Vec<_> = paragraph
-            .children()
-            .iter()
-            .filter_map(|span| {
-                let start = offset;
-                offset += span.apply_text_transform().encode_utf16().count();
-                (span.has_ruby() || span.is_warichu()).then_some(start..offset)
-            })
-            .collect();
-        let mut units = Vec::new();
-        let mut start = 0;
-        for at in 1..graphemes.len() {
-            if policy.break_allowed(graphemes[at - 1].1, graphemes[at].1) {
-                if at - start > 1 {
-                    units.push(start..at);
-                }
-                start = at;
-            }
-        }
-        if graphemes.len() - start > 1 {
-            units.push(start..graphemes.len());
-        }
-        if units.is_empty() {
-            return;
-        }
-        let measured = measure(self);
-        let line_starts: Vec<_> = measured
-            .get_line_metrics()
-            .iter()
-            .skip(1)
-            .map(|line| line.start_index)
-            .collect();
-        let mut emergency = Vec::new();
-        for unit in units {
-            let first = graphemes[unit.start].0;
-            let last = graphemes.get(unit.end).map_or(utf16, |g| g.0);
-            let from = self.offsets.transformed_to_builder(first);
-            let to = self.offsets.transformed_to_builder(last);
-            let split = line_starts.partition_point(|at| *at <= from);
-            if line_starts.get(split).is_none_or(|at| *at >= to) {
-                continue;
-            }
-            let extent: f32 = measured
-                .get_rects_for_range(from..to, RectHeightStyle::Tight, RectWidthStyle::Tight)
-                .iter()
-                .map(|box_| box_.rect.width())
-                .sum();
-            if extent <= width + 0.01 {
-                continue;
-            }
-            for at in unit.start + 1..unit.end {
-                let (position, after) = graphemes[at];
-                let before = graphemes[at - 1].1;
-                if policy.emergency_break_allowed(before, after)
-                    && protected
-                        .get(protected.partition_point(|r| r.end <= position))
-                        .is_none_or(|r| position <= r.start)
-                {
-                    emergency.push(position);
-                }
-            }
-        }
-        if !emergency.is_empty() {
-            *self = Self::with_emergency_breaks(paragraph, &emergency);
-        }
-    }
-
-    fn with_emergency_breaks(paragraph: &Paragraph, emergency: &[usize]) -> Self {
+    pub(super) fn with_emergency_breaks(paragraph: &Paragraph, emergency: &[usize]) -> Self {
         Self::with_breaks(paragraph, emergency, &[])
     }
 
-    fn with_breaks(paragraph: &Paragraph, emergency: &[usize], forced: &[usize]) -> Self {
+    pub(super) fn with_breaks(
+        paragraph: &Paragraph,
+        emergency: &[usize],
+        forced: &[usize],
+    ) -> Self {
         let (texts, offset_map) = layout_span_texts_with_breaks(paragraph, emergency, forced);
         let pair_spacing = horizontal_pair_spacing(paragraph, &texts);
         let ruby_spacing = horizontal_ruby_spacing(paragraph, &texts, &offset_map);

render-wasm/src/shapes/text_horizontal/overflow.rs

@@ -1,9 +1,119 @@
 //! Repair SkParagraph's emergency splits without changing the wrap budget.
 
 use super::*;
+use crate::shapes::line_breaks::PolicyBoundaries;
 use crate::shapes::text_paragraph::LayoutParagraph;
 
