Merge origin/main

Tests only: both sides added tests at the same place; both kept.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SqmHYx8FiWM5PcsoiErgp9

ROADMAP.md

@@ -1,5 +1,9 @@
 # PdfCraft roadmap
 
+> **Last reviewed:** 2026-11-03 | **Last updated:** 2026-11-03 | **Change:** patch (M5 #5850 review: indirect typewriter intent and composer edits preserve outer whitespace; movement regression commits by outside click; estimates unchanged)
+
+> **Last reviewed:** 2026-11-03 | **Last updated:** 2026-11-03 | **Change:** patch (M5 #5850: Enter adds Fill & Sign line breaks; outside clicks commit, Escape cancels, and saved multiline appearances retain their text and height; estimates unchanged)
+
 > **Last reviewed:** 2026-11-01 | **Last updated:** 2026-11-01 | **Change:** patch (M7, PR #2155 further review: inherited drawing/text state, affine image gestures, paste/selection ownership, no-op scripts and menu permissions; estimates unchanged)
 
 > **Last reviewed:** 2026-11-01 | **Last updated:** 2026-11-01 | **Change:** patch (M7, PR #2155 latest comments: explicit image-paste events, mixed-size field refresh, and documented mixed-script normalization with Undo; estimates unchanged)
@@ -219,6 +223,10 @@ Ranked; detail and the full milestone table in [docs/roadmap.md](docs/roadmap.md
 
 ## Progress log
 
+- **2026-11-03 (M5, #5850 review):** Resolve indirect IT when recognizing typewriter contents and Fill & Sign marks. Comment listing and property lookup retain their existing direct-Subtype behavior. A real double-click Edit composer preserves the typewriter’s text as written; ordinary comment edits keep trimming. Regressions cover direct Subtype with indirect intent, save/reopen, raw posted contents and same-index comments on different pages. The existing live-movement test now commits its setup with an outside click rather than Enter. Estimates unchanged.
+
+- **2026-11-03 (M5, #5850):** Fill & Sign Enter inserts a line break in a growing editor; clicking outside commits and Escape cancels. Nonempty text keeps leading, internal and trailing line breaks. The headless fill_sign_add path sizes its typewriter box for the same wrapped lines so saved appearances show every row. Production-order keyboard and engine save/reopen/render regressions exercise the UI and headless callers. Estimates unchanged.
+
 - **2026-11-03 (M7, #5835):** In image crop mode, a right-click on the image, its crop box or its handles keeps the crop and opens no menu, so the menu no longer switches as the pointer moves. A right-click elsewhere clears the selection and opens the page menu. UI tests cover all four page rotations, overlapping images, a 45° image, and a check that the Done/Cancel/Remove Crop pills at a page edge still keep the crop.
 
 - **2026-11-02 (M12, #1297):** `model_add` embeds a U3D or PRC file, recognised by its header, as a 3D annotation with a chosen activation and a poster (an image or PDF page fitted in the box, or a labelled frame); an older file's catalog `/Version` is raised to what the model needs. Views, a rendered poster and a UI command are later steps. Estimates unchanged.
@@ -751,6 +759,8 @@ Newest first: the date, what moved, and the new overall estimate.
 
 | Date | Change | Summary |
 |---|---|---|
+| 2026-11-03 | patch | M5 #5850 review: indirect typewriter intent and composer edits preserve outer whitespace; movement regression commits by outside click; estimates unchanged. |
+| 2026-11-03 | patch | M5 #5850: Enter adds Fill & Sign line breaks; outside clicks commit, Escape cancels, and saved multiline appearances retain their text and height; estimates unchanged. |
 | 2026-11-01 | patch | M7 #5657: header/footer, watermark and background operations commit all selected pages together; later-page errors preserve marks, content, shared/inherited resources and pending edits; estimates unchanged. |
 | 2026-10-28 | patch | M2, PR #4172: integrate main 5af10276 and retain both roadmap histories; viewer-only CJK and print/output font policies and estimates unchanged. |
 | 2026-10-27 | minor | M10: OCR stages SHA-256-verified ONNX models after the upstream RTEN URL brownout; discovery prefers a complete ONNX pair per directory, falls back to a complete RTEN pair, and never mixes formats or directories; language/web limits and estimates unchanged. |

crates/annot/src/info.rs

@@ -47,8 +47,12 @@ pub(crate) fn text_value(doc: &Document, d: &Dict, key: &[u8]) -> Option<String>
         Object::Name(n) => String::from_utf8_lossy(n).into_owned(),
         _ => return None,
     };
-    let s = s.trim_matches('\0').trim().to_string();
-    (!s.is_empty()).then_some(s)
+    let s = s.trim_matches('\0');
+    // Typewriter outer whitespace belongs to the text (keep in step with render's inspector).
+    let name_is = |key: &[u8], expected: &[u8]| d.get(key).is_some_and(|o| doc.resolve(o).as_name() == Some(expected));
+    let keep_breaks = key == b"Contents" && name_is(b"Subtype", b"FreeText") && name_is(b"IT", b"FreeTextTypeWriter");
+    let s = if keep_breaks { s } else { s.trim() };
+    (!s.trim().is_empty()).then(|| s.to_owned())
 }
 
 /// Whether an annotation subtype is a comment. Links, form widgets, pop-ups and the non-markup

crates/annot/src/lookup.rs

@@ -137,11 +137,11 @@ pub fn is_fill_sign(doc: &Document, d: &Dict) -> bool {
     if matches!(d.get(b"PCFillSign").map(|o| doc.resolve(o)).as_deref(), Some(Object::Bool(true))) {
         return true;
     }
-    let subtype = d.name(b"Subtype").unwrap_or_default();
-    if subtype == b"FreeText" && d.name(b"IT") == Some(b"FreeTextTypeWriter") {
+    let name_is = |key: &[u8], expected: &[u8]| d.get(key).is_some_and(|o| doc.resolve(o).as_name() == Some(expected));
+    if name_is(b"Subtype", b"FreeText") && name_is(b"IT", b"FreeTextTypeWriter") {
         return true;
     }
-    if subtype == b"Stamp" {
+    if name_is(b"Subtype", b"Stamp") {
         if matches!(d.name(b"Name"), Some(b"PCCheck" | b"PCCross" | b"PCDot" | b"PCLine" | b"PCTypedSignature")) {
             return true;
         }
@@ -151,7 +151,7 @@ pub fn is_fill_sign(doc: &Document, d: &Dict) -> bool {
             return true;
         }
     }
-    subtype == b"Ink" && text_value(doc, d, b"Subj").as_deref() == Some("Signature")
+    name_is(b"Subtype", b"Ink") && text_value(doc, d, b"Subj").as_deref() == Some("Signature")
 }
 
 pub(crate) fn annotation_flags(doc: &Document, d: &Dict) -> i64 {

crates/annot/src/tests.rs

@@ -2385,3 +2385,27 @@ fn a_3d_poster_fitted_to_its_rectangle_with_rounding_is_taken() {
     };
     assert_eq!(add_3d_model(&mut doc, &new, &meta("m")), Ok(1));
 }
+
+#[test]
+fn indirect_typewriter_names_keep_outer_whitespace_in_summaries() {
+    let mut doc = fixture();
+    let intent = Object::Ref(doc.add(Object::name("FreeTextTypeWriter")));
+    let page = page_refs(&doc).unwrap()[0];
+    let mut entries = annots(&doc, page);
+    let Object::Dict(annotation) = &mut entries[0] else { panic!("fixture annotation is inline") };
+    annotation.set(b"Subtype".to_vec(), Object::name("FreeText"));
+    annotation.set(b"IT".to_vec(), intent);
+    let contents = " \nAda\n\nLovelace\n\n ";
+    annotation.set(b"Contents".to_vec(), PdfString::text(contents));
+    set_annots(&mut doc, page, entries).unwrap();
+    let check = |doc: &Document| {
+        let list = summaries(doc);
+        let summary = list.iter().find(|a| a.page == 0 && a.index == 0).expect("direct Subtype is listed");
+        assert_eq!(summary.contents.as_deref(), Some(contents), "indirect typewriter intent preserves outer whitespace");
+        assert_eq!(summary.subtype, "FreeText");
+        assert_eq!(summary.intent.as_deref(), Some("FreeTextTypeWriter"));
+        assert!(summary.fill_sign, "indirect intent still identifies a Fill & Sign mark");
+    };
+    check(&doc);
+    check(&reopen(&doc));
+}

crates/audit/src/cover.rs

@@ -15,7 +15,7 @@
 //! the joined stream)`; rectangles inside a form XObject carry the `Do` operator's position.
 //! Positions compare lexicographically in paint order.
 
