Opening a DOCX no longer copies its pictures (#2441)

Each picture, and each embedded object or macro part kept for saving, was
copied out of the unzipped package (bytes.to_vec()), so a document with
pictures held them twice while it opened, and the package's copies were
left behind as freed memory. Package now holds its parts as Arc<Vec<u8>>,
the type the document keeps media and passthrough parts in, and hands them
over with get_shared. keep_unmodelled also searches the main part in place
instead of copying it.

A 92 MB document with 20 pictures on macOS 27: the app's footprint once
open 866 -> 772 MB (freed blocks kept 206 -> 118 MB, peak 915-932 ->
821-823 MB); wordcraft-cli info peak 319 -> 231 MB.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

ROADMAP.md

@@ -2,7 +2,7 @@
 
 **Stage: alpha** · next: beta, ~15 points (60% → 75% ready for real work) and ~150–230 h away
 
-> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-18 · **Change:** trivial (Add Font imports every selected font file, #1789); previously trivial (RTF export writes sections (gap 19); previously trivial: ODT export writes sections (gap 19); previously trivial: Read Aloud speaks each sentence in its own language, #1843; previously trivial: hint-reliant CJK outlines; previously trivial: Ctrl+wheel zoom no longer scrolls the document, #1743; previously trivial: progress log: Turkish case mapping in AutoCorrect, #1923; previously trivial: a mouse drag on a table leaves the view where it is, #1738; previously trivial: progress log: Hebrew field results and smart quotes, #1670; previously trivial: Backspace removes a selected table, rows or columns, #1844; previously trivial: progress log: mixed Hebrew/Latin text keeps its order in Word, #1543; previously trivial: progress log: web typing in non-Latin text no longer repeats, #1480; previously trivial: progress log: font family caches bounded and math font lookups follow font changes, #1354; previously trivial: `-->` AutoCorrect replacement, #2223; previously trivial: numbered headings, #1793; previously trivial: Navigation Find exposes Results, #2096; previously trivial: progress log: Columns on selected text, #2024; previously trivial: progress log: legacy outline, shadow, emboss and imprint text effects drawn, #2534; previously trivial: progress log: Indonesian proofing, #498; previously trivial: progress log: DOCX glossary documents (building blocks); previously trivial: progress log: Read Mode keeps the keyboard and shows whole, unobscured columns, #1772); previously trivial: progress log: font list entries of tall-ascent fonts no longer cut off, #2570); previously trivial: progress log: Split Cells and Convert Text to Table dialogs follow the Word check, #2592; previously trivial: a new source's Tag is made like Word's, three letters of the author and two digits of the year with a number added when taken, and an edit keeps the Tag, #2094; previously trivial: Web Layout fills the window width, #1876; previously trivial: progress log: Different First Page headers reachable from page 1, #2826; previously trivial: Language dialog keeps inherited proofing language and no-proof choices (#2717); previously trivial: progress log: web preferences kept between visits, #2132; previously trivial: progress log: undrawable private-use list bullets show the level's default bullet, #1683; previously trivial: progress log: `w:eastAsianLayout` kept on save; previously trivial: table style editor formats every part, #714); previously trivial: VML freeforms read and kept on save; previously trivial: progress log: consistent preferred cell widths; previously trivial: multi-paragraph field instructions no longer add blank lines; previously trivial: page-break-only list paragraphs hide their labels; previously trivial: progress log: grouped `flowChartOr` shapes draw their circle and crossing spokes; previously trivial: progress log: copies keep their formatting in other programs, #1523; previously trivial: progress log: privacy options on save, #1062; previously trivial: progress log: equation pictures in theme colours on Windows, #1002; previously trivial: symbol fonts, #776; previously trivial: the save prompt answers from the keyboard, #861; previously trivial: table column and row markers on the rulers, #700; previously trivial: table borders and the resize handle drag on the page, #217; previously trivial: formatting revisions landed, #41; previously trivial: Draw tab ink, #307, and Draw Table and Eraser, #303, landed: catalog 402/434; previously minor: readiness table with hours per audience; full number back to the additive weighted sum, 60%; mainstream and essentials numbers added; alpha gate checked; previously major: full re-measure against Word for Mac 16.113.4; restructured to the craftrules progress-docs standard) · **Target:** Microsoft Word (Microsoft 365) for Mac 16.113.4
+> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-18 · **Change:** trivial (opening a DOCX no longer copies its pictures; previously trivial: Add Font imports every selected font file, #1789); previously trivial (RTF export writes sections (gap 19); previously trivial: ODT export writes sections (gap 19); previously trivial: Read Aloud speaks each sentence in its own language, #1843; previously trivial: hint-reliant CJK outlines; previously trivial: Ctrl+wheel zoom no longer scrolls the document, #1743; previously trivial: progress log: Turkish case mapping in AutoCorrect, #1923; previously trivial: a mouse drag on a table leaves the view where it is, #1738; previously trivial: progress log: Hebrew field results and smart quotes, #1670; previously trivial: Backspace removes a selected table, rows or columns, #1844; previously trivial: progress log: mixed Hebrew/Latin text keeps its order in Word, #1543; previously trivial: progress log: web typing in non-Latin text no longer repeats, #1480; previously trivial: progress log: font family caches bounded and math font lookups follow font changes, #1354; previously trivial: `-->` AutoCorrect replacement, #2223; previously trivial: numbered headings, #1793; previously trivial: Navigation Find exposes Results, #2096; previously trivial: progress log: Columns on selected text, #2024; previously trivial: progress log: legacy outline, shadow, emboss and imprint text effects drawn, #2534; previously trivial: progress log: Indonesian proofing, #498; previously trivial: progress log: DOCX glossary documents (building blocks); previously trivial: progress log: Read Mode keeps the keyboard and shows whole, unobscured columns, #1772); previously trivial: progress log: font list entries of tall-ascent fonts no longer cut off, #2570); previously trivial: progress log: Split Cells and Convert Text to Table dialogs follow the Word check, #2592; previously trivial: a new source's Tag is made like Word's, three letters of the author and two digits of the year with a number added when taken, and an edit keeps the Tag, #2094; previously trivial: Web Layout fills the window width, #1876; previously trivial: progress log: Different First Page headers reachable from page 1, #2826; previously trivial: Language dialog keeps inherited proofing language and no-proof choices (#2717); previously trivial: progress log: web preferences kept between visits, #2132; previously trivial: progress log: undrawable private-use list bullets show the level's default bullet, #1683; previously trivial: progress log: `w:eastAsianLayout` kept on save; previously trivial: table style editor formats every part, #714); previously trivial: VML freeforms read and kept on save; previously trivial: progress log: consistent preferred cell widths; previously trivial: multi-paragraph field instructions no longer add blank lines; previously trivial: page-break-only list paragraphs hide their labels; previously trivial: progress log: grouped `flowChartOr` shapes draw their circle and crossing spokes; previously trivial: progress log: copies keep their formatting in other programs, #1523; previously trivial: progress log: privacy options on save, #1062; previously trivial: progress log: equation pictures in theme colours on Windows, #1002; previously trivial: symbol fonts, #776; previously trivial: the save prompt answers from the keyboard, #861; previously trivial: table column and row markers on the rulers, #700; previously trivial: table borders and the resize handle drag on the page, #217; previously trivial: formatting revisions landed, #41; previously trivial: Draw tab ink, #307, and Draw Table and Eraser, #303, landed: catalog 402/434; previously minor: readiness table with hours per audience; full number back to the additive weighted sum, 60%; mainstream and essentials numbers added; alpha gate checked; previously major: full re-measure against Word for Mac 16.113.4; restructured to the craftrules progress-docs standard) · **Target:** Microsoft Word (Microsoft 365) for Mac 16.113.4
 > **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-17 · **Change:** trivial (RTF export writes sections (gap 19); previously trivial: ODT export writes sections (gap 19); previously trivial: progress log: pasted blockquotes are indented paragraphs, #1270; previously trivial: progress log: print-area strings translated in every catalog; previously trivial: progress log: comments readable in the dark theme, #2045; previously trivial: progress log: ODT sections, #1379; previously trivial: progress log: table deletes, Change Case and Unlink Fields respect Track Changes and locked content controls, #1329; previously trivial: progress log: rounded rectangle corners, #1476; previously trivial: progress log: Save As type list, #1399; previously trivial: progress log: privacy options on save, #1062; previously trivial: progress log: equation pictures in theme colours on Windows, #1002; previously trivial: symbol fonts, #776; previously trivial: the save prompt answers from the keyboard, #861; previously trivial: table column and row markers on the rulers, #700; previously trivial: table borders and the resize handle drag on the page, #217; previously trivial: formatting revisions landed, #41; previously trivial: Draw tab ink, #307, and Draw Table and Eraser, #303, landed: catalog 402/434; previously minor: readiness table with hours per audience; full number back to the additive weighted sum, 60%; mainstream and essentials numbers added; alpha gate checked; previously major: full re-measure against Word for Mac 16.113.4; restructured to the craftrules progress-docs standard) · **Target:** Microsoft Word (Microsoft 365) for Mac 16.113.4
 > **Printing update:** 2026-10-12 · direct desktop printing through craft-print; details below.
 
