Fonts: reuse shared MATH construction records (#4660)

Co-Authored-By: OPUZZLES <109076338+OPUZZLES@users.noreply.github.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

ROADMAP.md

@@ -4,7 +4,7 @@
 
 **Stage: alpha** · next: beta, ~15 points (60% → 75% ready for real work) and ~150–230 h away
 
-> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-11-05 · **Change:** trivial (DOCX `w:adjustRightInd` is preserved through style resolution and ordered DOCX saves); previously trivial (per-page ink object lookups, #3614); previously trivial (progress log: ODT display settings removed); previously trivial (progress log: dialogs are modal except the five Word keeps modeless, and a press beside one shakes it, #3859); previously trivial (progress log: picture brightness, contrast and transparency, ODT element audit, ODT checked against 220 files from other applications, ODT display styles); previously trivial (progress log: the dialogs of the mirrored Arabic interface read right to left: label-and-field rows, tables (list columns, the Models table), tile galleries and the Add/Remove arrows follow the reading direction, through one `rtl::Table` helper that draws exactly as before in English; no ranked gap closed, readiness estimates unchanged); previously trivial (progress log: fonts that aren't installed are flagged in the font box and status bar); previously trivial (progress log: ODT sections inside text boxes); previously trivial (progress log: ODT definitions sharing a name, sections around tables, stacking order); previously trivial (progress log: ODT forms and indexes written as read); previously trivial (progress log: ODT table styles saved, WordCraft-origin documents checked in LibreOffice); previously trivial (progress log: ODT interpretation profile); previously trivial (progress log: ODT controls, custom shapes, in-line placement); previously trivial (progress log: ODT charts kept, frame outlines, header height); previously trivial (progress log: ODT header rows); previously trivial (progress log: ODT layout setting, padding, columns); previously trivial (progress log: ODT tables, header spacing, picture paragraphs); previously trivial (progress log: ODT round trips keep their page count); previously trivial (progress log: emphasis marks, SET fields, ODT page backgrounds and tables); previously trivial (progress log: ODT checked against LibreOffice's templates); previously trivial (progress log: ODT at ~90%); previously trivial (progress log: ODT read and written nearly whole); previously trivial (progress log: ODT text boxes); previously trivial (progress log: ODT layout settings); previously trivial (progress log: ODT character styles); previously trivial (progress log: ODT custom styles); previously trivial (progress log: ODT equations saved); previously trivial (progress log: ODT floating pictures); previously trivial (progress log: ODT paragraph borders); previously trivial (progress log: ODT heading numbering); previously trivial (progress log: ODT table borders); previously trivial (progress log: ODT heading styles); previously trivial (progress log: ODT tab stops); previously trivial (progress log: ODT character formatting); previously trivial (progress log: ODT lists); previously trivial (progress log: ODT comments); previously trivial (progress log: ODT footnotes and endnotes); previously trivial (progress log: ODT headers, footers and fields); previously trivial (progress log: ODT paragraph spacing); previously trivial (progress log: Options › Language redesign, CP4.3); previously trivial (progress log: the missing-pack bar on a fresh install, the caret's typing language, and the spelling menu heading in the interface language, CP4.3); previously trivial (progress log: Language pack languages in every language list, CP4.3); previously trivial (progress log: craft-proof pin, French elision and Valencian, CP4.3); previously trivial (progress log: a long comment's card and balloon stop at ten lines, #3382); previously trivial (progress log: a pasted text box story is charged for the formatting it copies); previously trivial (progress log: a .doc table row keeps one override per cell); previously trivial (progress log: a .doc file's sections find their paragraphs by index); previously minor (Arabic interface language with mirrored UI, #1195); previously trivial (progress log: Replace and Remove Table of Contents with Word's title above a section break, #4041); previously trivial (progress log: a password-protected file named on the command line asks for its password after the startup document is replaced, #5299); previously trivial (progress log: language-packs-2 published and pinned, 75 Language packs); previously trivial (progress log: language-packs-1 published and pinned, CP4.4); previously trivial (progress log: Language packs, CP4);; previously trivial (DOCX table rows and cells inside selected alternate-content branches survive open and repeated saves (#5219); broader corpus qualification and readiness estimates unchanged); previously trivial (progress log: a list inside a Markdown block quote stays quoted); previously trivial (progress log: arrowheads keep their kind, width and length and are drawn at Word's sizes); previously trivial (progress log: the line between columnLine truncated
+> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-11-05 · **Change:** trivial (MATH construction reuse, #3630); previously trivial (DOCX `w:adjustRightInd` is preserved through style resolution and ordered DOCX saves); previously trivial (per-page ink object lookups, #3614); previously trivial (progress log: ODT display settings removed); previously trivial (progress log: dialogs are modal except the five Word keeps modeless, and a press beside one shakes it, #3859); previously trivial (progress log: picture brightness, contrast and transparency, ODT element audit, ODT checked against 220 files from other applications, ODT display styles); previously trivial (progress log: the dialogs of the mirrored Arabic interface read right to left: label-and-field rows, tables (list columns, the Models table), tile galleries and the Add/Remove arrows follow the reading direction, through one `rtl::Table` helper that draws exactly as before in English; no ranked gap closed, readiness estimates unchanged); previously trivial (progress log: fonts that aren't installed are flagged in the font box and status bar); previously trivial (progress log: ODT sections inside text boxes); previously trivial (progress log: ODT definitions sharing a name, sections around tables, stacking order); previously trivial (progress log: ODT forms and indexes written as read); previously trivial (progress log: ODT table styles saved, WordCraft-origin documents checked in LibreOffice); previously trivial (progress log: ODT interpretation profile); previously trivial (progress log: ODT controls, custom shapes, in-line placement); previously trivial (progress log: ODT charts kept, frame outlines, header height); previously trivial (progress log: ODT header rows); previously trivial (progress log: ODT layout setting, padding, columns); previously trivial (progress log: ODT tables, header spacing, picture paragraphs); previously trivial (progress log: ODT round trips keep their page count); previously trivial (progress log: emphasis marks, SET fields, ODT page backgrounds and tables); previously trivial (progress log: ODT checked against LibreOffice's templates); previously trivial (progress log: ODT at ~90%); previously trivial (progress log: ODT read and written nearly whole); previously trivial (progress log: ODT text boxes); previously trivial (progress log: ODT layout settings); previously trivial (progress log: ODT character styles); previously trivial (progress log: ODT custom styles); previously trivial (progress log: ODT equations saved); previously trivial (progress log: ODT floating pictures); previously trivial (progress log: ODT paragraph borders); previously trivial (progress log: ODT heading numbering); previously trivial (progress log: ODT table borders); previously trivial (progress log: ODT heading styles); previously trivial (progress log: ODT tab stops); previously trivial (progress log: ODT character formatting); previously trivial (progress log: ODT lists); previously trivial (progress log: ODT comments); previously trivial (progress log: ODT footnotes and endnotes); previously trivial (progress log: ODT headers, footers and fields); previously trivial (progress log: ODT paragraph spacing); previously trivial (progress log: Options › Language redesign, CP4.3); previously trivial (progress log: the missing-pack bar on a fresh install, the caret's typing language, and the spelling menu heading in the interface language, CP4.3); previously trivial (progress log: Language pack languages in every language list, CP4.3); previously trivial (progress log: craft-proof pin, French elision and Valencian, CP4.3); previously trivial (progress log: a long comment's card and balloon stop at ten lines, #3382); previously trivial (progress log: a pasted text box story is charged for the formatting it copies); previously trivial (progress log: a .doc table row keeps one override per cell); previously trivial (progress log: a .doc file's sections find their paragraphs by index); previously minor (Arabic interface language with mirrored UI, #1195); previously trivial (progress log: Replace and Remove Table of Contents with Word's title above a section break, #4041); previously trivial (progress log: a password-protected file named on the command line asks for its password after the startup document is replaced, #5299); previously trivial (progress log: language-packs-2 published and pinned, 75 Language packs); previously trivial (progress log: language-packs-1 published and pinned, CP4.4); previously trivial (progress log: Language packs, CP4);; previously trivial (DOCX table rows and cells inside selected alternate-content branches survive open and repeated saves (#5219); broader corpus qualification and readiness estimates unchanged); previously trivial (progress log: a list inside a Markdown block quote stays quoted); previously trivial (progress log: arrowheads keep their kind, width and length and are drawn at Word's sizes); prevLine truncated
 > **Printing update:** 2026-10-12 · direct desktop printing through craft-print; details below.
 