-use pdfcraft_content::{Matrix, Op, Pieces, parse};
+use pdfcraft_content::{Matrix, Op, Pieces, ops};
 use pdfcraft_cos::{DecodeBudget, Dict, Document, Object};
 
 use crate::AuditError;
@@ -123,12 +123,13 @@ impl<'a> Walker<'a> {
         Ok(())
     }
 
-    /// Walk one operator slice with the given graphics state. `at_do` is the `Do` position to
+    /// Walk one stream's operators, read one at a time (a stream of millions of short operators
+    /// is never held whole), with the given graphics state. `at_do` is the `Do` position to
     /// stamp on rectangles found inside a form XObject, or `None` at page level (where each
     /// fill operator's own position is used).
     fn walk_ops(
         &mut self,
-        ops: &[Op],
+        ops: impl Iterator<Item = Op>,
         gs: &mut Gs,
         stack: &mut Vec<Gs>,
         resources: &Dict,
@@ -136,8 +137,9 @@ impl<'a> Walker<'a> {
         at_do: Option<(usize, usize)>,
     ) -> Result<(), AuditError> {
         let mut path = Path::Empty;
-        for (i, op) in ops.iter().enumerate() {
+        for (i, op) in ops.enumerate() {
             self.bump()?;
+            let op = &op;
             // Paint position of this operator: its own `(piece, joined index)` at page level,
             // the enclosing `Do`'s position inside a form XObject.
             let pos = match at_do {
@@ -257,8 +259,9 @@ impl<'a> Walker<'a> {
     /// changes inside never leak out; every rectangle found is stamped with the `Do`
     /// position `pos`.
     fn walk_xobject(&mut self, name: &[u8], gs: &Gs, resources: &Dict, depth: usize, pos: (usize, usize)) -> Result<(), AuditError> {
-        let xobjects = resources.get(b"XObject").map(|o| self.doc.resolve(o)).and_then(|o| o.as_dict().cloned()).unwrap_or_default();
-        let Some(entry) = xobjects.get(name) else { return Ok(()) };
+        // Borrowed, not copied: a `Do` must not cost the size of the whole XObject table.
+        let xobjects = resources.get(b"XObject").map(|o| self.doc.resolve(o));
+        let Some(entry) = xobjects.as_deref().and_then(Object::as_dict).and_then(|d| d.get(name)) else { return Ok(()) };
         let resolved = self.doc.resolve(entry);
         let Object::Stream(stream) = &*resolved else { return Ok(()) };
         let dict = &stream.dict;
@@ -274,18 +277,18 @@ impl<'a> Walker<'a> {
                 (v.len() == 6 && v.iter().all(|x| x.is_finite())).then(|| Matrix([v[0], v[1], v[2], v[3], v[4], v[5]]))
             })
             .unwrap_or(Matrix::IDENTITY);
-        let sub_resources = dict.get(b"Resources").map(|o| self.doc.resolve(o)).and_then(|o| o.as_dict().cloned()).unwrap_or_default();
+        let (sub_resources, empty) = (dict.get(b"Resources").map(|o| self.doc.resolve(o)), Dict::new());
+        let sub_resources = sub_resources.as_deref().and_then(Object::as_dict).unwrap_or(&empty);
         if !self.budget.enter_form() {
             return Err(AuditError::TooManyForms);
         }
         // A form stream that cannot be decoded fails the audit: it might hide a cover.
         let data = self.budget.decode_strict(stream)?;
-        let ops = parse(&data).ops;
         // The XObject matrix maps form space into the user space active at `Do`.
         let mut sub_gs = gs.clone();
         sub_gs.ctm = matrix.then(&gs.ctm);
         let mut sub_stack = Vec::new();
-        self.walk_ops(&ops, &mut sub_gs, &mut sub_stack, &sub_resources, depth + 1, Some(pos))
+        self.walk_ops(ops(&data), &mut sub_gs, &mut sub_stack, sub_resources, depth + 1, Some(pos))
     }
 }
 
@@ -298,13 +301,10 @@ fn user_point(gs: &Gs, op: &Op) -> Option<[f64; 2]> {
 
 /// The non-stroking alpha (`/ca`) of a named ExtGState, defaulting to opaque.
 fn ext_alpha(doc: &Document, resources: &Dict, name: &[u8]) -> f64 {
-    let alpha = resources
-        .get(b"ExtGState")
-        .map(|o| doc.resolve(o))
-        .and_then(|o| o.as_dict().cloned())
-        .and_then(|d| d.get(name).map(|o| doc.resolve(o)))
-        .and_then(|o| o.as_dict().cloned())
-        .and_then(|d| d.get(b"ca").map(|o| doc.resolve(o)).and_then(|o| o.as_f64()));
+    // Borrowed, not copied: a `gs` must not cost the size of the whole ExtGState table.
+    let states = resources.get(b"ExtGState").map(|o| doc.resolve(o));
+    let state = states.as_deref().and_then(Object::as_dict).and_then(|d| d.get(name)).map(|o| doc.resolve(o));
+    let alpha = state.as_deref().and_then(Object::as_dict).and_then(|d| d.get(b"ca").map(|o| doc.resolve(o)).and_then(|o| o.as_f64()));
     alpha.filter(|a| a.is_finite()).map(|a| a.clamp(0.0, 1.0)).unwrap_or(1.0)
 }
 
@@ -330,10 +330,11 @@ pub fn filled_rects(doc: &Document, page: &Dict) -> Result<Vec<FilledRect>, Audi
         }
     }
     let pieces = Pieces::join(&streams);
-    let parsed = pieces.parse();
     let mut w = Walker { doc, pieces: &pieces, out: Vec::new(), ops_seen: 0, budget };
     let mut gs = Gs::default();
     let mut stack = Vec::new();
-    w.walk_ops(&parsed.ops, &mut gs, &mut stack, &resources, 0, None)?;
+    // Read one operator at a time: the joined stream can hold millions of them within the
+    // decode budget, and the walk only ever needs the current one.
+    w.walk_ops(ops(pieces.data()), &mut gs, &mut stack, &resources, 0, None)?;
     Ok(w.out)
 }

crates/audit/src/lib.rs

@@ -99,6 +99,9 @@ const COVER_FRACTION: f64 = 0.4;
 const WORD_FRACTION: f64 = 0.5;
 /// WCAG contrast ratio below which the text is unreadable against the cover.
 const COVER_CONTRAST: f64 = 3.0;
+/// Bytes of a hidden line's text read (copied or trimmed) that cost one comparison: a byte scan
+/// is far cheaper than a box comparison, but a long padded line under many covers adds up.
+const TEXT_BYTES_PER_COMPARISON: usize = 16;
 
 /// Audit every page for faux redactions and unapplied redaction marks. Fails closed: any
 /// content that cannot be decoded, or that needs more work or report than the audit's bounds, is
@@ -157,15 +160,24 @@ impl Budget {
         cover: &[f64; 4],
         hides: impl Fn(&TextLine) -> bool,
     ) -> Result<Option<String>, AuditError> {
-        self.comparisons = self.comparisons.saturating_add(self.page_cost);
-        if self.comparisons > self.limits.comparisons {
-            return Err(AuditError::TooManyComparisons);
-        }
+        self.spend(self.page_cost)?;
         if !copying {
-            return Ok(lines.iter().any(|line| hides(line) && !line.text.trim().is_empty()).then(String::new));
+            let mut hidden = false;
+            for line in lines.iter().filter(|l| hides(l)) {
+                // Trimming walks the line's text (padding and all): it costs its bytes.
+                self.spend(line.text.len() / TEXT_BYTES_PER_COMPARISON)?;
+                if !line.text.trim().is_empty() {
+                    hidden = true;
+                    break;
+                }
+            }
+            return Ok(hidden.then(String::new));
         }
         let mut covered = String::new();
         for line in lines.iter().filter(|l| hides(l)) {
+            // Copying and trimming the line walks its text once per cover: a long padded
+            // line under many covers would otherwise be read far more than its runs suggest.
+            self.spend(line.text.len() / TEXT_BYTES_PER_COMPARISON)?;
             if !covered.is_empty() {
                 covered.push(' ');
             }
@@ -174,11 +186,21 @@ impl Budget {
         Ok((!covered.trim().is_empty()).then_some(covered))
     }
 