+/// The paragraph's span texts after text transform, joined.
+fn transformed_text(paragraph: &Paragraph) -> String {
+    paragraph
+        .children()
+        .iter()
+        .map(TextSpan::apply_text_transform)
+        .collect()
+}
+
+/// UTF-16 ranges, in the transformed text, of the spans whose ruby or
+/// warichu must stay whole.
+fn annotation_ranges(paragraph: &Paragraph) -> Vec<std::ops::Range<usize>> {
+    let mut offset = 0;
+    paragraph
+        .children()
+        .iter()
+        .filter_map(|span| {
+            let start = offset;
+            offset += span.apply_text_transform().encode_utf16().count();
+            (span.has_ruby() || span.is_warichu()).then_some(start..offset)
+        })
+        .collect()
+}
+
+/// True when `position` falls strictly inside one of the ascending
+/// `annotations` ranges.
+fn within_annotation(annotations: &[std::ops::Range<usize>], position: usize) -> bool {
+    annotations
+        .get(annotations.partition_point(|r| r.end <= position))
+        .is_some_and(|r| r.start < position)
+}
+
+/// A finite, real wrap width; auto-width and unbounded layout skip repairs.
+fn is_wrap_width(width: f32) -> bool {
+    width.is_finite() && width > 0.0 && width != f32::MAX
+}
+
 impl HorizontalParagraphPlan {
+    /// SkParagraph's emergency fallback can ignore punctuation joins inside
+    /// an overlong word. Measure whole units once, then reopen only safe
+    /// grapheme boundaries of units that cannot fit an empty line. The
+    /// replacement markers have identical UTF-8/UTF-16 lengths, so every
+    /// source/builder offset remains identical to the unrelaxed plan.
+    pub(crate) fn relax_overlong_words(
+        &mut self,
+        paragraph: &Paragraph,
+        width: f32,
+        measure: impl FnOnce(&Self) -> LayoutParagraph,
+    ) {
+        let policy = paragraph.layout_policy();
+        if !policy.reopens_overlong_words() || !is_wrap_width(width) {
+            return;
+        }
+        let boundaries = PolicyBoundaries::new(&transformed_text(paragraph), policy);
+        let graphemes = &boundaries.graphemes;
+        let annotations = annotation_ranges(paragraph);
+        let mut units = Vec::new();
+        let mut start = 0;
+        for (at, grapheme) in graphemes.iter().enumerate().skip(1) {
+            if grapheme.allowed {
+                if at - start > 1 {
+                    units.push(start..at);
+                }
+                start = at;
+            }
+        }
+        if graphemes.len() - start > 1 {
+            units.push(start..graphemes.len());
+        }
+        if units.is_empty() {
+            return;
+        }
+        let measured = measure(self);
+        let line_starts: Vec<_> = measured
+            .get_line_metrics()
+            .iter()
+            .skip(1)
+            .map(|line| line.start_index)
+            .collect();
+        let mut emergency = Vec::new();
+        for unit in units {
+            let first = graphemes[unit.start].utf16;
+            let last = graphemes
+                .get(unit.end)
+                .map_or(boundaries.length, |g| g.utf16);
+            let from = self.offsets.transformed_to_builder(first);
+            let to = self.offsets.transformed_to_builder(last);
+            let split = line_starts.partition_point(|at| *at <= from);
+            if line_starts.get(split).is_none_or(|at| *at >= to) {
+                continue;
+            }
+            let extent: f32 = measured
+                .get_rects_for_range(from..to, RectHeightStyle::Tight, RectWidthStyle::Tight)
+                .iter()
+                .map(|box_| box_.rect.width())
+                .sum();
+            if extent <= width + 0.01 {
+                continue;
+            }
+            for grapheme in &graphemes[unit.start + 1..unit.end] {
+                if grapheme.emergency && !within_annotation(&annotations, grapheme.utf16) {