@@ -200,6 +200,7 @@ Imported Word 97–2003 `.doc`/`.dot` files now ask for a supported DOCX target
 | 2026-10-14 | Consistent preferred cell widths override a stale table grid when both totals agree, for fixed and AutoFit tables. Conflicting widths, spans and percentages remain unchanged. |
 | 2026-10-14 | Legacy VML freeforms (`v:shape` with its own `path`), such as the rule lines in many older headers, are read as freeform shapes: they show on screen and in PDFs and are saved back as DrawingML custom geometry instead of vanishing |
 | 2026-10-14 | Field instructions spanning paragraphs no longer add blank lines during pagination; empty result paragraphs still take a line. |
+| 2026-10-16 | Opening a DOCX no longer copies its pictures: each picture, and each embedded object or macro part kept for saving, was copied out of the unzipped package, so a document with pictures held them twice while it opened and left the package's copies behind as freed memory. The document now shares the package's bytes: a 92 MB document with 20 pictures opens in about 90 MB less (866 → 772 MB in the app on macOS; the command line's peak, 319 → 231 MB) |
 | 2026-10-14 | Page-break-only list paragraphs no longer print stray bullets or numbers. The next item keeps its number, matching the reference export. |
 | 2026-10-12 | Backspace at the start of the paragraph after a table nested in a cell moves into the nested table's last cell, no longer into the last cell of the table around it |
 | 2026-10-14 | Grouped `flowChartOr` shapes draw their circle and crossing spokes, as in Word |
@@ -518,6 +519,7 @@ Imported Word 97–2003 `.doc`/`.dot` files now ask for a supported DOCX target
 | 2026-10-12 | minor | Czech proofing review: short-word splits rank after deletion, NOSUGGEST excludes suggestions while accepting typed forms, affix comments apply only to headers, ratios keep spaces and repeated-word messages use Czech quotes; synthetic engine end-to-end and latency checks |
 | 2026-10-12 | minor | Czech spelling from the installed Hunspell dictionary, runtime-table suggestions and generic checks; Czech interface writes cs-CZ documents; no Czech hyphenation patterns bundled; estimates unchanged |
 | 2026-10-12 | trivial | Progress log: one selected cell's border drags in its row only (#962) |
+| 2026-10-16 | trivial | Progress log: opening a DOCX no longer copies its pictures |
 | 2026-10-12 | minor | Progress log: Word default shortcuts expanded to 140 commands on macOS (169 combinations) and 158 on Linux/Windows (188 combinations), addressing gap #24 |
 | 2026-10-13 | trivial | Web/Draft long-page blur fixed with visible full-resolution raster tiles |
 | 2026-10-14 | trivial | Progress log: VML freeforms read and kept on save |

crates/docx/src/package.rs

@@ -2,6 +2,7 @@
 
 use std::collections::{BTreeMap, HashMap};
 use std::io::{Cursor, Read, Write};
+use std::sync::Arc;
 
 use crate::DocxError;
 use crate::xml::{self, El};
@@ -186,9 +187,10 @@ pub fn rel_is(t: &str, full: &str) -> bool {
     t == full || tail(t) == tail(full)
 }
 