+    /// `n` units of work done against the comparison bound.
+    fn spend(&mut self, n: usize) -> Result<(), AuditError> {
+        self.comparisons = self.comparisons.saturating_add(n);
+        if self.comparisons > self.limits.comparisons {
+            return Err(AuditError::TooManyComparisons);
+        }
+        Ok(())
+    }
+
     /// The text a finding reports for `covered`: none when copying is withheld. Charges what it
-    /// reports against the output bound.
+    /// reports, and the finding's own record, against the output bound: a finding with no text
+    /// (copying withheld, or a shared annotation list on many pages) still takes memory.
     fn report_text(&mut self, covered: String, copying: bool) -> Result<String, AuditError> {
         let reported = if copying { covered } else { String::new() };
-        self.covered_text = self.covered_text.saturating_add(reported.len());
+        self.covered_text = self.covered_text.saturating_add(reported.len()).saturating_add(std::mem::size_of::<AuditFinding>());
         if self.covered_text > self.limits.covered_text {
             return Err(AuditError::TooMuchOutput);
         }
@@ -321,17 +343,14 @@ enum AnnotationCover {
 fn annots(doc: &Document, page: &Dict) -> Result<Vec<AnnotationCover>, AuditError> {
     let Some(a) = page.get(b"Annots") else { return Ok(Vec::new()) };
     let resolved = doc.resolve(a);
-    let items: Vec<pdfcraft_cos::Object> = match &*resolved {
-        pdfcraft_cos::Object::Array(items) => items.clone(),
-        // A missing /Annots means no marks; a malformed one fails the audit.
-        _ => return Err(AuditError::BadAnnots),
-    };
+    // A missing /Annots means no marks; a malformed one fails the audit.
+    let Some(items) = resolved.as_array() else { return Err(AuditError::BadAnnots) };
     let mut out = Vec::new();
-    for a in &items {
-        let dict = match &*doc.resolve(a) {
-            pdfcraft_cos::Object::Dict(d) => d.clone(),
-            _ => continue,
-        };
+    for a in items {
+        // Borrowed from the resolved object: one large dictionary listed many times is not
+        // copied once per listing.
+        let resolved = doc.resolve(a);
+        let Some(dict) = resolved.as_dict() else { continue };
         let v: Vec<f64> = match dict.get(b"Rect") {
             None => continue,
             Some(r) => doc.resolve(r).as_array().map(|arr| arr.iter().filter_map(|o| doc.resolve(o).as_f64()).collect()).unwrap_or_default(),
@@ -340,7 +359,7 @@ fn annots(doc: &Document, page: &Dict) -> Result<Vec<AnnotationCover>, AuditErro
             continue;
         }
         let rect = [v[0].min(v[2]), v[1].min(v[3]), v[0].max(v[2]), v[1].max(v[3])];
-        match subtype(doc, &dict)?.as_deref() {
+        match subtype(doc, dict)?.as_deref() {
             Some(b"Redact") => out.push(AnnotationCover::Redact(rect)),
             Some(b"Square") => {
                 let flags = dict.get(b"F").and_then(|f| doc.resolve(f).as_int()).unwrap_or(0);
@@ -349,7 +368,7 @@ fn annots(doc: &Document, page: &Dict) -> Result<Vec<AnnotationCover>, AuditErro
                 }
                 let alpha = dict.get(b"CA").map(|o| doc.resolve(o)).and_then(|o| o.as_f64()).unwrap_or(1.0);
                 if alpha >= 1.0
-                    && let Some(rgb) = annotation_color(doc, &dict, b"IC")
+                    && let Some(rgb) = annotation_color(doc, dict, b"IC")
                 {
                     out.push(AnnotationCover::Square { rect, rgb });
                 }

crates/audit/src/tests.rs

@@ -720,9 +720,11 @@ fn runs_without_words_count_against_the_comparison_bound() {
     let doc = open(one_page(&content, ""));
     let lines = pdfcraft_edit::glyph_text_lines(&doc, &pdfcraft_model::pages(&doc)[0]).expect("text");
     assert_eq!(lines.iter().map(|l| (l.ink.len(), l.runs().count())).collect::<Vec<_>>(), [(1, 51)]);
-    let limits = crate::Limits { comparisons: 50, covered_text: usize::MAX };
+    // Plus the hidden line's text, read once to copy it.
+    let text = lines[0].text.len() / crate::TEXT_BYTES_PER_COMPARISON;
+    let limits = crate::Limits { comparisons: 50 + text, covered_text: usize::MAX };
     assert!(matches!(crate::audit(&doc, limits), Err(crate::AuditError::TooManyComparisons)));
-    let limits = crate::Limits { comparisons: 51, covered_text: usize::MAX };
+    let limits = crate::Limits { comparisons: 51 + text, covered_text: usize::MAX };
     assert_eq!(crate::audit(&doc, limits).expect("within the budget").len(), 1);
 }
 
@@ -746,14 +748,18 @@ fn copy_restricted_documents_withhold_covered_text() {
     let findings = audit_redactions(&user).expect("audit must run");
     assert_eq!(findings.len(), 1, "the cover is still reported: {findings:?}");
     assert_eq!(findings[0].covered_text, "");
-    // The output bound counts the text reported, so withheld text costs nothing.
-    let nothing = crate::Limits { comparisons: usize::MAX, covered_text: 0 };
+    // The output bound counts the text reported and the finding's record, so withheld text
+    // costs only the record.
+    let record = std::mem::size_of::<AuditFinding>();
+    let nothing = crate::Limits { comparisons: usize::MAX, covered_text: record };
     assert_eq!(crate::audit(&user, nothing).expect("withheld text is not output").len(), 1);
+    let none = crate::Limits { covered_text: record - 1, ..nothing };
+    assert!(matches!(crate::audit(&user, none), Err(crate::AuditError::TooMuchOutput)));
     let owner = Document::open_with_password(bytes, Some("owner")).expect("open");
     let findings = audit_redactions(&owner).expect("audit must run");
     assert_eq!(findings.len(), 1, "{findings:?}");
     assert_eq!(findings[0].covered_text, "SECRET DATA 123");
-    let exact = crate::Limits { comparisons: usize::MAX, covered_text: "SECRET DATA 123".len() };
+    let exact = crate::Limits { comparisons: usize::MAX, covered_text: "SECRET DATA 123".len() + record };
     assert!(crate::audit(&owner, exact).is_ok());
     let short = crate::Limits { covered_text: exact.covered_text - 1, ..exact };
     assert!(matches!(crate::audit(&owner, short), Err(crate::AuditError::TooMuchOutput)));
@@ -820,3 +826,52 @@ fn form_entries_past_the_page_allowance_fail_the_audit() {
     let page: Dict = [(b"Contents".to_vec(), Object::Ref(draws)), (b"Resources".to_vec(), Object::Dict(resources))].into_iter().collect();
     assert!(matches!(crate::cover::filled_rects(&doc, &page), Err(crate::AuditError::TooManyForms)));
 }
+
+/// Copying a covered line's text walks all of it, padding included, so a long padded line under
+/// many covers costs its bytes once per cover, not one run per cover (CodeRabbit).
+#[test]
+fn covered_text_bytes_count_against_the_comparison_bound() {
+    // One run: 20 KiB of spaces around a word, which trims to the word alone.
+    let mut content = b"BT /F1 24 Tf 100 700 Td (".to_vec();
+    content.extend_from_slice(&[b' '; 10_000]);
+    content.extend_from_slice(b"SECRET");
+    content.extend_from_slice(&[b' '; 10_000]);
+    content.extend_from_slice(b") Tj ET\n0 g\n");
+    for _ in 0..10 {
+        content.extend_from_slice(b"-1 688 1000000 32 re f\n");
+    }
+    let doc = open(one_page(&content, ""));
+    let lines = pdfcraft_edit::glyph_text_lines(&doc, &pdfcraft_model::pages(&doc)[0]).expect("text");
+    assert_eq!(lines.len(), 1);
+    let bytes = lines[0].text.len();
+    assert!(bytes > 20_000, "{bytes} bytes of text");
+    // Each cover: one comparison for the line, plus its bytes when it is hidden.
+    let exact = crate::Limits { comparisons: 10 * (1 + bytes / crate::TEXT_BYTES_PER_COMPARISON), covered_text: usize::MAX };
+    let findings = crate::audit(&doc, exact).expect("within the budget");
+    assert_eq!(findings.len(), 10, "{findings:?}");
+    assert!(findings.iter().all(|f| f.covered_text == "SECRET"), "{findings:?}");
+    let short = crate::Limits { comparisons: exact.comparisons - 1, ..exact };
+    assert!(matches!(crate::audit(&doc, short), Err(crate::AuditError::TooManyComparisons)));
+}
+
+/// One annotation list shared by many pages makes a finding on each: every finding's record
+/// counts against the output bound, text or no text (CodeRabbit).
+#[test]
+fn every_finding_record_counts_against_the_output_bound() {
+    let content = b"BT /F1 24 Tf 100 700 Td (SECRET) Tj ET\n";
+    let mut b = Builder::new();
+    b.add("<< /Type /Catalog /Pages 2 0 R >>");
+    b.add("<< /Type /Pages /Kids [3 0 R 7 0 R 8 0 R] /Count 3 >>");
+    b.add("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Contents 4 0 R /Resources << /Font << /F1 5 0 R >> >> /Annots 6 0 R >>");
+    b.add_stream(content);
+    b.add(FONT);
+    b.add("[<< /Type /Annot /Subtype /Redact /Rect [95 685 200 725] >>]");
+    b.add("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Contents 4 0 R /Resources << /Font << /F1 5 0 R >> >> /Annots 6 0 R >>");
+    b.add("<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Contents 4 0 R /Resources << /Font << /F1 5 0 R >> >> /Annots 6 0 R >>");
+    let doc = open(b.build());
+    let one = "SECRET".len() + std::mem::size_of::<AuditFinding>();
+    let exact = crate::Limits { comparisons: usize::MAX, covered_text: 3 * one };
+    assert_eq!(crate::audit(&doc, exact).expect("within the budget").len(), 3);
+    let short = crate::Limits { covered_text: 3 * one - 1, ..exact };
+    assert!(matches!(crate::audit(&doc, short), Err(crate::AuditError::TooMuchOutput)));
+}