 WordCraft aims for complete parity with Microsoft Word, then goes further on speed, openness and
@@ -759,6 +759,7 @@ Imported Word 97–2003 `.doc`/`.dot` files now ask for a supported DOCX target
 Footer page-table height and body wrapping are corrected. Other pagination gaps remain open.
 
 Section page-numbering settings preserve `w:chapStyle` values across the full unsigned-byte range.
+MATH glyphs that reference the same construction now share its immutable variants and assembly data (#3630). Glyph mappings, values and the existing partial-record fallback stay unchanged. Repeated glyph IDs keep their last valid reference without retaining superseded constructions. Distinct overlapping constructions and assembly-only sharing remain separate.
 
 ## Revision history
 
@@ -776,6 +777,7 @@ Section page-numbering settings preserve `w:chapStyle` values across the full un
 |---|---|---|
 | 2026-11-05 | trivial | DOCX paragraph `w:adjustRightInd` is preserved with true, false and absent values; OOXML properties follow schema order. Grid layout remains open. |
 | 2026-10-28 | trivial | Per-page ink object lookups, #3614. Drawing order, geometry and saved data retain their behavior; no readiness estimate change |
+| 2026-10-28 | trivial | Shared MATH construction records reuse variants and assembly data without changing glyph mappings or existing limits (#3630); readiness estimates unchanged |
 | 2026-11-04 | trivial | Progress log: OpenDocument display settings removed (gap 19) |
 | 2026-10-31 | trivial | Progress log: dialogs are modal except the five Word keeps modeless; a press beside one shakes it (#3859) |
 | 2026-11-04 | trivial | Progress log: picture brightness, contrast and transparency (#1711, #1766) |

crates/fonts/src/math.rs

@@ -88,7 +88,7 @@ pub struct Construction {
     pub parts: Vec<GlyphPart>,
 }
 
-/// A parsed `MATH` table. Values are font units.
+/// A parsed `MATH` table. Values are font units; glyphs can share immutable constructions.
 #[derive(Clone, Debug, Default)]
 pub struct MathTable {
     pub script_percent: f32,
@@ -100,8 +100,8 @@ pub struct MathTable {
     pub italics: HashMap<u32, f32>,
     pub top_accent: HashMap<u32, f32>,
     pub min_connector_overlap: f32,
-    pub vert: HashMap<u32, Construction>,
-    pub horiz: HashMap<u32, Construction>,
+    pub vert: HashMap<u32, Arc<Construction>>,
+    pub horiz: HashMap<u32, Arc<Construction>>,
 }
 
 impl MathTable {
@@ -185,41 +185,59 @@ fn value_map(b: &[u8], at: usize) -> HashMap<u32, f32> {
     m
 }
 