+                    emergency.push(grapheme.utf16);
+                }
+            }
+        }
+        if !emergency.is_empty() {
+            *self = Self::with_emergency_breaks(paragraph, &emergency);
+        }
+    }
+
     /// Keep an indivisible unit together even when Skia's last-resort wrapping
     /// ignores its joiners. Isolate overflowing units, measure the surrounding
     /// text at the original width, and freeze those legal line starts before
@@ -14,57 +124,32 @@ impl HorizontalParagraphPlan {
         width: f32,
         mut measure: impl FnMut(&Self, f32) -> LayoutParagraph,
     ) {
-        use unicode_segmentation::UnicodeSegmentation;
         let policy = paragraph.layout_policy();
-        if !policy.keeps_overlong_units() || !width.is_finite() || width <= 0.0 || width == f32::MAX
-        {
+        if !policy.keeps_overlong_units() || !is_wrap_width(width) {
             return;
         }
-        let source: String = paragraph
-            .children()
-            .iter()
-            .map(TextSpan::apply_text_transform)
-            .collect();
-        let mut length = 0;
-        let graphemes: Vec<_> = source
-            .graphemes(true)
-            .map(|g| {
-                let start = length;
-                length += g.encode_utf16().count();
-                (start, g.chars().next().unwrap())
-            })
-            .collect();
-        let mut offset = 0;
-        let protected: Vec<_> = paragraph
-            .children()
-            .iter()
-            .filter_map(|span| {
-                let start = offset;
-                offset += span.apply_text_transform().encode_utf16().count();
-                (span.has_ruby() || span.is_warichu()).then_some(start..offset)
-            })
-            .collect();
+        let boundaries = PolicyBoundaries::new(&transformed_text(paragraph), policy);
+        let graphemes = &boundaries.graphemes;
+        let length = boundaries.length;
+        let annotations = annotation_ranges(paragraph);
         let hard_break = |c| matches!(c, '\n' | '\r' | '\u{2028}' | '\u{2029}');
         let mut units = Vec::new();
         let mut start = 0;
         for at in 1..graphemes.len() {
-            let (position, after) = graphemes[at];
-            let before = graphemes[at - 1].1;
-            let annotation = protected
-                .get(protected.partition_point(|r| r.end <= position))
-                .is_some_and(|r| r.start < position);
+            let position = graphemes[at].utf16;
+            let (before, after) = (graphemes[at - 1].first, graphemes[at].first);
             if hard_break(before)
                 || hard_break(after)
-                || (!annotation && policy.emergency_break_allowed(before, after))
+                || (!within_annotation(&annotations, position) && graphemes[at].emergency)
             {
                 if at - start > 1 {
-                    units.push(graphemes[start].0..position);
+                    units.push(graphemes[start].utf16..position);
                 }
                 start = at;
             }
         }
         if graphemes.len() - start > 1 {
-            units.push(graphemes[start].0..length);
+            units.push(graphemes[start].utf16..length);
         }
         if units.is_empty() {
             return;
@@ -107,9 +192,11 @@ impl HorizontalParagraphPlan {
             at > 0
                 && at < length
                 && graphemes
-                    .binary_search_by_key(&at, |g| g.0)
+                    .binary_search_by_key(&at, |g| g.utf16)
                     .ok()
-                    .is_some_and(|i| !hard_break(graphemes[i].1) && !hard_break(graphemes[i - 1].1))
+                    .is_some_and(|i| {
+                        !hard_break(graphemes[i].first) && !hard_break(graphemes[i - 1].first)
+                    })
         };
         let mut forced: Vec<_> = units
             .iter()