crates/automation/README.md

@@ -5,6 +5,8 @@ Agent control for PdfCraft: a headless tool table over the engine, and an opt-in
 - **Layer:** L7 (architecture §13), but headless. It depends on `engine`, `render` and `organize`, never on a UI toolkit.
 - **Status:** over a hundred tools; `pdfcraft-cli tools` lists the current set with their schemas (the count isn't repeated here, so it can't go stale). They cover inspection, rendering, text, page edits and organizing, page labels, bookmarks, comments and their review, form filling and authoring (fields, properties, scripts, data exchange, detection), redaction and sanitizing, password protection and signatures, calibrated 2D measurements and CSV export, adding text, images, links and stamps, optimizing, OCR, printing, export, metadata, undo/redo, save, combine, extract and split. Comment tools take geometry in the same top-left-origin points as everything else, and `comment_add` can mark text by searching for it (`find`). The MCP server also serves the open documents as resources, page images included. The running app's UI is driven separately, through the `ui.*` control channel in `pdfcraft-ui-egui`.
 
+`fill_sign_add` text accepts line breaks, including blank lines. Its typewriter box grows to hold all wrapped rows, and saving/reopening preserves the contents and appearance.
+
 ## API
 
 ```rust

crates/automation/src/a11y/documents.rs

@@ -6,6 +6,13 @@ impl Automation {
         let ext = path.extension().map(|e| e.to_string_lossy().into_owned()).unwrap_or_default();
         let format = pdfcraft_engine::compare::OfficeFormat::from_extension(&ext)
             .ok_or_else(|| ToolError::InvalidArgs(format!("unsupported extension {ext:?} (docx, pptx, html, rtf or md)")))?;
+        let html_mode = a.opt_str("html_mode")?.unwrap_or("rich");
+        let format = match (format, html_mode) {
+            (pdfcraft_engine::compare::OfficeFormat::Html, "rich") => pdfcraft_engine::compare::OfficeFormat::Html,
+            (pdfcraft_engine::compare::OfficeFormat::Html, "simple") => pdfcraft_engine::compare::OfficeFormat::HtmlSimple,
+            (_, "rich") => format,
+            (_, _) => return Err(ToolError::InvalidArgs("html_mode is only available for HTML exports (rich or simple)".into())),
+        };
         let export = self.doc_to_copy(a)?.export_office_checked(format);
         write_atomic(&path, &export.bytes)?;
         // Text left out to keep a hostile page bounded is said, not silent (#3120).

crates/automation/src/bates.rs

@@ -0,0 +1,214 @@
+use std::io::Write;
+
+use pdfcraft_engine::{Edit, HeaderFooter};
+use serde_json::{Value, json};
+
+use super::error::{Result, ToolError, failed};
+use super::json::Args;
+use super::paths::read_input;
+use super::{Arc, Automation, Path, PathBuf};
+
+struct FilePlan {
+    input: String,
+    source: PathBuf,
+    password: Option<String>,
+    output: PathBuf,
+    pages: usize,
+    first: u64,
+    last: u64,
+}
+
+impl Automation {
+    pub(crate) fn doc_bates_multi(&mut self, args: &Args) -> Result<Value> {
+        let files = args.0.get("files").and_then(Value::as_array).ok_or_else(|| ToolError::InvalidArgs("files must be an array".into()))?;
+        if files.is_empty() || files.len() > 1000 {
+            return Err(ToolError::InvalidArgs("files must contain 1 to 1000 entries".into()));
+        }
+        let start = match args.0.get("start_number") {
+            None | Some(Value::Null) => 1,
+            Some(value) => value
+                .as_u64()
+                .filter(|n| *n >= 1)
+                .ok_or_else(|| ToolError::InvalidArgs("start_number must be a whole number of at least 1".into()))?,
+        };
+        let digits = args.opt_int("digits")?.unwrap_or(6);
+        let digits =
+            u8::try_from(digits).ok().filter(|d| (1..=12).contains(d)).ok_or_else(|| ToolError::InvalidArgs("digits must be from 1 to 12".into()))?;
+        let prefix = args.opt_str("prefix")?.unwrap_or("");
+        let suffix = args.opt_str("suffix")?.unwrap_or("");
+        // `#` separates the token's fields and `>>` ends it: either would cut the token short.
+        if prefix.len() > 128 || suffix.len() > 128 || [prefix, suffix].iter().any(|s| s.contains('#') || s.contains(">>")) {
+            return Err(ToolError::InvalidArgs("prefix and suffix must be at most 128 bytes and cannot contain # or >>".into()));
+        }
+        let dry_run = args.opt_bool("dry_run")?.unwrap_or(false);
+
+        // Resolve and reserve the complete path list before opening or writing anything.
+        let mut sources = Vec::with_capacity(files.len());
+        let mut targets = Vec::with_capacity(files.len());
+        let mut passwords = Vec::with_capacity(files.len());
+        for entry in files {
+            let input = entry
+                .get("input")
+                .and_then(Value::as_str)
+                .filter(|s| !s.is_empty())
+                .ok_or_else(|| ToolError::InvalidArgs("each file needs a non-empty input path".into()))?;
+            let output = entry
+                .get("output")
+                .and_then(Value::as_str)
+                .filter(|s| !s.is_empty())
+                .ok_or_else(|| ToolError::InvalidArgs("each file needs a non-empty output path".into()))?;
+            let source = self.resolve(input, false)?;
+            let target = self.resolve(output, true)?;
+            if target.exists() {
+                return Err(failed(format!("output already exists; refusing to replace it: {output}")));
+            }
+            let parent = target.parent().ok_or_else(|| failed(format!("output has no parent directory: {output}")))?;
+            if !parent.is_dir() {
+                return Err(failed(format!("output directory does not exist: {}", parent.display())));
+            }
+            sources.push((input.to_string(), source));
+            targets.push((output.to_string(), target));
+            passwords.push(entry.get("password").and_then(Value::as_str).map(str::to_owned));
+        }
+        let input_paths: std::collections::HashSet<_> = sources.iter().map(|(_, path)| path.clone()).collect();
+        let mut output_paths = std::collections::HashSet::new();
+        for (name, path) in &targets {
+            if input_paths.contains(path) {
+                return Err(failed(format!("output {name} names an input file; choose a separate output path")));
+            }
+            if !output_paths.insert(path.clone()) {
+                return Err(failed(format!("two files use the same output path: {name}")));
+            }
+        }
+
+        let mut plans = Vec::with_capacity(files.len());
+        let mut next = start;
+        let mut input_bytes = 0u64;
+        for (index, ((input_name, input_path), (_, output_path))) in sources.iter().zip(&targets).enumerate() {
+            let size = std::fs::metadata(input_path).map_err(failed)?.len();
+            input_bytes = input_bytes
+                .checked_add(size)
+                .filter(|total| *total <= 512 * 1024 * 1024)
+                .ok_or_else(|| failed("combined Bates input exceeds the 512 MiB safety limit"))?;
+            let bytes = read_input(input_path)?;
+            input_bytes = input_bytes
+                .saturating_sub(size)
+                .checked_add(u64::try_from(bytes.len()).unwrap_or(u64::MAX))
+                .filter(|total| *total <= 512 * 1024 * 1024)
+                .ok_or_else(|| failed("combined Bates input exceeds the 512 MiB safety limit"))?;