-fn constructions(b: &[u8], variants: usize, cov_at: usize, count: usize, offsets_at: usize) -> HashMap<u32, Construction> {
+fn construction(b: &[u8], c: usize) -> Construction {
+    let mut con = Construction::default();
+    let nv = u16_at(b, c + 2).unwrap_or(0) as usize;
+    for k in 0..nv.min(256) {
+        let r = c + 4 + k * 4;
+        let (Some(vg), Some(adv)) = (u16_at(b, r), u16_at(b, r + 2)) else { break };
+        con.variants.push((vg as u32, adv as f32));
+    }
+    if let Some(a) = sub(b, c, c) {
+        let np = u16_at(b, a + 4).unwrap_or(0) as usize;
+        for k in 0..np.min(64) {
+            let r = a + 6 + k * 10;
+            let (Some(pg), Some(s), Some(e), Some(full), Some(fl)) =
+                (u16_at(b, r), u16_at(b, r + 2), u16_at(b, r + 4), u16_at(b, r + 6), u16_at(b, r + 8))
+            else {
+                break;
+            };
+            con.parts.push(GlyphPart {
+                gid: pg as u32,
+                start_connector: s as f32,
+                end_connector: e as f32,
+                full_advance: full as f32,
+                extender: fl & 1 != 0,
+            });
+        }
+    }
+    con
+}
+
+fn constructions(
+    b: &[u8],
+    variants: usize,
+    cov_at: usize,
+    count: usize,
+    offsets_at: usize,
+    cache: &mut HashMap<usize, Arc<Construction>>,
+) -> HashMap<u32, Arc<Construction>> {
     let mut m = HashMap::new();
     let Some(cov) = sub(b, variants, cov_at) else { return m };
     let glyphs = coverage(b, cov);
+    let mut offsets = HashMap::new();
     for (i, g) in glyphs.iter().enumerate().take(count) {
         let Some(c) = sub(b, variants, offsets_at + i * 2) else { continue };
-        let mut con = Construction::default();
-        let nv = u16_at(b, c + 2).unwrap_or(0) as usize;
-        for k in 0..nv.min(256) {
-            let r = c + 4 + k * 4;
-            let (Some(vg), Some(adv)) = (u16_at(b, r), u16_at(b, r + 2)) else { break };
-            con.variants.push((vg as u32, adv as f32));
-        }
-        if let Some(a) = sub(b, c, c) {
-            let np = u16_at(b, a + 4).unwrap_or(0) as usize;
-            for k in 0..np.min(64) {
-                let r = a + 6 + k * 10;
-                let (Some(pg), Some(s), Some(e), Some(full), Some(fl)) =
-                    (u16_at(b, r), u16_at(b, r + 2), u16_at(b, r + 4), u16_at(b, r + 6), u16_at(b, r + 8))
-                else {
-                    break;
-                };
-                con.parts.push(GlyphPart {
-                    gid: pg as u32,
-                    start_connector: s as f32,
-                    end_connector: e as f32,
-                    full_advance: full as f32,
-                    extender: fl & 1 != 0,
-                });
-            }
-        }
         if *g != u32::MAX {
-            m.insert(*g, con);
+            offsets.insert(*g, c);
         }
     }
+    // Malformed coverage may repeat a glyph. Decode only its last valid reference so
+    // superseded records cannot accumulate in the shared cache.
+    for (g, c) in offsets {
+        let con = cache.entry(c).or_insert_with(|| Arc::new(construction(b, c)));
+        m.insert(g, Arc::clone(con));
+    }
     m
 }
 
@@ -250,8 +268,9 @@ pub fn parse(b: &[u8]) -> Option<MathTable> {
         t.min_connector_overlap = u16_at(b, v).unwrap_or(0) as f32;
         let nv = u16_at(b, v + 6).unwrap_or(0) as usize;
         let nh = u16_at(b, v + 8).unwrap_or(0) as usize;
-        t.vert = constructions(b, v, v + 2, nv, v + 10);
-        t.horiz = constructions(b, v, v + 4, nh, v + 10 + nv * 2);
+        let mut cache = HashMap::new();
+        t.vert = constructions(b, v, v + 2, nv, v + 10, &mut cache);
+        t.horiz = constructions(b, v, v + 4, nh, v + 10 + nv * 2, &mut cache);
     }
     Some(t)
 }
@@ -425,6 +444,126 @@ mod tests {
         assert!(!cached(&long), "absurdly long names must not be cached");
     }
 