render-wasm/src/shapes/text_horizontal/ruby.rs

@@ -1,6 +1,5 @@
 use crate::render::FallbackFonts;
 use crate::shapes::text_paragraph::LayoutParagraph;
-use std::collections::HashMap;
 
 use super::*;
 
@@ -17,6 +16,25 @@ pub(crate) fn horizontal_zhuyin_origin(
     )
 }
 
+/// Zhuyin glyphs of a horizontal base whose builder text starts at UTF-16
+/// `base_start` and whose first line rect is `rect`, with the strip's
+/// top-left in the laid-out paragraph. Position data, painting and bounds
+/// share it.
+fn horizontal_zhuyin_glyphs(
+    span: &TextSpan,
+    syllable: &super::super::text_zhuyin::Syllable,
+    laid_out: &LayoutParagraph,
+    lines: &[skia::textlayout::LineMetrics],
+    base_start: usize,
+    rect: skia::Rect,
+    fonts: (&skia::textlayout::TypefaceFontProvider, &FallbackFonts),
+) -> (skia::Point, Vec<super::super::text_zhuyin::Glyph>) {
+    let em_top = horizontal_base_em_top(laid_out, lines, base_start, span.font_size, rect);
+    let origin = horizontal_zhuyin_origin(span, rect, em_top);
+    let glyphs = super::super::text_zhuyin::shape(span, syllable, fonts.0, fonts.1);
+    (origin, glyphs)
+}
+
 pub(crate) fn horizontal_zhuyin_entries(
     paragraph: &Paragraph,
     offsets: &HorizontalOffsets,
@@ -40,11 +58,17 @@ pub(crate) fn horizontal_zhuyin_entries(
         let Some(rect) = rects.first() else {
             continue;
         };
-        let em_top =
-            horizontal_base_em_top(laid_out, &lines, base_start, span.font_size, rect.rect);
-        let origin = horizontal_zhuyin_origin(span, rect.rect, em_top);
+        let (origin, glyphs) = horizontal_zhuyin_glyphs(
+            span,
+            &syllable,
+            laid_out,
+            &lines,
+            base_start,
+            rect.rect,
+            (provider, fallbacks),
+        );
         let size = span.ruby_font_size();
-        for glyph in super::super::text_zhuyin::shape(span, &syllable, provider, fallbacks) {
+        for glyph in glyphs {
             entries.push(crate::shapes::PositionData {
                 paragraph: 0,
                 span: index as u32,
@@ -173,8 +197,7 @@ pub(super) fn horizontal_ruby_spacing_with(
         .flat_map(|run| run.advances.iter().copied())
         .collect()
     };
-    // Overhang an earlier reading already takes on a character after it.
-    let mut claimed: HashMap<(usize, usize), f32> = HashMap::new();
+    let mut claims = OverhangClaims::new();
     let mut span_start = 0usize;
     for (index, (span, text)) in spans.iter().zip(texts).enumerate() {
         let text_start = span_start;
@@ -240,16 +263,14 @@ pub(super) fn horizontal_ruby_spacing_with(
                 )
             })
         };
-        let taken = previous
-            .and_then(|(owner, char_index, _)| claimed.get(&(owner, char_index)))
-            .copied()
-            .unwrap_or(0.0);
-        let rooms = ((room(previous) - taken).max(0.0), room(next));
+        let key = |(owner, char_index, _): (usize, usize, char)| (owner, char_index);
+        let rooms = claims.room(
+            previous.map(key),
+            next.map(key),
+            (room(previous), room(next)),
+            ruby_length - base_width,
+        );
         spacing.rooms[index] = rooms;
-        if let Some((owner, char_index, _)) = next {
-            let hang_after = ruby_overhangs(ruby_length - base_width, rooms).1;
-            claimed.insert((owner, char_index), hang_after);
-        }
         let gap = long_ruby_gap(ruby_length, base_width, rooms, base.len());
         if gap <= 0.0 {
             continue;
@@ -534,15 +555,16 @@ fn visit_horizontal_ruby_glyphs(
         );
         if let Some(syllable) = super::super::text_zhuyin::syllable(span) {
             if let Some(rect) = rects.first() {
-                let em_top =
-                    horizontal_base_em_top(laid_out, &lines, base_start, span.font_size, rect.rect);
-                let origin = horizontal_zhuyin_origin(span, rect.rect, em_top);
-                for glyph in super::super::text_zhuyin::shape(
+                let (origin, glyphs) = horizontal_zhuyin_glyphs(
                     span,
                     &syllable,
-                    font_provider,
-                    fallback_families,
-                ) {
+                    laid_out,
+                    &lines,
+                    base_start,
+                    rect.rect,
+                    (font_provider, fallback_families),
+                );
+                for glyph in glyphs {
                     visit(
                         &glyph.run.font,
                         glyph.run.glyphs[glyph.glyph],

render-wasm/src/shapes/text_policy.rs

@@ -2,6 +2,9 @@
 //! Regional presets use CLREQ (2026-10-07); Korean restrictions use KLREQ
 //! (2026-03-21). These are named presets, not universal regional mandates.
 