-/// The decompressed parts of a package (bounded).
+/// The decompressed parts of a package (bounded). Shared, so a part the document keeps as it is (a
+/// picture, an embedded object) is handed over rather than copied.
 pub struct Package {
-    files: BTreeMap<String, Vec<u8>>,
+    files: BTreeMap<String, Arc<Vec<u8>>>,
 }
 
 impl Package {
@@ -230,7 +232,7 @@ impl Package {
                 return Err(DocxError::Limit(format!("part {name} decompresses past the size limit")));
             }
             total = total.saturating_add(buf.len() as u64);
-            files.insert(name, buf);
+            files.insert(name, Arc::new(buf));
         }
         Ok(Package { files })
     }
@@ -238,16 +240,22 @@ impl Package {
     /// A package from parts already in memory (a Flat OPC file, a converted document), by part
     /// name without the leading `/`.
     pub fn from_parts(files: BTreeMap<String, Vec<u8>>) -> Package {
-        Package { files }
+        Package { files: files.into_iter().map(|(name, bytes)| (name, Arc::new(bytes))).collect() }
     }
 
     /// Part bytes by name (leading `/` ignored; case-insensitive fallback).
     pub fn get(&self, name: &str) -> Option<&[u8]> {
+        self.get_shared(name).map(|v| v.as_slice())
+    }
+
+    /// [`Package::get`], as the shared bytes themselves: for a part the document keeps whole, which
+    /// then costs no copy (a picture document's pictures used to be held twice while it opened).
+    pub fn get_shared(&self, name: &str) -> Option<&Arc<Vec<u8>>> {
         let n = name.trim_start_matches('/');
         if let Some(v) = self.files.get(n) {
             return Some(v);
         }
-        self.files.iter().find(|(k, _)| k.eq_ignore_ascii_case(n)).map(|(_, v)| v.as_slice())
+        self.files.iter().find(|(k, _)| k.eq_ignore_ascii_case(n)).map(|(_, v)| v)
     }
 
     pub fn xml(&self, name: &str) -> Result<Option<El>, DocxError> {
@@ -432,6 +440,27 @@ mod tests {
         }
     }
 
+    #[test]
+    fn the_document_shares_its_pictures_with_the_package() {
+        // Each picture was copied out of the package, so a picture document held its pictures
+        // twice while it opened, and left the package's copies behind as freed memory.
+        let ns = r#"xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main" xmlns:r="http://schemas.openxmlformats.org/officeDocument/2006/relationships" xmlns:wp="http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing" xmlns:a="http://schemas.openxmlformats.org/drawingml/2006/main" xmlns:pic="http://schemas.openxmlformats.org/drawingml/2006/picture""#;
+        let document = format!(
+            r#"<w:document {ns}><w:body><w:p><w:r><w:drawing><wp:inline><wp:extent cx="635000" cy="635000"/><a:graphic><a:graphicData uri="http://schemas.openxmlformats.org/drawingml/2006/picture"><pic:pic><pic:blipFill><a:blip r:embed="rIdImg"/></pic:blipFill></pic:pic></a:graphicData></a:graphic></wp:inline></w:drawing></w:r></w:p></w:body></w:document>"#
+        );
+        let rels = r#"<Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rIdImg" Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image" Target="media/pic.png"/></Relationships>"#;
+        let z = zip_of(&[
+            ("word/document.xml", document.into_bytes()),
+            ("word/_rels/document.xml.rels", rels.as_bytes().to_vec()),
+            ("word/media/pic.png", b"\x89PNG picture bytes".to_vec()),
+        ]);
+        let pkg = Package::open(&z).unwrap();
+        let doc = crate::read::read_opened(&pkg).unwrap();
+        let held = doc.media.values().next().expect("the picture is read");
+        assert!(Arc::ptr_eq(held, pkg.get_shared("word/media/pic.png").unwrap()), "the same bytes, not a copy");
+        assert_eq!(pkg.get("/WORD/media/PIC.png"), Some(&b"\x89PNG picture bytes"[..]), "lookups are unchanged");
+    }
+
     #[test]
     fn resolves_targets() {
         assert_eq!(resolve("word", "media/image1.png"), "word/media/image1.png");