+    /// A small MathVariants table with two directions and shared or distinct records.
+    fn construction_bytes(count: u16, distinct: bool) -> Vec<u8> {
+        let cov = 10 + 4 * usize::from(count);
+        let con = cov + 10;
+        let mut b = vec![0; con];
+        let put = |b: &mut Vec<u8>, at: usize, value: u16| b[at..at + 2].copy_from_slice(&value.to_be_bytes());
+        put(&mut b, 0, 50);
+        put(&mut b, 2, cov as u16);
+        put(&mut b, 4, cov as u16);
+        put(&mut b, 6, count);
+        put(&mut b, 8, count);
+        for i in 0..2 * usize::from(count) {
+            put(&mut b, 10 + 2 * i, (con + if distinct && i % 2 == 1 { 1674 } else { 0 }) as u16);
+        }
+        for (i, v) in [2, 1, 0, count - 1, 0].into_iter().enumerate() {
+            put(&mut b, cov + 2 * i, v);
+        }
+        for shift in 0..if distinct { 2 } else { 1 } {
+            b.extend_from_slice(&1028u16.to_be_bytes());
+            b.extend_from_slice(&256u16.to_be_bytes());
+            for k in 0..256 {
+                b.extend_from_slice(&7u16.to_be_bytes());
+                b.extend_from_slice(&(500u16 + k + shift).to_be_bytes());
+            }
+            for v in [0u16, 0, 64] {
+                b.extend_from_slice(&v.to_be_bytes());
+            }
+            for _ in 0..64 {
+                for v in [8u16, 10, 20, 600 + shift, 1] {
+                    b.extend_from_slice(&v.to_be_bytes());
+                }
+            }
+        }
+        b
+    }
+
+    fn construction_math(variants: &[u8]) -> Vec<u8> {
+        let v = 10 + 8 + VALUE_COUNT * 4 + 2;
+        let mut b = vec![0; v];
+        b[1] = 1;
+        b[5] = 10;
+        b[8..10].copy_from_slice(&(v as u16).to_be_bytes());
+        b.extend_from_slice(variants);
+        b
+    }
+
+    #[test]
+    fn shared_constructions_reuse_variant_and_assembly_storage() {
+        // 16,000 references per direction fit in the table's 16-bit offset domain.
+        let table = parse(&construction_math(&construction_bytes(16_000, false))).unwrap();
+        assert_eq!(table.vert.len(), 16_000);
+        assert_eq!(table.horiz.len(), 16_000);
+        let first = table.vert.get(&0).unwrap();
+        assert_eq!(first.variants.len(), 256);
+        assert_eq!(first.variants[255], (7, 755.0));
+        assert_eq!(first.parts.len(), 64);
+        assert_eq!(first.parts[0], GlyphPart { gid: 8, start_connector: 10.0, end_connector: 20.0, full_advance: 600.0, extender: true });
+        for con in table.vert.values().chain(table.horiz.values()) {
+            assert_eq!(con.variants.as_ptr(), first.variants.as_ptr(), "one source record must have one variant allocation");
+            assert_eq!(con.parts.as_ptr(), first.parts.as_ptr(), "one source record must have one assembly allocation");
+        }
+    }
+
+    #[test]
+    fn distinct_constructions_keep_values_and_direction_mappings() {
+        let table = parse(&construction_math(&construction_bytes(4, true))).unwrap();
+        for glyph in 0..4 {
+            let con = table.vert.get(&glyph).unwrap();
+            let other = table.horiz.get(&glyph).unwrap();
+            assert_eq!(con.variants[0], (7, 500.0 + (glyph % 2) as f32));
+            assert_eq!(con.parts[0].full_advance, 600.0 + (glyph % 2) as f32);
+            assert_eq!(con.variants.as_ptr(), other.variants.as_ptr());
+        }
+        assert_ne!(table.vert[&0].variants.as_ptr(), table.vert[&1].variants.as_ptr());
+    }
+
+    #[test]
+    fn shared_constructions_keep_truncated_records_and_absent_offsets() {
+        let mut variants = construction_bytes(4, false);
+        let con = u16_at(&variants, 10).unwrap() as usize;
+        variants.truncate(con + 4 + 10);
+        // An absent source reference remains absent, rather than aliasing record zero.
+        variants[10..12].copy_from_slice(&0u16.to_be_bytes());
+        let table = parse(&construction_math(&variants)).unwrap();
+        assert!(!table.vert.contains_key(&0));
+        let first = table.vert.get(&1).unwrap();
+        assert_eq!(first.variants, [(7, 500.0), (7, 501.0)]);
+        assert!(first.parts.is_empty());
+        for con in table.vert.values().chain(table.horiz.values()) {
+            assert_eq!(con.variants.as_ptr(), first.variants.as_ptr());
+        }
+    }
+
+    #[test]
+    fn duplicate_coverage_retains_only_surviving_constructions() {
+        let mut variants = construction_bytes(3, true);
+        let cov = u16_at(&variants, 2).unwrap() as usize;
+        // Format 1 accepts repeated glyph IDs. Each direction keeps its last valid offset.
+        for (i, value) in [1u16, 3, 7, 7, 7].into_iter().enumerate() {
+            variants[cov + i * 2..cov + i * 2 + 2].copy_from_slice(&value.to_be_bytes());
+        }
+        let mut cache = HashMap::new();
+        let vert = constructions(&variants, 0, 2, 3, 10, &mut cache);
+        assert_eq!(vert.len(), 1);
+        assert_eq!(vert[&7].variants[0], (7, 500.0));
+        assert_eq!(cache.len(), 1, "superseded records must not stay alive in the cache");
+        let horiz = constructions(&variants, 0, 4, 3, 16, &mut cache);
+        assert_eq!(horiz.len(), 1);
+        assert_eq!(horiz[&7].variants[0], (7, 501.0));
+        assert_eq!(cache.len(), 2, "each surviving direction can retain its own record");
+
+        // A later absent offset must not remove the preceding valid mapping.
+        variants[14..16].copy_from_slice(&0u16.to_be_bytes());
+        let table = parse(&construction_math(&variants)).unwrap();
+        assert_eq!(table.vert.len(), 1);
+        assert_eq!(table.vert[&7].variants[0], (7, 501.0));
+        assert_eq!(table.horiz[&7].variants[0], (7, 501.0));
+        assert!(Arc::ptr_eq(&table.vert[&7], &table.horiz[&7]));
+    }
+
     #[test]
     fn garbage_tables_never_panic() {
         assert!(parse(&[]).is_none());