+use super::cjk_chars::{
+    is_han, COMMAS_AND_FULL_STOPS, FULLWIDTH_CLOSING_BRACKETS, FULLWIDTH_OPENING_BRACKETS,
+};
 use super::jlreq::{
     break_allowed_between, forbidden_at_line_end, forbidden_at_line_start,
     keeps_together_classified, pair_rule, JlreqClass,
@@ -40,6 +43,15 @@ pub enum TextLayoutPolicy {
     Korean(KoreanWrap),
 }
 
+/// Spacing between two adjacent characters of a pairwise policy, in em.
+#[derive(Debug, Default, Clone, Copy, PartialEq)]
+pub struct PairSpacing {
+    /// Removable punctuation side bearings.
+    pub compression: f32,
+    /// Han/Latin gap to add.
+    pub script_gap: f32,
+}
+
 /// The parts of a BCP 47 language tag the layout and font choices read.
 /// Subtags may be separated by `-` or `_`; matching ignores case.
 #[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
@@ -93,20 +105,18 @@ impl<'a> LanguageTag<'a> {
     }
 }
 
-pub fn is_han(ch: char) -> bool {
-    matches!(ch as u32, 0x3400..=0x4dbf | 0x4e00..=0x9fff | 0xf900..=0xfaff | 0x20000..=0x323af)
-}
-
 fn numeric(ch: char) -> bool {
     ch.is_numeric() || "一二三四五六七八九十百千萬万億亿〇零".contains(ch)
 }
 
 fn opening(ch: char) -> bool {
-    "（［｛〈《「『【〔〖〘〚‘“([{".contains(ch)
+    FULLWIDTH_OPENING_BRACKETS.contains(ch) || "‘“([{".contains(ch)
 }
 
 fn closing(ch: char) -> bool {
-    "，。、．！？：；）］｝〉》」』】〕〗〙〛’”)]},.!?:;".contains(ch)
+    COMMAS_AND_FULL_STOPS.contains(ch)
+        || FULLWIDTH_CLOSING_BRACKETS.contains(ch)
+        || "！？：；’”)]},.!?:;".contains(ch)
 }
 
 fn protected(ch: char) -> bool {
@@ -263,13 +273,13 @@ impl TextLayoutPolicy {
         else {
             return (0.0, 0.0);
         };
-        if "（［｛〈《「『【〔〖〘〚".contains(ch) {
+        if FULLWIDTH_OPENING_BRACKETS.contains(ch) {
             return (0.5, 0.0);
         }
-        if "）］｝〉》」』】〕〗〙〛".contains(ch) {
+        if FULLWIDTH_CLOSING_BRACKETS.contains(ch) {
             return (0.0, 0.5);
         }
-        let adjustable = "，。、．".contains(ch)
+        let adjustable = COMMAS_AND_FULL_STOPS.contains(ch)
             || (!vertical && region == ChineseRegion::Mainland && "：；！？".contains(ch))
             || (!vertical && region == ChineseRegion::HongKong && "：；".contains(ch));
         if !adjustable {
@@ -294,6 +304,16 @@ impl TextLayoutPolicy {
         }
     }
 
+    /// Pairwise spacing between two adjacent characters, in em: punctuation
+    /// compression to remove and the script gap to add. Each axis applies
+    /// its own caps and skips.
+    pub fn pair_spacing_em(self, before: char, after: char, vertical: bool) -> PairSpacing {
+        PairSpacing {
+            compression: self.compression_em(before, after, vertical),
+            script_gap: self.script_spacing_em(before, after),
+        }
+    }
+
     /// This preset permits hanging only in vertical traditional composition.
     /// Consecutive marks are excluded by the planner. Horizontal CN uses
     /// compression and push-out; no preset silently enables horizontal hanging.
@@ -303,7 +323,7 @@ impl TextLayoutPolicy {
                 self,
                 Self::Chinese(ChineseRegion::Taiwan | ChineseRegion::HongKong)
             )
-            && "，。、．".contains(ch)
+            && COMMAS_AND_FULL_STOPS.contains(ch)
         {
             0.5
         } else {

render-wasm/src/shapes/text_ruby.rs

@@ -1,6 +1,9 @@
 //! Ruby (furigana, Pinyin) piece, distribution and overhang rules shared by
 //! the horizontal and vertical engines (JLREQ §3.3).
 