crates/docx/src/read/embed.rs

@@ -94,7 +94,8 @@ impl Reader<'_> {
     /// relationship the main document's markup refers to by id (its target goes with the
     /// element, or not at all).
     pub(super) fn keep_unmodelled(&mut self, package: &Rels, document: &Rels) {
-        let main = self.pkg.get(&self.main).map(|b| b.to_vec()).unwrap_or_default();
+        let pkg = self.pkg;
+        let main = pkg.get(&self.main).unwrap_or_default();
         let referenced = |id: &str| {
             let needle = format!("\"{id}\"");
             main.windows(needle.len()).any(|w| w == needle.as_bytes())
@@ -195,14 +196,14 @@ impl Reader<'_> {
             log::warn!("docx: more than {MAX_EMBEDDED_PARTS} parts for embedded objects; the rest are dropped");
             return false;
         }
-        let Some(bytes) = self.pkg.get(part) else {
+        let Some(bytes) = self.pkg.get_shared(part) else {
             log::warn!("docx: part {part} of an embedded object is missing");
             return false;
         };
         let pkg = self.pkg;
         let ct = self.content_types.get_or_insert_with(|| ContentTypes::read(pkg)).of(part).unwrap_or("application/octet-stream");
         let content_type = if manifest_field_ok(ct) { ct.to_string() } else { "application/octet-stream".to_string() };
-        self.doc.passthrough.entry(part.to_string()).or_insert_with(|| Arc::new(bytes.to_vec()));
+        self.doc.passthrough.entry(part.to_string()).or_insert_with(|| bytes.clone());
         self.embedded.parts.insert(part.to_string(), crate::package::EmbeddedPart { content_type, rels: Vec::new() });
         true
     }

crates/docx/src/read/mod.rs

@@ -282,9 +282,9 @@ pub(crate) fn read_opened(pkg: &Package) -> Result<Document, DocxError> {
     // bytes, never parsed or run, so a .docm/.dotm saved again carries exactly what the file had.
     // Only the relationships and content types are read.
     if let Some(v) = part(rt::VBA_PROJECT)
-        && let Some(b) = pkg.get(&v).filter(|b| !b.is_empty())
+        && let Some(b) = pkg.get_shared(&v).filter(|b| !b.is_empty())
     {
-        r.doc.passthrough.insert(VBA_PROJECT_PART.into(), Arc::new(b.to_vec()));
+        r.doc.passthrough.insert(VBA_PROJECT_PART.into(), b.clone());
         r.read_vba_related(&v);
     }
     r.read_glossary(&rels);
@@ -579,7 +579,7 @@ impl Reader<'_> {
                 log::warn!("docx: not carrying VBA-related part {path:?}");
                 continue;
             }
-            let Some(bytes) = self.pkg.get(path) else {
+            let Some(bytes) = self.pkg.get_shared(path) else {
                 log::warn!("docx: VBA-related part {path} is missing; keeping the project without it");
                 continue;
             };
@@ -589,7 +589,7 @@ impl Reader<'_> {
             }
             kept += 1;
             manifest.push_str(&format!("{}\t{path}\t{ct}\n", rel.kind));
-            self.doc.passthrough.entry(path.to_string()).or_insert_with(|| Arc::new(bytes.to_vec()));
+            self.doc.passthrough.entry(path.to_string()).or_insert_with(|| bytes.clone());
         }
         if !manifest.is_empty() {
             self.doc.passthrough.insert(VBA_RELATED.into(), Arc::new(manifest.into_bytes()));
@@ -644,7 +644,7 @@ impl Reader<'_> {
         if let Some(k) = self.media_by_path.get(&rel.target) {
             return Some(k.clone());
         }
-        let bytes = self.pkg.get(&rel.target)?;
+        let bytes = self.pkg.get_shared(&rel.target)?;
         let base: String =
             rel.target.rsplit('/').next().unwrap_or("image").chars().filter(|c| c.is_ascii_alphanumeric() || matches!(c, '.' | '_' | '-')).collect();
         let base = if base.is_empty() || base.starts_with('.') { format!("image{base}") } else { base };
@@ -657,7 +657,7 @@ impl Reader<'_> {
                 None => format!("{base}_{n}"),
             };
         }
-        self.doc.media.insert(key.clone(), Arc::new(bytes.to_vec()));
+        self.doc.media.insert(key.clone(), bytes.clone());
         self.media_by_path.insert(rel.target.clone(), key.clone());
         Some(key)
     }