docs/gaps.md

@@ -1,6 +1,6 @@
 # Where WordCraft falls short of Word
 
-> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-11-05 · **Change:** trivial (DOCX `w:adjustRightInd` defaults true and is written in schema order); previously trivial (ink page lookups reuse scanned paragraph prefixes, #3614); previously trivial (#19: ODT settings for what the standard leaves open removed); previously trivial (#19: ODT element audit; files from other applications checked); previously trivial (#11: the dialogs of the mirrored Arabic interface read right to left: label-and-field rows, tables (list columns, the Models table), tile galleries and the Add/Remove arrows follow the reading direction, through one `rtl::Table` helper that draws exactly as before in English; no ranked gap closed, readiness estimates unchanged); previously trivial (#4: fonts that aren't installed are flagged in the font box and status bar); previously trivial (#19: ODT sections inside text boxes kept as read); previously trivial (#19: an ODT saved without edits keeps what WordCraft doesn't model); previously trivial (#19: ODT interpretation profile, WordCraft / LibreOffice / custom); previously trivial (#19: ODT form controls placed and multi-line; custom shapes keep their geometry); previously trivial (#19: ODT in-line frames placed against the text; template instantiation listed as left); previously trivial (#19: ODT charts kept as read; frame outlines; default header height); previously trivial (#19: ODT tables no longer gain repeated header rows); previously trivial (#19: ODT padding without a border, unequal columns, column breaks, picture-only lines); previously trivial (#19: ODT tables' margins and spacing, header spacing, pictures alone in a paragraph); previously trivial (#19: no LibreOffice template changes page count through a save); previously trivial (#19: ODT variables, emphasis marks, page backgrounds, fixed tables, default size); previously trivial (#19: ODT checked against LibreOffice's own templates: placeholders, first-page styles, unsized pages, metafile logos, `.ott`); previously trivial (#19: ODT at ~90%: custom shapes, cell formulas, macros kept, turned text, control sizes and buttons); previously trivial (#19 narrowed: ODT read and written nearly whole; estimate 18–27 h → 10–16 h); previously trivial (#10 narrowed: text in a turned text box is drawn turned with it); previously trivial (gap 19 narrowed: ODT heading numbering read and written); previously trivial (gap 19 narrowed: ODT table borders and padding read and written); previously trivial (gap 19 narrowed: ODT heading style looks read and written); previously trivial (gap 19 narrowed: ODT tab stops read and written); previously trivial (gap 19 narrowed: ODT character formatting read and written); previously trivial (gap 19 narrowed: ODT list formats read and written); previously trivial (gap 19 narrowed: ODT comments read and written); previously trivial (gap 19 narrowed: ODT footnotes and endnotes read and written); previously trivial (gap 19 narrowed: ODT headers and footers written, text fields and paragraph spacing read and written); previously trivial (#8: Options › Language lists Your languages and Add a language… replaces the Language packs table); previously trivial (#8: the missing-pack bar works on a fresh install and counts the caret's language); previously trivial (#8: Language pack languages are in Set Proofing Language, Review › Language and the status bar; built-in proofing is named as such); previously trivial (#20: layout resource caps and linear-time passes (#3867, #3866)); previously trivial (#3 narrowed: a pasted text box story is charged for the formatting it copies); previously trivial (#3 narrowed: a .doc table row keeps one override per cell); previously trivial (#3 narrowed: a .doc file's sections find their paragraphs by index); previously minor (#11: Arabic interface with mirrored shell landed, #1195; hi/id/pt-PT and Hebrew UI remain); previously trivial (#8: language-packs-2 is published and pinned: 75 packs, adding Danish, German (Germany, Austria, Switzerland), Esperanto, Estonian, Persian, Croatian, Hungarian, Norwegian (Bokmål, Nynorsk), Dutch and Swedish (Sweden, Finland); no ranked gap changed); previously trivial (#8: language-packs-1 is published and pinned, 61 packs: Lithuanian, Romanian, Catalan, French, Greek, Slovak, Serbian and Latvian install as Language packs; Dutch, German and the other compounding languages wait on craft-proof; no ranked gap changed); previously minor (#8: Language packs: every other language's spelling and hyphenation installs from File › Options › Language or the missing-pack bar, through the font store; Lithuanian and Romanian are available as a Language pack once language-packs-1 is published; Dutch, Catalan, French, Greek, Slovak, Serbian and Latvian packs wait on craft-proof engine support); previously trivial (gap 8: English spelling reads SCOWL 2026, one dictionary per English, #40, #520, #5205; no ranked gap changed);; previously trivial (DLine truncated
+> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-11-05 · **Change:** trivial (MATH construction reuse, #3630); previously trivial (DOCX `w:adjustRightInd` defaults true and is written in schema order); previously trivial (ink page lookups reuse scanned paragraph prefixes, #3614); previously trivial (#19: ODT settings for what the standard leaves open removed); previously trivial (#19: ODT element audit; files from other applications checked); previously trivial (#11: the dialogs of the mirrored Arabic interface read right to left: label-and-field rows, tables (list columns, the Models table), tile galleries and the Add/Remove arrows follow the reading direction, through one `rtl::Table` helper that draws exactly as before in English; no ranked gap closed, readiness estimates unchanged); previously trivial (#4: fonts that aren't installed are flagged in the font box and status bar); previously trivial (#19: ODT sections inside text boxes kept as read); previously trivial (#19: an ODT saved without edits keeps what WordCraft doesn't model); previously trivial (#19: ODT interpretation profile, WordCraft / LibreOffice / custom); previously trivial (#19: ODT form controls placed and multi-line; custom shapes keep their geometry); previously trivial (#19: ODT in-line frames placed against the text; template instantiation listed as left); previously trivial (#19: ODT charts kept as read; frame outlines; default header height); previously trivial (#19: ODT tables no longer gain repeated header rows); previously trivial (#19: ODT padding without a border, unequal columns, column breaks, picture-only lines); previously trivial (#19: ODT tables' margins and spacing, header spacing, pictures alone in a paragraph); previously trivial (#19: no LibreOffice template changes page count through a save); previously trivial (#19: ODT variables, emphasis marks, page backgrounds, fixed tables, default size); previously trivial (#19: ODT checked against LibreOffice's own templates: placeholders, first-page styles, unsized pages, metafile logos, `.ott`); previously trivial (#19: ODT at ~90%: custom shapes, cell formulas, macros kept, turned text, control sizes and buttons); previously trivial (#19 narrowed: ODT read and written nearly whole; estimate 18–27 h → 10–16 h); previously trivial (#10 narrowed: text in a turned text box is drawn turned with it); previously trivial (gap 19 narrowed: ODT heading numbering read and written); previously trivial (gap 19 narrowed: ODT table borders and padding read and written); previously trivial (gap 19 narrowed: ODT heading style looks read and written); previously trivial (gap 19 narrowed: ODT tab stops read and written); previously trivial (gap 19 narrowed: ODT character formatting read and written); previously trivial (gap 19 narrowed: ODT list formats read and written); previously trivial (gap 19 narrowed: ODT comments read and written); previously trivial (gap 19 narrowed: ODT footnotes and endnotes read and written); previously trivial (gap 19 narrowed: ODT headers and footers written, text fields and paragraph spacing read and written); previously trivial (#8: Options › Language lists Your languages and Add a language… replaces the Language packs table); previously trivial (#8: the missing-pack bar works on a fresh install and counts the caret's language); previously trivial (#8: Language pack languages are in Set Proofing Language, Review › Language and the status bar; built-in proofing is named as such); previously trivial (#20: layout resource caps and linear-time passes (#3867, #3866)); previously trivial (#3 narrowed: a pasted text box story is charged for the formatting it copies); previously trivial (#3 narrowed: a .doc table row keeps one override per cell); previously trivial (#3 narrowed: a .doc file's sections find their paragraphs by index); previously minor (#11: Arabic interface with mirrored shell landed, #1195; hi/id/pt-PT and Hebrew UI remain); previously trivial (#8: language-packs-2 is published and pinned: 75 packs, adding Danish, German (Germany, Austria, Switzerland), Esperanto, Estonian, Persian, Croatian, Hungarian, Norwegian (Bokmål, Nynorsk), Dutch and Swedish (Sweden, Finland); no ranked gap changed); previously trivial (#8: language-packs-1 is published and pinned, 61 packs: Lithuanian, Romanian, Catalan, French, Greek, Slovak, Serbian and Latvian install as Language packs; Dutch, German and the other compounding languages wait on craft-proof; no ranked gap changed); previously minor (#8: Language packs: every other language's spelling and hyphenation installs from File › Options › Language or the missing-pack bar, through the font store; Lithuanian and Romanian are available as a Language pack once language-packs-1 is published; Dutch, Catalan, French, Greek, Slovak, Serbian and Latvian packs wait on craft-proof engine support); previously trivial (gap 8: English spelling reads SCOWL 2026, one dictionary per English, #40, #520, Line truncated
 > **Highlight update:** 2026-10-25 · Highlight shortcut toggles the last chosen colour off when the whole selection has it; explicit colour commands remain setters (#4085). Readiness estimates unchanged.
 > **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-29 · **Change:** trivial (aligned-copy review fixes retain the document axis through scrolling and include secondary selections (#4764)); previously trivial (#15: bibliography sources keep a corporate author, #4848); previously trivial (Table Properties applies to the selected rows, columns and cells, #4857); previously trivial (Relative font-size commands preserve mixed sizes in the selection (#4903); existing button steps and one-point shortcuts are unchanged; readiness estimates unchanged); previously trivial (Arabic Backspace removes one combining mark and keeps its base letter (#4757); previously trivial (Recipient labels continue across product-sized sheets instead of stopping after the first sheet (#4880)); previously trivial (#7: Custom Document Properties dialog, #2982; previously trivial: egui 0.37 migration: the interface catalogs stay shared and pinned by `craft-i18n.rev` at `9205108cfdced5177706264038848b68b2422d2b` (27 languages, English fallback); input-method handling has new regressions but managed CI passed, and native IME acceptance plus native-speaker catalog review remain open; previously trivial: Markdown code spans keep line-broken text literal, #3738; previously trivial: small follow-ups from the field, table and AutoFormat batch, #4232; previously trivial: highlight target includes selected paragraph marks, #4085; previously trivial: full-target highlight check beyond 10,000 runs, #4085; previously trivial: Exact-spacing table contact uses the specified line height rather than paragraph-mark metrics (#3977); previously trivial: cu20we: restrict pre-spacing float origins to paragraph anchors; document cap helpers and regressions; previously trivial: a command that changes nothing is not an edit: no undo step, no unsaved changes, no relayout, and the redo stack survives it; previously minor: #8 narrowed: Spanish proofing (Spain, Mexico, Latin America, Uruguay) from the installed Hunspell dictionaries, #3360, #3553; previously trivial: gap 10: group connector rerouting preserves transformed free and unresolved endpoints under nonuniform scaling; obstacle routing still missing; readiness estimates unchanged; previously trivial: a word with no break opportunity in it no longer makes the layout quadratic, #3412; previously trivial: the ranked gap list held two rows each for #19, #20 and #21; each gap appears once now; a document with many relationships opens in the time its part takes, #3411; previously trivial: the file commands that change the document are checked on a protected document, the protection of a protected document needs an explicit stop, the session author can't change through `file.setAuthor` under protection (the CLI's `--author` uses the same protection check), and the post-check compares the settings and the metadata; previously trivial: explicit documents keep the editor when recovery discovery finishes; readable recovery controls; previously trivial: text mini toolbar available with the right-click menu, #2959; previously trivial: personal templates in File › New and Save as Word Template, #2852; previously trivial: tracked table/row/cell/section property changes record in WordCraft, narrowing gap 5; previously trivial: a document with many content controls saves in the time the controls take, #13; previously trivial: Format Page Numbers dialog carries Word's number formats and Continue from previous section / Start at, #1113; previously trivial: centred list labels after leading page breaks, #2964; previously trivial: fixed font inputs in four layout controls; previously trivial: a right, centre or decimal tab whose text doesn't fit, or a left tab past the right indent, goes to the next line with its text, #1840; previously trivial: Bullets and Numbering toggle off style-inherited lists at their effective level, including basedOn chains and numbered ListParagraph styles; other paragraphs and list definitions stay unchanged (related to #3459); previously trivial: Russian sorting keeps ё with е and recognises decomposed ё/й; previously trivial: #8: English dialects — en-GB, en-CA, en-AU… flag each other's spellings, #1281; previously trivial: #1: shape gradients and transparency, #1500; previously trivial: text typed into a content control takes the control's formatting; localized placeholder styles recognised; previously trivial: forms protection keeps its content controls: no inserting, Properties or Remove Content Control; previously trivial: clearing a no-match font query keeps the recent heading visible, #2068; previously trivial: line breaking: a slash opening a word travels whole to the next line, #2963; previously trivial: recipient list search and merged-only view, #1576; previously trivial: #2 impact restated: only ink rendering and OLE editing remain; previously trivial: #2 narrowed: native ink nested in groups and drawing canvases kept on save; previously trivial: opening a DOCX no lonLine truncated
 Recovery discovery no longer interrupts an explicitly created or opened document. Older copies remain in File › Recover, whose buttons use the interface palette. See [local recovery](recovery.md).
@@ -142,6 +142,7 @@ Small follow-ups from the 2026-10-25 field, table and AutoFormat batch (#4232):
 - The ribbon's Accept and Reject buttons only open their menus; Word's main half runs Accept (Reject) and Move to Next. The ribbon needs a big split button (Paste needs one too) (#4205).
 