+            let display_name = input_path.file_name().map(|n| n.to_string_lossy().into_owned()).unwrap_or_default();
+            let password = passwords.get(index).and_then(Option::as_deref);
+            let id = self.session.open(display_name, Some(input_path.to_string_lossy().into_owned()), Arc::new(bytes), password).map_err(failed)?;
+            let pages =
+                self.session.get(id).map(|doc| doc.info.pages.len()).ok_or_else(|| failed("opened document disappeared during Bates preflight"))?;
+            self.session.close(id);
+            if pages == 0 {
+                return Err(failed(format!("{input_name} has no pages to number")));
+            }
+            let count = u64::try_from(pages).map_err(|_| failed(format!("page count is too large for {input_name}")))?;
+            let last = next.checked_add(count.saturating_sub(1)).ok_or_else(|| failed(format!("Bates number range overflows for {input_name}")))?;
+            plans.push(FilePlan {
+                input: input_name.clone(),
+                source: input_path.clone(),
+                password: passwords.get(index).cloned().flatten(),
+                output: output_path.clone(),
+                pages,
+                first: next,
+                last,
+            });
+            if index + 1 < sources.len() {
+                next = last.checked_add(1).ok_or_else(|| failed(format!("Bates sequence overflows after {input_name}")))?;
+            }
+        }
+
+        let intervals: Vec<Value> = plans
+            .iter()
+            .zip(&targets)
+            .map(|(p, (name, _))| json!({"input":p.input,"output":name,"pages":p.pages,"first":p.first,"last":p.last}))
+            .collect();
+        if dry_run {
+            return Ok(json!({"dry_run":true,"files":intervals}));
+        }
+
+        let mut staged = Vec::with_capacity(plans.len());
+        let mut results = Vec::with_capacity(plans.len());
+        let mut output_bytes = 0u64;
+        for plan in &plans {
+            let source_bytes = match read_input(&plan.source) {
+                Ok(bytes) => bytes,
+                Err(error) => {
+                    results.push(json!({"input":plan.input,"output":plan.output,"status":"error","pages":plan.pages,"first":plan.first,"last":plan.last,"error":error.to_string()}));
+                    continue;
+                }
+            };
+            let name = plan.source.file_name().map(|n| n.to_string_lossy().into_owned()).unwrap_or_default();
+            let id = match self.session.open(name, Some(plan.source.to_string_lossy().into_owned()), Arc::new(source_bytes), plan.password.as_deref())
+            {
+                Ok(id) => id,
+                Err(error) => {
+                    results.push(json!({"input":plan.input,"output":plan.output,"status":"error","pages":plan.pages,"first":plan.first,"last":plan.last,"error":error.to_string()}));
+                    continue;
+                }
+            };
+            if self.session.get(id).is_none_or(|doc| doc.info.pages.len() != plan.pages) {
+                self.session.close(id);
+                results.push(json!({"input":plan.input,"output":plan.output,"status":"error","pages":plan.pages,"first":plan.first,"last":plan.last,"error":"input changed after preflight; page count differs"}));
+                continue;
+            }
+            let template = format!("<<Bates Number#{digits}#{}#{prefix}#{suffix}>>", plan.first);
+            let mut settings = HeaderFooter::default();
+            settings.text[4] = template;
+            let pages: Vec<usize> = (0..plan.pages).collect();
+            let applied = self.session.apply(id, Edit::AddHeaderFooter { pages, settings, replace: false });
+            let result = applied.and_then(|()| self.session.save_full_bytes(id));
+            match result {
+                Ok(bytes) => {
+                    let total = output_bytes.saturating_add(u64::try_from(bytes.len()).unwrap_or(u64::MAX));
+                    if total > 512 * 1024 * 1024 {
+                        results.push(json!({"input":plan.input,"output":plan.output,"status":"error","pages":plan.pages,"first":plan.first,"last":plan.last,"error":"combined Bates output exceeds the 512 MiB safety limit"}));
+                    } else {
+                        output_bytes = total;
+                        staged.push((plan.output.clone(), bytes));
+                    }
+                }
+                Err(error) => results.push(json!({"input":plan.input,"output":plan.output,"status":"error","pages":plan.pages,"first":plan.first,"last":plan.last,"error":error.to_string()})),
+            }
+            self.session.close(id);
+        }
+        for (path, bytes) in staged {
+            match publish_new(&path, &bytes) {
+                Ok(()) => {
+                    let plan = plans.iter().find(|p| p.output == path);
+                    if let Some(plan) = plan {
+                        results.push(json!({"input":plan.input,"output":path,"status":"ok","first":plan.first,"last":plan.last,"pages":plan.pages}));
+                    }
+                }
+                Err(error) => {
+                    let plan = plans.iter().find(|p| p.output == path);
+                    if let Some(plan) = plan {
+                        results.push(json!({"input":plan.input,"output":path,"status":"error","pages":plan.pages,"first":plan.first,"last":plan.last,"error":error.to_string()}));
+                    }
+                }
+            }
+        }
+        results.sort_by_key(|result| plans.iter().position(|plan| result["output"] == json!(plan.output)).unwrap_or(usize::MAX));
+        let completed = results.iter().filter(|r| r["status"] == "ok").count();
+        let failed_count = results.len().saturating_sub(completed);
+        Ok(json!({"dry_run":false,"completed":completed,"failed":failed_count,"files":results}))
+    }
+}
+
+/// Publish a PDF atomically while refusing a target created by another process after preflight.
+fn publish_new(path: &Path, bytes: &[u8]) -> Result<()> {
+    let parent = path.parent().ok_or_else(|| failed("output has no parent directory"))?;
+    let mut temp = tempfile::NamedTempFile::new_in(parent).map_err(failed)?;
+    temp.write_all(bytes).map_err(failed)?;
+    temp.as_file().sync_all().map_err(failed)?;
+    temp.persist_noclobber(path).map(|_| ()).map_err(|error| {
+        if error.error.kind() == std::io::ErrorKind::AlreadyExists {
+            failed(format!("output appeared after preflight; refusing to replace it: {}", path.display()))
+        } else {
+            failed(error.error)
+        }
+    })
+}

crates/automation/src/comments.rs

@@ -539,8 +539,10 @@ impl Automation {
             return self.apply(a, edit);
         }
         let text_at = |t: &str| {
-            let w = (pdfcraft_engine::annot_text::text_width(t, size) + 8.0).clamp(20.0, 600.0);
-            let h = size * 1.2 + 6.0;
+            let longest = t.lines().map(|line| pdfcraft_engine::annot_text::text_width(line, size)).fold(0.0, f64::max);
+            let w = (longest + 8.0).clamp(20.0, 600.0);
+            let rows = pdfcraft_engine::annot_text::wrap(t, size, w - 4.0).len().max(1);
+            let h = rows as f64 * size * 1.2 + 6.0;
             let rect = pdfcraft_engine::text_box_at(i64::from(info.rotation), at, w, h).unwrap_or([at[0], at[1] - h, at[0] + w, at[1]]);
             Shape::Typewriter { rect, font_size: size }
         };