docs/gaps.md

@@ -1,6 +1,6 @@
 # Where WordCraft falls short of Word
 
-> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-18 · **Change:** trivial (RTF export writes sections (gap 19); previously trivial: ODT export writes sections (gap 19); previously trivial: automatic Asian/Latin and Asian/number spacing laid out, #2358; previously trivial: hint-reliant CJK outlines; previously trivial: numbered headings, #1793; previously trivial: legacy outline, shadow, emboss and imprint text effects drawn, #2534; previously trivial: #8 narrowed: Indonesian proofing from the installed Hunspell dictionary, #498; previously trivial: glossary/building blocks closed; previously trivial: Split Cells and Convert Text to Table follow the Word check, #2592; previously trivial: Language dialog keeps inherited proofing language and no-proof choices (#2717); previously trivial: #29: a copied equation also goes in the MathML clipboard formats, #1297; previously trivial: #24: the Spike, Ctrl+F3 and Ctrl+Shift+F3; previously trivial: VML freeforms read and kept on save; previously trivial: consistent preferred cell widths; previously trivial: multi-paragraph field instructions no longer add blank lines; previously trivial: page-break-only list paragraphs hide their labels; previously trivial: floating-table overlap ignores empty line-break lines; previously trivial: copies keep their formatting on the clipboard as HTML, #1523; previously trivial: sample-text formulas on Enter, #1067; previously trivial: #29: copied equations as MathML, OMML open as #1297, #816; previously trivial: the Font list no longer keeps every previewed font in memory, #848; previously trivial: table column and row markers on the rulers, #700; previously trivial: table borders and the resize handle drag on the page, #217; previously trivial: formatting revisions landed, #41; previously trivial: Draw Table and Eraser landed, #303; previously minor: beta distance restated: ~15 points, ~150–230 h; previously: first version; every known shortfall from the 2026-10-10 re-measure, ranked) · **Target:** Microsoft Word (Microsoft 365) for Mac 16.113.4
+> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-18 · **Change:** trivial (opening a DOCX no longer copies its pictures; previously trivial: RTF export writes sections (gap 19); previously trivial: ODT export writes sections (gap 19); previously trivial: automatic Asian/Latin and Asian/number spacing laid out, #2358; previously trivial: hint-reliant CJK outlines; previously trivial: numbered headings, #1793; previously trivial: legacy outline, shadow, emboss and imprint text effects drawn, #2534; previously trivial: #8 narrowed: Indonesian proofing from the installed Hunspell dictionary, #498; previously trivial: glossary/building blocks closed; previously trivial: Split Cells and Convert Text to Table follow the Word check, #2592; previously trivial: Language dialog keeps inherited proofing language and no-proof choices (#2717); previously trivial: #29: a copied equation also goes in the MathML clipboard formats, #1297; previously trivial: #24: the Spike, Ctrl+F3 and Ctrl+Shift+F3; previously trivial: VML freeforms read and kept on save; previously trivial: consistent preferred cell widths; previously trivial: multi-paragraph field instructions no longer add blank lines; previously trivial: page-break-only list paragraphs hide their labels; previously trivial: floating-table overlap ignores empty line-break lines; previously trivial: copies keep their formatting on the clipboard as HTML, #1523; previously trivial: sample-text formulas on Enter, #1067; previously trivial: #29: copied equations as MathML, OMML open as #1297, #816; previously trivial: the Font list no longer keeps every previewed font in memory, #848; previously trivial: table column and row markers on the rulers, #700; previously trivial: table borders and the resize handle drag on the page, #217; previously trivial: formatting revisions landed, #41; previously trivial: Draw Table and Eraser landed, #303; previously minor: beta distance restated: ~15 points, ~150–230 h; previously: first version; every known shortfall from the 2026-10-10 re-measure, ranked) · **Target:** Microsoft Word (Microsoft 365) for Mac 16.113.4
 
 > **Printing update:** 2026-10-12 · direct desktop printing through craft-print; details below.
 