+use std::collections::HashMap;
+use std::hash::Hash;
+
 use crate::shapes::jlreq::{classify, JlreqClass};
 use crate::shapes::{RubyAlign, RubyOverhang};
 
@@ -54,6 +57,43 @@ pub(crate) fn ruby_overhangs(overflow: f32, room: (f32, f32)) -> (f32, f32) {
     (hang_before, (overflow - hang_before).clamp(0.0, after))
 }
 
+/// Overhang that earlier readings already take on the character after them.
+/// A kana between two readings lends the later one only what the earlier one
+/// leaves (JLREQ Fig 138), so the two readings never meet over it. `K`
+/// identifies a neighbouring character.
+pub(crate) struct OverhangClaims<K> {
+    claimed: HashMap<K, f32>,
+}
+
+impl<K: Eq + Hash> OverhangClaims<K> {
+    pub(crate) fn new() -> Self {
+        Self {
+            claimed: HashMap::new(),
+        }
+    }
+
+    /// Room (before, after) left for a reading `overflow` longer than its
+    /// base, given the `room` its neighbours `before` and `after` offer, in
+    /// reading order. Records how far the reading hangs over `after`.
+    pub(crate) fn room(
+        &mut self,
+        before: Option<K>,
+        after: Option<K>,
+        room: (f32, f32),
+        overflow: f32,
+    ) -> (f32, f32) {
+        let taken = before
+            .and_then(|key| self.claimed.get(&key))
+            .copied()
+            .unwrap_or(0.0);
+        let room = ((room.0 - taken).max(0.0), room.1);
+        if let Some(after) = after {
+            self.claimed.insert(after, ruby_overhangs(overflow, room).1);
+        }
+        room
+    }
+}
+
 /// Gap that spreads a base under a reading longer than itself (JLREQ
 /// §3.3.8). What the reading set solid (`ruby_length`) overflows the base
 /// by, less the overhang `room`, becomes one gap per base character: a full

render-wasm/src/shapes/text_vertical/annotations.rs

@@ -9,7 +9,7 @@ use skia_safe::{self as skia, Font};
 
 use crate::shapes::text_attrs::{emphasis_char_allowed, EMPHASIS_FONT_SCALE};
 use crate::shapes::text_ruby::{
-    distribute_ruby_pieces, long_ruby_gap, ruby_overhang_room, ruby_overhangs, ruby_pieces,
+    distribute_ruby_pieces, long_ruby_gap, ruby_overhang_room, ruby_pieces, OverhangClaims,
 };
 use crate::shapes::{
     AnnotationSide, AppliedTextTransform, FontFeatures, Paragraph, RubyOverhang, TextContent,
@@ -208,7 +208,7 @@ pub(super) fn set_ruby_overhang_rooms(flow: &[FlowCell], ruby_units: &mut [RubyB
             .is_some_and(|unit| ruby_units[*unit].annotates(cell))
     };
     let indices = annotated_indices(flow.iter().map(|item| &item.cell), ruby_units);
-    let mut claimed: HashMap<usize, f32> = HashMap::new();
+    let mut claims = OverhangClaims::new();
     let rooms: Vec<(f32, f32)> = ruby_units
         .iter()
         .zip(&indices)
@@ -229,13 +229,13 @@ pub(super) fn set_ruby_overhang_rooms(flow: &[FlowCell], ruby_units: &mut [RubyB
             };
             let before = first.checked_sub(1);
             let after = Some(last + 1).filter(|index| *index < flow.len());
-            let taken = before.and_then(|index| claimed.get(&index)).copied();
-            let room = ((room(before) - taken.unwrap_or(0.0)).max(0.0), room(after));
-            if let Some(after) = after {
-                let base: f32 = indices.iter().map(|index| flow[*index].cell.extent).sum();
-                claimed.insert(after, ruby_overhangs(unit.ruby_line() - base, room).1);
-            }
-            room
+            let base: f32 = indices.iter().map(|index| flow[*index].cell.extent).sum();
+            claims.room(
+                before,
+                after,
+                (room(before), room(after)),
+                unit.ruby_line() - base,
+            )
         })
         .collect();
     for (unit, room) in ruby_units.iter_mut().zip(rooms) {

render-wasm/src/shapes/text_vertical/flow.rs

@@ -3,12 +3,16 @@
 // (aki, oikomi, inter-script spacing) and pick column breaks (kinsoku,
 // burasage, oidashi).
 