crates/automation/src/dispatch.rs

@@ -53,6 +53,7 @@ impl Automation {
         tools::check_args(def, args)?;
         let a = Args(args);
         let out = match name {
+            "doc_bates_multi" => self.doc_bates_multi(&a)?,
             "doc_open" => self.doc_open(&a)?,
             "doc_batch" => self.doc_batch(&a)?,
             "doc_list" => json!({ "documents": self.session.docs().iter().map(summary).collect::<Vec<_>>() }),

crates/automation/src/lib.rs

@@ -16,6 +16,7 @@
 
 mod a11y;
 mod batch;
+mod bates;
 mod comments;
 mod content;
 mod conventions;

crates/automation/src/tools.rs

@@ -121,6 +121,14 @@ pub fn tools() -> Vec<ToolDef> {
     let save_out = json!({ "type": "string", "description": "File to write. Omit to open the result as a new unsaved document instead." });
     let open = json!({ "type": "boolean", "description": "Also open the result as a new document (default: only when out is omitted)." });
     vec![
+        t("doc_bates_multi", "Add Bates numbering across files", "Apply one consecutive Bates sequence to an ordered list of PDFs. Each file entry has input and output paths and an optional open or permissions password. The operation first validates every input/output and computes inclusive numeric intervals; dry_run returns that allocation without writing. Existing output files are never replaced. On application, files that cannot be edited or saved are reported individually and later files continue.")
+            .destructive()
+            .with(schema(json!({
+                "files": {"type":"array","minItems":1,"maxItems":1000,"description":"PDFs in Bates sequence order.","items":{"type":"object","properties":{"input":{"type":"string","minLength":1,"description":"Input PDF path."},"output":{"type":"string","minLength":1,"description":"New output PDF path; must not exist."},"password":{"type":"string","description":"Open or permissions password, if needed."}},"required":["input","output"],"additionalProperties":false}},
+                "start_number": {"type":"integer","minimum":1,"description":"First number in the sequence (default 1)."}, "digits": {"type":"integer","minimum":1,"maximum":12,"description":"Zero-padded number width (default 6)."},
+                "prefix": {"type":"string","maxLength":128,"description":"Text before the number: at most 128 UTF-8 bytes, without # or >>."}, "suffix": {"type":"string","maxLength":128,"description":"Text after the number: at most 128 UTF-8 bytes, without # or >>."},
+                "dry_run": {"type":"boolean","description":"Return each file's inclusive number interval without writing (default false)."}
+            }), &["files"])),
         t("doc_batch", "Process files and folders", "Run convert, ocr and/or optimize over files and folders, one file at a time. recursive defaults to false. Preserve paths relative to each selected folder inside out_dir; duplicate inputs run once. Skip unsupported files, links and the output subtree. Existing outputs receive numbered names and are never replaced. Continue after per-file errors unless stop_on_error is true. Return completed/failed/skipped counts and per-file results. Discovery is bounded to 10000 entries and 64 folder levels. Convert must be first; each step can appear once. OCR uses installed models. Optimize uses Reduce File Size settings, optionally overriding target ppi (threshold 1.5 times ppi; black-and-white images go to the larger of ppi and 300) and JPEG quality; a larger candidate keeps the original bytes. Sources and open documents are unchanged.")
             .with(schema(json!({
                 "paths": {"type":"array","items":{"type":"string"},"minItems":1,"maxItems":1000},
@@ -1208,9 +1216,9 @@ pub fn tools() -> Vec<ToolDef> {
                 json!({ "doc": doc(), "field": { "type": "string" }, "event": { "type": "string", "enum": ["keystroke", "format", "validate", "calculate", "mouse_up"] }, "script": { "type": "string" } }),
                 &["doc", "field", "event"],
             )),
-        t("doc_export_office", "Export to Word, PowerPoint, HTML, RTF or Markdown", "Export a PDF ▸ Word (.docx), PowerPoint (.pptx, one slide per page), HTML (.html, one file with images inline), RTF (.rtf) or Markdown (.md or .markdown: pipe tables, images left out), chosen by path's extension: text, headings, bold and italic. Word uses borderless inferred tables and positioned editable shapes on pages with artwork, rotated or large text; each source page retains its size. Font matching and vector graphics remain limited. Export reads forms 50 deep and up to 50 000 a page, counting forms the page draws itself. Pages that go past either limit, or that have pictures past what one export may hold or decode, are listed in truncated_pages, with a warning; what was left out is missing.")
+        t("doc_export_office", "Export to Word, PowerPoint, HTML, RTF or Markdown", "Export a PDF ▸ Word (.docx), PowerPoint (.pptx, one slide per page), HTML (.html, one file with images inline), RTF (.rtf) or Markdown (.md or .markdown: pipe tables, images left out), chosen by path's extension: text, headings, bold and italic. HTML uses rich styling by default; set html_mode to simple for CSS-free headings, paragraphs, emphasis and tables. Simple HTML keeps text and inline images but omits colour and layout styling. Word uses borderless inferred tables and positioned editable shapes on pages with artwork, rotated or large text; each source page retains its size. Font matching and vector graphics remain limited. Export reads forms 50 deep and up to 50 000 a page, counting forms the page draws itself. Pages that go past either limit, or that have pictures past what one export may hold or decode, are listed in truncated_pages, with a warning; what was left out is missing.")
             .cmd("export.docx")
-            .with(schema(json!({ "doc": doc(), "path": { "type": "string" } }), &["doc", "path"])),
+            .with(schema(json!({ "doc": doc(), "path": { "type": "string" }, "html_mode": { "type": "string", "enum": ["rich", "simple"], "default": "rich" } }), &["doc", "path"])),
         t("pdfa_verify", "Verify PDF/A", "Standards ▸ Verify PDF/A compliance: the PDF/A-2b or 3b rules the document breaks (ISO 19005 clause, message, page, whether Save as PDF/A can fix it), plus what it declares.")
             .ro()
             .cmd("standards.pdfa")

crates/automation/tests/it/automation.rs

@@ -2832,6 +2832,74 @@ fn headers_watermarks_and_backgrounds_through_tools() {
     assert!(matches!(refused, Err(ToolError::InvalidArgs(e)) if e.contains("Courier")));
 }
 
+#[test]
+fn bates_numbers_continue_across_files_and_report_preflight_and_permission_failures() {
+    let dir = workdir("bates-multi");
+    let files = json!([
+        {"input":"b.pdf","output":"b-numbered.pdf"},
+        {"input":"a.pdf","output":"a-numbered.pdf"}
+    ]);
+    let mut a = auto(&dir);
+    let allocation = ok(&mut a, "doc_bates_multi", json!({"files":files,"start_number":50,"dry_run":true}));
+    assert_eq!(allocation["files"][0]["first"], 50);
+    assert_eq!(allocation["files"][0]["last"], 51);
+    assert_eq!(allocation["files"][1]["first"], 52);
+    assert_eq!(allocation["files"][1]["last"], 54);
+    assert!(!dir.join("a-numbered.pdf").exists());
+
+    let applied = ok(&mut a, "doc_bates_multi", json!({"files":files,"start_number":17,"prefix":"CASE-"}));
+    assert_eq!(applied["completed"], 2);
+    let mut out = auto(&dir);
+    let first = ok(&mut out, "doc_open", json!({"path":"a-numbered.pdf"}))["doc"].as_u64().unwrap();