@@ -49,7 +49,7 @@ Legacy Word imports (#2445): interactive Save for `.doc`/`.dot` uses the existin
 | 17 | **East Asian typography**: hint-reliant DFKai-SB contours repaired (local Windows regression; no readiness estimate change); Asian Typography options (Phonetic Guide landed, #288; Enclose Characters landed, #297; document grid landed, #391; vertical text landed, #491) | | `para.asianTypography` is live (Paragraph dialog Asian Typography tab, #323), automatic Asian/Latin and Asian/number spacing is laid out (#2358), and no East Asian item is left in the catalog's missing list; the remaining depth is not measured | Chinese, Japanese and Korean documents | 11–19 | [layout-parity.md](layout-parity.md) |
 | 18 | **Screen-reader access** to the document canvas: exposed, not yet tried with real screen readers | | The canvas describes paragraphs, headings, lists, tables, links, pictures and the caret to AccessKit (#488), tested headlessly; Windows Magnifier follows the caret, onto a new line too (#1119); no NVDA, VoiceOver or Orca run yet | Blind and low-vision users; public-sector procurement | 3–6 | [ui-parity.md](ui-parity.md) |
 | 19 | **RTF and ODT depth**: notes, comments, revisions (RTF and ODT insertions and deletions done), sections (RTF and ODT import and export done, #1378, #1379), RTL. ODT export writes sections back (a master page per new-page section, a columned `text:section` per continuous one); RTF export writes every section (`\sect`, `\sectd`) | | ~2,100 (RTF) and ~1,600 (ODT) lines in `crates/formats`; RTF code pages (#896) and page setup (#902), ODT page setup (#829) landed; open PRs: RTF notes and comments (#1056), ODT notes and comments (#1057), ODT headers and footers (#873) | Users exchanging with LibreOffice or older tools | 18–27 | [file-format-parity.md](file-format-parity.md) |
-| 20 | **Performance on large real-world documents** unmeasured (500+ pages, many pictures) | | Only the 188-page sample benchmark; scrolling a 1,096-page one-table document went from ~600 ms to ~2–5 ms a frame (#811), but opening it still lays out for seconds | Thesis and book authors | 10–20 | [layout-parity.md](layout-parity.md) |
+| 20 | **Performance on large real-world documents** unmeasured (500+ pages, many pictures) | | Only the 188-page sample benchmark; scrolling a 1,096-page one-table document went from ~600 ms to ~2–5 ms a frame (#811), but opening it still lays out for seconds; opening a document with pictures no longer copies them (92 MB with 20 pictures: about 90 MB less) | Thesis and book authors | 10–20 | [layout-parity.md](layout-parity.md) |
 | 21 | **Mail merge data sources**: no Outlook/Contacts, Access or database sources, no email merge; CSV, typed lists (#247) and Excel/OpenDocument spreadsheets (#334, legacy .xls #778) only | | Select Recipients reads .csv/.tsv/.txt in any common encoding (UTF-8/16, GBK, Big5, Shift_JIS…), .xlsx/.xlsm, .xls and .ods; Edit Recipient List filters and sorts (#778) | Office administrators | 4–8 | [target-app-parity.md](target-app-parity.md) |
 | 22 | **Ecosystem**: no VBA execution (macros preserved only), no Office add-ins, no cloud storage, no real-time co-authoring | | `.docm` keeps macros (#172) | Enterprise users with macro templates; teams | 40–80 + owner decisions | [target-app-parity.md](target-app-parity.md) |
 | 23 | **PDF Reflow** (open a PDF as an editable document, Word for Windows) | | | Users who edit PDFs they receive | 30–50 | [file-format-parity.md](file-format-parity.md) |
@@ -121,6 +121,7 @@ The Language dialog reads effective character formatting, including document def
 | 2026-10-14 | trivial | Row table border exceptions fixed; other row table exceptions remain unsupported |
 | 2026-10-12 | minor | Czech proofing review: short-word splits rank after deletion, NOSUGGEST excludes suggestions while accepting typed forms, affix comments apply only to headers, ratios keep spaces and repeated-word messages use Czech quotes; synthetic engine end-to-end and latency checks |
 | 2026-10-12 | minor | Czech spelling from the installed Hunspell dictionary, runtime-table suggestions and generic checks; Czech interface writes cs-CZ documents; no Czech hyphenation patterns bundled; estimates unchanged |
+| 2026-10-16 | trivial | Gap #20: opening a DOCX no longer copies its pictures |
 | 2026-10-12 | minor | Gap #24 default shortcuts expanded (140 commands on macOS / 158 on Linux/Windows); stale gaps #9 (DOCX agile encryption already works, only encrypted .doc missing) and #15 (DOCX bibliography sources already saved, only desktop integrations missing) updated |
 | 2026-10-13 | trivial | Previously untracked long Web/Draft page blur resolved with visible high-resolution raster tiling; no gap remains for this issue |
 | 2026-10-14 | trivial | #1: VML freeforms (rule lines in headers) read and kept on save |