 Ink replay, eraser picking, lasso selection and arrow-candidate discovery reuse the paragraph prefixes they have already scanned (#3614). Page item order, stroke geometry and saved data retain their behavior. This narrows the stability work; ink import fidelity, native pen input and broader corpus qualification remain open.
+MATH glyphs that reference the same construction now share its immutable variants and assembly data (#3630). Glyph mappings, values and the existing partial-record fallback stay unchanged. Repeated glyph IDs keep their last valid reference without retaining superseded constructions. Distinct overlapping constructions and assembly-only sharing remain separate.
 Recipient label generation (#4880) now includes all selected recipients across as many sheets as needed, retaining filtered/sorted order and the chosen product geometry. This fixes the one-sheet omission; Update Labels and Finish & Merge template behavior remain separate, and gap 21 remains open. Readiness estimates are unchanged.
 
 File > Print offers 2, 4, 6, 9 or 16 logical pages per physical sheet side and left/right-bound booklets (#641). Preview and print PDF share the sheet plan, selected pages, parity, copy padding and markup geometry. Custom-range changes reset the preview to the first sheet; Print Markup invalidates preview layout and textures. Native jobs omit link annotations as required by upstream #4734, while saved print PDFs keep their remapped links. Native printing submits copies to the spooler without multiplying them in the PDF. Ordinary PDF export remains single-page by default. Translations are reviewed separately in craft-i18n#38. Physical-printer qualification and driver-specific options remain unverified.
@@ -166,6 +167,7 @@ Table rows and cells inside `mc:AlternateContent` use the same selected-branch p
 |---|---|---|
 | 2026-11-05 | trivial | DOCX preserves `w:adjustRightInd` with its true default and writes it in schema order; grid layout remains open |
 | 2026-10-28 | trivial | Ink page lookups reuse scanned paragraph prefixes, #3614. Drawing order, geometry and saved data retain their behavior; no readiness estimate change |
+| 2026-10-28 | trivial | Shared MATH construction records reuse variants and assembly data without changing glyph mappings or existing limits (#3630); readiness estimates unchanged |
 | 2026-11-04 | trivial | #19: ODT settings for what the standard leaves open removed: one choice for each case |
 | 2026-11-04 | trivial | #19: ODT element audit; files from other applications checked |
 | 2026-11-04 | trivial | PR #4609: relative REF keeps context already displayed by the own or a needed ancestor label; six synthetic legal-template cases and separate-list context follow the Word for Windows measurements reported by @pakiat in PR discussion. Existing unknown-switch/cache boundaries and readiness estimates unchanged; no new native Word run. |

docs/layout-parity.md

@@ -1,6 +1,6 @@
 # Layout and pagination parity with Microsoft Word
 
-> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-11-04 · **Change:** trivial (ODT documents: a line of only pictures keeps its font's height; no setting for it any more); previously trivial (ODT documents: padding is the standard's box model always; picture-only line height is an ODF interpretation setting); previously trivial (行頭禁則 also covers the small kana and the prolonged sound mark at ordinary break opportunities, as JIS X 4051 §4 specifies, inferred from the standard and synthetic tests rather than checked in Word); previously trivial (a missing font is drawn in an installed metric twin before any other substitute: Arial for Helvetica on Windows, Tinos before macOS Times for Times New Roman); previously trivial (ODT documents: padding without a border takes room); previously trivial (ODT documents: a line of only pictures is as tall as they are); previously trivial (ODT documents: a table's own space above and below; header and footer spacing kept past a tall header); previously trivial (a page break before a table's first row moves the table; emphasis marks drawn with room in the line); previously trivial (text in a turned text box turns with it; previously trivial: an in-line picture, shape or chart moves with its run's raise or lowering, Word's `w:position`; previously trivial (layout work bounds: header-row repeats budgeted per layout, cached picture outlines charged to the decode budget, linked-box chain widths exact up to a work cap, #3867); previously trivial (the line between columns runs as tall as the section's text, so columns for selected text have it, #5178); previously trivial (a table row that would leave a cell empty on a page that already holds something, its picture or first line not fitting there, moves on whole instead of leaving the other cells' text beside the empty cell, #4272); previously trivial (a middle or bottom cell of a row split across pages lines up on the first page); previously trivial (tab leader dots spaced and placed as in Word); previously trivial (a level above that hasn't counted yet takes its first number at the first deeper item); previously trivial (paragraph side borders lie beyond their distance from the text, #5036); previously trivial (with a header taller than the top margin, Margin-anchored drawings, header-cell pins and header-table obstacles use the rendered body top; TopMargin and InsideMargin keep their declared band, #2485; previously trivial: connector rerouting indexes laid-out object positions once per batch (#3631); previously trivial (field results take the `\* DollarText`, `\* Hex` and `\* ArabicDash` switches, #4226; previously trivial: a cell spanning columns shares its extra minimum width by the columns' preferred widths, equally in automatic tables, #1883; previously trivial: continuous sections one after another go onto new pages instead of running off the page; empty marks after another empty section mark take a line each until the page is full, then stay at its bottom drawing nothing, not even a list label; an empty mark that takes no room draws no list label either; a kept-together paragraph is measured against the page it would move to, so a tall different first-page header no longer splits it, #3907; previously trivial: a justified line ending in a manual line break is stretched to the margin unless `w:doNotExpandShiftReturn` is set, #3924; previously trivial: a no-break space is as wide as a space, #4192; previously trivial: a bookmark or other zero-width marker is no line-break opportunity, #4193; previously trivial: Arabic list labels: `arabicAlpha` and `arabicAbjad` draw the 28 Arabic letters (alphabetical and abjad order), doubling past the last letter; Hebrew, Hindi, Thai and Vietnamese still show decimal (`doc::section::tests::arabic_letters_count_in_both_orders`); previously trivial: Preserve grapheme edit/wrap boundaries across independently drawn font/style pieces; Exact-spacing table contact uses the specified line height rather than paragraph-mark metrics (#3977); previously trivial: cu20we: restrict pre-spacing float origins to paragraph anchors; document cap helpers and regressions; previously trivial: cu20we: outline caps; previously trivial: cu20we: box float paragraph origin; previously trivial: group connector rerouting preserves transformed free and unresolved endpoints under nonuniform scaling; readiness estimates unchanged; previously trivial: a word with no break opportunity in it no longer makes the layout quadratic, #3412; previously trivial: an empty paragraph whose mark is hidden shows nothing, not even its list number, which is still counted, #3982; previously trivial: a Thai run with Latin finds one installed font covering it all on Linux too, #1549; previously trivial: symbol fonts keep their line height without the font; previously trivial: an equation-array row without `&` is centred, #2617; previously trivial: a superscript or subscript sizes its line at the full font size, withouLine truncated
+> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-11-04 · **Change:** trivial (MATH construction reuse, #3630; previously trivial: continuous sections one after another go onto new pages instead of running off the page; empty marks after another empty section mark take a line each until the page is full, then stay at its bottom drawing nothing, not even a list label; an empty mark that takes no room draws no list label either; a kept-together paragraph is measured against the page it would move to, so a tall different first-page header no longer splits it, #3907; previously trivial: a justified line ending in a manual line break is stretched to the margin unless `w:doNotExpandShiftReturn` is set, #3924; previously trivial: a no-break space is as wide as a space, #4192; previously trivial: a bookmark or other zero-width marker is no line-break opportunity, #4193; previously trivial: Arabic list labels: `arabicAlpha` and `arabicAbjad` draw the 28 Arabic letters (alphabetical and abjad order), doubling past the last letter; Hebrew, Hindi, Thai and Vietnamese still show decimal (`doc::section::tests::arabic_letters_count_in_both_orders`); previously trivial (ODT documents: a line of only pictures keeps its font's height; no setting for it any more); previously trivial (ODT documents: padding is the standard's box model always; picture-only line height is an ODF interpretation setting); previously trivial (行頭禁則 also covers the small kana and the prolonged sound mark at ordinary break opportunities, as JIS X 4051 §4 specifies, inferred from the standard and synthetic tests rather than checked in Word); previously trivial (a missing font is drawn in an installed metric twin before any other substitute: Arial for Helvetica on Windows, Tinos before macOS Times for Times New Roman); previously trivial (ODT documents: padding without a border takes room); previously trivial (ODT documents: a line of only pictures is as tall as they are); previously trivial (ODT documents: a table's own space above and below; header and footer spacing kept past a tall header); previously trivial (a page break before a table's first row moves the table; emphasis marks drawn with room in the line); previously trivial (text in a turned text box turns with it; previously trivial: an in-line picture, shape or chart moves with its run's raise or lowering, Word's `w:position`; previously trivial (layout work bounds: header-row repeats budgeted per layout, cached picture outlines charged to the decode budget, linked-box chain widths exact up to a work cap, #3867); previously trivial (the line between columns runs as tall as the section's text, so columns for selected text have it, #5178); previously trivial (a table row that would leave a cell empty on a page that already holds something, its picture or first line not fitting there, moves on whole instead of leaving the other cells' text beside the empty cell, #4272); previously trivial (a middle or bottom cell of a row split across pages lines up on the first page); previously trivial (tab leader dots spaced and placed as in Word); previously trivial (a level above that hasn't counted yet takes its first number at the first deeper item); previously trivial (paragraph side borders lie beyond their distance from the text, #5036); previously trivial (with a header taller than the top margin, Margin-anchored drawings, header-cell pins and header-table obstacles use the rendered body top; TopMargin and InsideMargin keep their declared band, #2485; previously trivial: connector rerouting indexes laid-out object positions once per batch (#3631); previously trivial (field results take the `\* DollarText`, `\* Hex` and `\* ArabicDash` switches, #4226; previously trivial: a cell spanning columns shares its extra minimum width by the columns' preferred widths, equally in automatic tables, #1883; previously trivial: continuous sections one after another go onto new pages instead of running off the page; empty marks after another empty section mark take a line each until the page is full, then stay at its bottom drawing nothing, not even a list label; an empty mark that takes no room draws no list label either; a kept-together paragraph is measured against the page it would move to, so a tall different first-page header no longer splits it, #3907; previously trivial: a justified line ending in a manual line break is stretched to the margin unless `w:doNotExpandShiftReturn` is set, #3924; previously trivial: a no-break space is as wide as a space, #4192; previously trivial: a bookmark or other zero-width marker is no line-break opportunity, #4193; previously trivial: Arabic list labels: `arabicAlpha` and `arabicAbjad` draw the 28 Arabic letters (alphabetical and abjad order), doubling past the last letter; Hebrew, Hindi, Thai and Vietnamese still show decimal (`doc::section::tests::arabic_letters_count_in_both_orders`); previously trivial: Preserve grapheme edit/wrap boundaries across inLine truncated
 