+    let second = ok(&mut out, "doc_open", json!({"path":"b-numbered.pdf"}))["doc"].as_u64().unwrap();
+    assert!(page_text(&mut out, second)[0].contains("CASE-000017"));
+    assert!(page_text(&mut out, first)[0].contains("CASE-000019"));
+
+    let overflow = err(&mut a, "doc_bates_multi", json!({"files":[{"input":"b.pdf","output":"overflow.pdf"}],"start_number":u64::MAX}));
+    assert!(overflow.to_string().contains("overflows"));
+    assert!(!dir.join("overflow.pdf").exists());
+    // A range may end at u64::MAX itself.
+    let at_max = ok(&mut a, "doc_bates_multi", json!({"files":[{"input":"b.pdf","output":"max.pdf"}],"start_number":u64::MAX - 1,"dry_run":true}));
+    assert_eq!(at_max["files"][0]["last"], u64::MAX);
+    for bad in
+        [json!({"start_number":-3}), json!({"start_number":2.5}), json!({"start_number":"7"}), json!({"prefix":"CASE>>"}), json!({"suffix":"A>>B"})]
+    {
+        let mut args = json!({"files":[{"input":"b.pdf","output":"bad.pdf"}],"dry_run":true});
+        args.as_object_mut().unwrap().extend(bad.as_object().unwrap().clone());
+        assert!(a.call("doc_bates_multi", &args).is_err(), "{bad}");
+    }
+
+    use pdfcraft_engine::{Changes, Edit, Protection, Session};
+    let mut session = Session::new();
+    let id = session.open("locked.pdf", None, std::sync::Arc::new(fixture(1)), None).unwrap();
+    session
+        .apply(
+            id,
+            Edit::Protect(Protection {
+                open_password: Some("open".into()),
+                permissions_password: Some("owner".into()),
+                changes: Changes::None,
+                ..Default::default()
+            }),
+        )
+        .unwrap();
+    std::fs::write(dir.join("locked.pdf"), session.save_bytes(id).unwrap().as_ref()).unwrap();
+    let result = ok(
+        &mut a,
+        "doc_bates_multi",
+        json!({"files":[
+            {"input":"b.pdf","output":"partial-numbered.pdf"},
+            {"input":"locked.pdf","output":"locked-numbered.pdf","password":"open"}
+        ]}),
+    );
+    assert_eq!(result["completed"], 1);
+    assert_eq!(result["failed"], 1);
+    assert_eq!(result["files"][1]["first"], 3);
+    assert!(result["files"][1]["error"].as_str().unwrap().contains("owner password"), "{result}");
+    assert!(dir.join("partial-numbered.pdf").exists());
+    assert!(!dir.join("locked-numbered.pdf").exists());
+}
+
 #[test]
 fn a_shrinking_header_and_footer_moves_the_page_content_out_of_their_way_through_tools() {
     let dir = workdir("hf-shrink");
@@ -7615,6 +7683,17 @@ fn exporting_to_word_html_and_rtf() {
     }
     let html = std::fs::read_to_string(dir.join("a.html")).unwrap();
     assert!(html.contains("Page 1") && html.contains("Page 3") && html.matches("<hr>").count() == 2, "{html}");
+    ok(&mut a, "doc_export_office", json!({ "doc": doc, "path": "simple.html", "html_mode": "simple" }));
+    let simple_html = std::fs::read_to_string(dir.join("simple.html")).unwrap();
+    assert!(
+        simple_html.contains("Page 1")
+            && !simple_html.contains("<style")
+            && !simple_html.contains("style=")
+            && !simple_html.contains("<span")
+            && !simple_html.contains("<div")
+    );
+    assert!(a.call("doc_export_office", &json!({ "doc": doc, "path": "bad.html", "html_mode": "compact" })).is_err());
+    assert!(a.call("doc_export_office", &json!({ "doc": doc, "path": "bad.docx", "html_mode": "simple" })).is_err());
     assert!(std::fs::read(dir.join("a.docx")).unwrap().starts_with(b"PK"));
     assert!(std::fs::read(dir.join("a.pptx")).unwrap().starts_with(b"PK"));
     assert!(std::fs::read_to_string(dir.join("a.rtf")).unwrap().contains("Page 2"));
@@ -10534,3 +10613,32 @@ fn mixed_script_tools_normalize_but_undo_restores_the_exact_mixture() {
         assert_eq!(page_text(&mut a, reopened)[0], plain);
     }
 }
+
+#[test]
+fn multiline_fill_sign_engine_round_trip_preserves_lines_and_appearance() {
+    let dir = workdir("fill-sign-multiline");
+    let mut a = auto(&dir);
+    let doc = ok(&mut a, "doc_open", json!({"path":"a.pdf"}))["doc"].as_u64().unwrap();
+    let text = "\nAda\n\nLovelace\n";
+    ok(&mut a, "fill_sign_add", json!({"doc":doc,"page":1,"type":"text","at":[20,20],"text":text}));
+    let id = pdfcraft_engine::DocId(doc);
+    let annot = &a.session().get(id).unwrap().info.annotations[0];
+    assert_eq!(annot.contents.as_deref(), Some(text), "typewriter contents keep outer line breaks");
+    assert!(annot.rect[3] - annot.rect[1] >= 66.0, "five text lines need their full height, including blank lines: {:?}", annot.rect);
+    let pixels = a.render_pixels(doc, 1, 1.0).unwrap();
+    // The fixture's page text is at y=150; only the typewriter draws at y=65..80.
+    assert!(
+        (65..80).any(|y| (20..80).any(|x| {
+            let i = (y * pixels.0 as usize + x) * 4;
+            pixels.2[i..i + 3].iter().all(|v| *v < 100)
+        })),
+        "the lower nonempty line appears inside its box"
+    );
+    let rendered = a.call("page_render", &json!({"doc":doc,"page":1,"dpi":72})).unwrap();
+    ok(&mut a, "doc_save", json!({"doc":doc,"path":"multiline.pdf"}));
+    let mut re = auto(&dir);
+    let reopened = ok(&mut re, "doc_open", json!({"path":"multiline.pdf"}))["doc"].as_u64().unwrap();
+    assert_eq!(re.session().get(pdfcraft_engine::DocId(reopened)).unwrap().info.annotations[0].contents.as_deref(), Some(text));
+    assert_eq!(rendered, re.call("page_render", &json!({"doc":reopened,"page":1,"dpi":72})).unwrap(), "saved appearances are identical");
+    std::fs::remove_dir_all(dir).unwrap();
+}

crates/cos/src/document.rs

@@ -27,7 +27,7 @@ pub(crate) use state::ObjectReader;
 use state::{ObjStm, Slot};
 pub use state::{ReadState, SourceIdentity};
 #[cfg(test)]
-use state::{SECTION_READS, objstm_member, owned_bytes};
+use state::{SECTION_READS, held_bytes, objstm_member};
 use xref::XrefIndex;
 pub use xref::{Revision, XrefEntry};
 
@@ -47,7 +47,7 @@ pub struct Document {
     /// `fill_xref_gaps`.
     located: Arc<BTreeMap<u32, (u64, u16)>>,
     repair_log: Arc<Vec<String>>,
-    cache: Arc<Mutex<HashMap<u32, Arc<Object>>>>,
+    cache: Arc<Mutex<state::ObjectCache>>,
     objstms: Arc<Mutex<HashMap<u32, Arc<ObjStm>>>>,
     /// What `/JBIG2Globals` streams of the file decode to, by object number: the images of a scan
     /// share them, and each would otherwise decode them again (see `Document::jbig2_globals`).

crates/cos/src/document/objects.rs

@@ -1,5 +1,5 @@
 use super::reader::find;
-use super::state::{LENGTHS, LoadGuard, ObjStm, ObjectReader, ReadState, Slot, SourceIdentity, objstm_member, owned_bytes};
+use super::state::{LENGTHS, LoadGuard, ObjStm, ObjectReader, ReadState, Slot, SourceIdentity, held_bytes, objstm_member};
 use super::xref::{Revision, XrefEntry};
 use super::{
     Arc, CosError, Cow, Dict, Document, Lexer, MAX_DECODED, MAX_JBIG2_GLOBALS, ObjRef, Object, Stream, is_whitespace, parse_indirect_shared,
@@ -399,8 +399,9 @@ impl Document {
         let obj = Arc::new(obj);
         // The editing cache is bounded like the scan reader's: re-parsing a file-backed view
         // is cheap, and returned objects outlive eviction.
-        self.cache.lock().map_err(|_| CosError::Poisoned)?.insert(num, obj.clone());
-        self.evict_editor_cache_if_over_budget(num, &obj)?;
+        let owned = held_bytes(&obj, &self.data, 0);
+        self.cache.lock().map_err(|_| CosError::Poisoned)?.insert(num, obj.clone(), owned);