+use crate::shapes::cjk_chars::is_punctuation;
 use crate::shapes::jlreq::{
     aki_class, break_allowed_between, classify, pair_rule, punctuation_aki_sheds, AkiShed,
     JlreqClass,
 };
 use crate::shapes::jlreq::{forbidden_at_line_end, forbidden_at_line_start};
-use crate::shapes::text_allocation::capped_equal_allocations;
+use crate::shapes::line_breaks::PolicyBoundaries;
+use crate::shapes::text_allocation::{
+    allocate_bounded_expansion, bounded_expansion, capped_equal_allocations,
+};
 use crate::shapes::text_policy::TextLayoutPolicy;
 use crate::shapes::{LineAdjustment, TextAlign};
 use unicode_segmentation::UnicodeSegmentation;
@@ -293,34 +297,21 @@ pub(super) fn plan_pairwise_columns(
     policy: TextLayoutPolicy,
     max_height: f32,
 ) -> Vec<(usize, f32)> {
-    let mut boundaries = Vec::new();
-    let mut offset = 0;
-    let mut previous = None;
-    for grapheme in text.graphemes(true) {
-        let first = grapheme.chars().next().unwrap();
-        let allowed = previous.is_none_or(|before| policy.break_allowed(before, first));
-        let emergency = previous.is_none_or(|before| policy.emergency_break_allowed(before, first));
-        boundaries.push((offset, allowed, emergency));
-        previous = Some(first);
-        offset += grapheme.encode_utf16().count();
-    }
-    boundaries.push((offset, true, true));
+    let boundaries = PolicyBoundaries::new(text, policy);
     let mut emergency_start = Vec::with_capacity(cells.len());
     let mut unit_start = Vec::with_capacity(cells.len());
     let mut prefix = vec![0.0];
     for (i, flow) in cells.iter().enumerate() {
-        let boundary = boundaries
-            .binary_search_by_key(&flow.cell.start, |b| b.0)
-            .ok();
+        let boundary = boundaries.at(flow.cell.start);
         let atomic =
             i > 0 && (flow.keep_with_previous || flow.shares_previous_box || boundary.is_none());
-        let emergency_joined = atomic || boundary.is_some_and(|at| !boundaries[at].2);
+        let emergency_joined = atomic || boundary.is_some_and(|(_, emergency)| !emergency);
         emergency_start.push(if i > 0 && emergency_joined {
             emergency_start[i - 1]
         } else {
             i
         });
-        let joined = atomic || boundary.is_some_and(|at| !boundaries[at].1);
+        let joined = atomic || boundary.is_some_and(|(allowed, _)| !allowed);
         unit_start.push(if i > 0 && joined {
             unit_start[i - 1]
         } else {
@@ -339,12 +330,7 @@ pub(super) fn plan_pairwise_columns(
     let mut placements = Vec::with_capacity(cells.len());
     let (mut column, mut column_start, mut cursor) = (0, 0, 0.0);
     for (i, flow) in cells.iter().enumerate() {
-        let punctuation = |at: usize| {
-            cells
-                .get(at)
-                .and_then(|c| c.ch)
-                .is_some_and(|c| "，。、．！？：；…‥—―「」『』（）〔〕［］｛｝〈〉《》【】〖〗〘〙〚〛‘’“”.,;!?:()[]{}".contains(c))
-        };
+        let punctuation = |at: usize| cells.get(at).and_then(|c| c.ch).is_some_and(is_punctuation);
         let solitary = (i == 0 || !punctuation(i - 1)) && !punctuation(i + 1);
         let hang = if solitary {
             flow.ch
@@ -425,7 +411,8 @@ pub(super) fn apply_pairwise_spacing(cells: &mut [FlowCell], text: &str, policy:
             - cells[i - 1].trailing_spacing
             - 0.5 * cells[i - 1].cell.font_size)
             .max(0.0);
-        let compression = (policy.compression_em(before, after, true) * em).min(available);
+        let spacing = policy.pair_spacing_em(before, after, true);
+        let compression = (spacing.compression * em).min(available);
         // A tate-chu-yoko composite sets in one ideograph's cell, so the
         // Han/Western gap of its digits or letters does not apply.
         let tate_chu_yoko =
@@ -433,7 +420,7 @@ pub(super) fn apply_pairwise_spacing(cells: &mut [FlowCell], text: &str, policy:
         let gap = if tate_chu_yoko(&cells[i - 1]) || tate_chu_yoko(&cells[i]) {
             0.0
         } else {
-            policy.script_spacing_em(before, after) * em
+            spacing.script_gap * em
         };
         cells[i - 1].cell.extent = (cells[i - 1].cell.extent + gap - compression)
             .max(cells[i - 1].trailing_spacing.max(0.0));
@@ -475,7 +462,7 @@ pub(super) fn pairwise_aligned_tops(
         if (align == TextAlign::Distribute || (align == TextAlign::Justify && column < last_column))
             && is_bounded(budget)
         {
-            let mut remaining = (budget - used).max(0.0);
+            let remaining = (budget - used).max(0.0);
             let mut extras = vec![0.0; end - start];
             let mut opportunities = Vec::new();
             for i in start..end - 1 {
@@ -495,34 +482,16 @@ pub(super) fn pairwise_aligned_tops(
                 {
                     continue;
                 }
-                let (priority, cap) = if align == TextAlign::Distribute {
-                    (
-                        1,
-                        budget
-                            / cells[i]
-                                .cell
-                                .font_size
-                                .min(cells[i + 1].cell.font_size)
-                                .max(0.01),
-                    )
-                } else {
-                    policy.expansion_em(before, after)
-                };
                 let em = cells[i].cell.font_size.min(cells[i + 1].cell.font_size);
-                if priority > 0 {
-                    opportunities.push((i, priority, cap * em));
+                let distribute = (align == TextAlign::Distribute).then_some(budget);
+                if let Some((priority, cap)) =
+                    bounded_expansion(policy, before, after, em, distribute)
+                {
+                    opportunities.push((i, priority, cap));
                 }
             }
-            for priority in 1..=2 {
-                let caps: Vec<_> = opportunities
-                    .iter()
-                    .filter(|(_, p, _)| *p == priority)
-                    .map(|(i, _, cap)| (*i, *cap))
-                    .collect();
-                for (i, amount) in capped_equal_allocations(&caps, remaining) {
-                    extras[i - start] += amount;
-                    remaining -= amount;
-                }
+            for (i, amount) in allocate_bounded_expansion(&opportunities, remaining) {
+                extras[i - start] += amount;
             }
             let mut extra = 0.0;
             for i in start..end {