 A document only "opens correctly" if its pages break where Word breaks them: the same lines, the
 same page count, footnotes and pictures on the same pages. This checklist covers line breaking,
@@ -263,6 +263,8 @@ new mixed-style emoji and mixed bundled-font combining-mark controls cover LTR/b
 hit stops, full selection, and narrow/overwide line wrapping. Independent review and the full
 layout suite are pending; this is not a new Word/native-IME acceptance claim.
 
+MATH glyphs that reference the same construction now share its immutable variants and assembly data (#3630). Glyph mappings, values and the existing partial-record fallback stay unchanged. Repeated glyph IDs keep their last valid reference without retaining superseded constructions. Distinct overlapping constructions and assembly-only sharing remain separate.
+
 ## Revision history
 
 > 2026-11-01 · Text in a turned text box is drawn turned with it, about its centre, on screen and in PNG and PDF export; flipped across, upside down; flipped side to side, still readable (as Word draws it). Clicks and the caret inside a turned box still go by its upright text (`layout::tests::a_turned_text_box_turns_its_text_with_it`).
@@ -284,6 +286,7 @@ layout suite are pending; this is not a new Word/native-IME acceptance claim.
 | Date | Change | Summary |
 |---|---|---|
 | 2026-11-04 | trivial | PR #4609: relative REF keeps context already displayed by the own or a needed ancestor label; six synthetic legal-template cases and separate-list context follow the Word for Windows measurements reported by @pakiat in PR discussion. Existing unknown-switch/cache boundaries and readiness estimates unchanged; no new native Word run. |
+| 2026-10-28 | trivial | Shared MATH construction records reuse variants and assembly data without changing glyph mappings or existing limits (#3630); readiness estimates unchanged |
 | 2026-11-01 | trivial | SEQ fields in text boxes are numbered where their boxes are anchored, in output and after F9, and a REF finds a bookmark in a text box |
 | 2026-11-01 | trivial | SEQ, STYLEREF and REF fields show fresh results in output; readiness estimates unchanged |
 | 2026-10-29 | trivial | Print Markup balloon anchor discovery walks paragraph objects once (#3614); readiness estimates unchanged |