render-wasm/src/shapes/text_vertical/orientation.rs

@@ -12,9 +12,10 @@ pub(crate) fn is_emoji_char(c: char) -> bool {
     )
 }
 
-/// Characters that stay upright in vertical flow: kana, kanji, CJK
-/// punctuation, full-width forms and emoji. Everything else rotates sideways
-/// under `text-orientation: mixed`.
+/// Characters that stay upright in vertical flow: kana, kanji, hangul,
+/// bopomofo, CJK punctuation, full-width forms and emoji. Everything else
+/// rotates sideways under `text-orientation: mixed`. A table of its own: it
+/// leaves half-width katakana sideways, unlike `cjk_chars::is_kana`.
 pub(crate) fn is_upright_char(c: char) -> bool {
     is_emoji_char(c)
         || matches!(u32::from(c),

render-wasm/src/shapes/text_zhuyin.rs

@@ -4,6 +4,7 @@ use std::ops::Range;
 
 use skia_safe::{self as skia, textlayout::TypefaceFontProvider, FontMgr};
 
+use super::cjk_chars::is_zhuyin_base;
 use super::text_vertical::{shape_segment_with_fallbacks, span_font_families, ShapedRun};
 use super::TextSpan;
 use crate::render::FallbackFonts;
@@ -89,7 +90,7 @@ pub(crate) fn single_han(text: &str) -> bool {
     let Some(first) = chars.next() else {
         return false;
     };
-    matches!(first as u32, 0x3400..=0x9fff | 0xf900..=0xfaff | 0x20000..=0x323af)
+    is_zhuyin_base(first)
         && chars.next().is_none_or(|ch| {
             matches!(ch as u32, 0xfe00..=0xfe0f | 0xe0100..=0xe01ef) && chars.next().is_none()
         })