+        self.evict_editor_cache_if_over_budget(num)?;
         Ok(obj)
     }
 
@@ -409,10 +410,10 @@ impl Document {
     /// document cannot pin its whole object graph. Only owned (decrypted/copied) bytes
     /// count; file-backed views re-parse for free.
     ///
-    /// Worst case per document: input bytes + up to `EDITOR_CACHE_OBJECTS` ×
-    /// `EDITOR_LARGE_OBJECT` (1024 × 256 KB = 256 MB) of small objects between full
-    /// accounting passes + `EDITOR_OBJSTM_BYTES` (32 MB) + unsaved `overlay` edits.
-    /// The count cap always holds; byte accounting between passes is approximate.
+    /// Worst case per document: input bytes + `EDITOR_CACHE_BYTES` (32 MB) of owned object
+    /// data + `EDITOR_OBJSTM_BYTES` (32 MB) + unsaved `overlay` edits. Each object is measured
+    /// once, as it is cached ([`ObjectCache`](super::state::ObjectCache)), so both caps hold
+    /// after every insertion.
     /// Clones share these caches, so one clone's eviction drops another's hot set;
     /// that only costs re-parsing, never correctness. Eviction keeps just the newcomer:
     /// file-backed views re-parse cheaply and editor working sets stay far below 1024.
@@ -425,37 +426,13 @@ impl Document {
 
     const EDITOR_OBJSTM_BYTES: usize = 32 * 1024 * 1024;
 
-    /// Newcomers above this force a full accounting pass even below the count cap:
-    /// `EDITOR_CACHE_BYTES / 128`, so 128 large newcomers always trigger a pass before the
-    /// byte budget can double.
-    const EDITOR_LARGE_OBJECT: usize = 256 * 1024;
-
-    fn evict_editor_cache_if_over_budget(&self, fresh_num: u32, fresh: &Object) -> Result<(), CosError> {
-        // Fast path: the full fold is O(cache) per insert, so skip it for a small newcomer
-        // below the count cap. `len` is copied out first; nothing here holds the lock.
-        let len = self.cache.lock().map_err(|_| CosError::Poisoned)?.len();
-        if len <= Self::EDITOR_CACHE_OBJECTS && owned_bytes(fresh, &self.data, 0) <= Self::EDITOR_LARGE_OBJECT {
-            return Ok(());
-        }
-        // Walk a snapshot off-lock; re-lock only to evict.
-        let live: Vec<(u32, Arc<Object>)> = self.cache.lock().map_err(|_| CosError::Poisoned)?.iter().map(|(n, o)| (*n, o.clone())).collect();
-        let data = &self.data;
-        let owned: usize = live.iter().fold(0usize, |n, (_, o)| n.saturating_add(owned_bytes(o, data, 0)));
-        if live.len() <= Self::EDITOR_CACHE_OBJECTS && owned <= Self::EDITOR_CACHE_BYTES {
-            return Ok(());
-        }
-        // Re-check under the final lock: another thread may have inserted or evicted since
-        // the snapshot. Keep its newcomers, and skip the eviction if back within budget.
-        let snapshot: std::collections::HashSet<u32> = live.into_iter().map(|(n, _)| n).collect();
+    fn evict_editor_cache_if_over_budget(&self, fresh_num: u32) -> Result<(), CosError> {
         let mut cache = self.cache.lock().map_err(|_| CosError::Poisoned)?;
-        if cache.len() <= Self::EDITOR_CACHE_OBJECTS {
-            let data = &self.data;
-            let owned: usize = cache.values().fold(0usize, |n, o| n.saturating_add(owned_bytes(o, data, 0)));
-            if owned <= Self::EDITOR_CACHE_BYTES {
-                return Ok(());
-            }
+        if cache.len() <= Self::EDITOR_CACHE_OBJECTS && cache.owned() <= Self::EDITOR_CACHE_BYTES {
+            return Ok(());
         }
-        cache.retain(|n, _| *n == fresh_num || !snapshot.contains(n));
+        // Eviction keeps just the newcomer: file-backed views re-parse cheaply.
+        cache.retain(|n| *n == fresh_num);
         Ok(())
     }
 

crates/cos/src/document/state.rs

@@ -163,11 +163,10 @@ impl ObjectReader {
         // input bytes, which the document keeps anyway, so they don't count; decrypted
         // streams and strings do. Checked after failed loads too: resolving a /Length may
         // have cached large objects.
-        if let Ok(mut cache) = self.document.cache.lock() {
-            let data = &self.document.data;
-            if cache.len() >= Self::OBJECTS || cache.values().fold(0usize, |n, o| n.saturating_add(owned_bytes(o, data, 0))) > Self::STREAM_BYTES {
-                cache.clear();
-            }
+        if let Ok(mut cache) = self.document.cache.lock()
+            && (cache.len() >= Self::OBJECTS || cache.owned() > Self::STREAM_BYTES)
+        {
+            cache.clear();
         }
         object
     }
@@ -179,26 +178,96 @@ thread_local! {
     pub(super) static SECTION_READS: std::cell::Cell<usize> = const { std::cell::Cell::new(0) };
 }
 
-/// Nesting [`owned_bytes`] walks: past the parser's own nesting limit, so every parsed object
+/// Parsed objects by number, each with what it holds ([`held_bytes`]) measured once, when it goes
+/// in, and their sum kept beside them.
+///
+/// Both caches are bounded by those bytes. Summing them on every read walked every cached object
+/// whole each time: a file whose page dictionaries carry kilobytes of private data (InDesign's
+/// `/PieceInfo`) took seconds to open, almost all of it re-measuring objects already measured
+/// (#5873).
+#[derive(Default)]
+pub(crate) struct ObjectCache {
+    entries: HashMap<u32, (Arc<Object>, usize)>,
+    owned: usize,
+}
+
+impl ObjectCache {
+    pub(crate) fn get(&self, num: &u32) -> Option<&Arc<Object>> {
+        self.entries.get(num).map(|(object, _)| object)
+    }
+
+    #[cfg(test)]
+    pub(crate) fn contains_key(&self, num: &u32) -> bool {
+        self.entries.contains_key(num)
+    }
+
+    /// Keep `object` as `num`, which holds `owned` bytes (what [`held_bytes`] says of it).
+    pub(crate) fn insert(&mut self, num: u32, object: Arc<Object>, owned: usize) {
+        if let Some((_, before)) = self.entries.insert(num, (object, owned)) {
+            self.owned = self.owned.saturating_sub(before);
+        }
+        self.owned = self.owned.saturating_add(owned);
+    }
+
+    pub(crate) fn len(&self) -> usize {
+        self.entries.len()
+    }
+
+    #[cfg(test)]
+    pub(crate) fn is_empty(&self) -> bool {
+        self.entries.is_empty()
+    }
+
+    /// The bytes the cached objects hold between them.
+    pub(crate) fn owned(&self) -> usize {
+        self.owned
+    }
+
+    pub(crate) fn clear(&mut self) {
+        self.entries.clear();
+        self.owned = 0;
+    }
+
+    pub(crate) fn retain(&mut self, mut keep: impl FnMut(&u32) -> bool) {
+        self.entries.retain(|num, _| keep(num));
+        self.owned = self.entries.values().fold(0usize, |n, (_, owned)| n.saturating_add(*owned));
+    }
+
+    #[cfg(test)]
+    pub(crate) fn values(&self) -> impl Iterator<Item = &Arc<Object>> {
+        self.entries.values().map(|(object, _)| object)
+    }
+}
+
+/// Nesting [`held_bytes`] walks: past the parser's own nesting limit, so every parsed object
 /// is measured in full.
 const OWNED_BYTES_DEPTH: usize = 256;
 
-/// Bytes `object` holds of its own: string and stream data that is not a view into `data`
-/// (the input file). An object nested deeper than [`OWNED_BYTES_DEPTH`] cannot be measured
-/// and counts as too large to keep.
-pub(super) fn owned_bytes(object: &Object, data: &Arc<Vec<u8>>, depth: usize) -> usize {
+/// What keeping `object` parsed costs: string and stream data that is not a view into `data`
+/// (the input file), and the parsed tree around it -- an [`Object`] per value and a key per
+/// dictionary entry. An object nested deeper than [`OWNED_BYTES_DEPTH`] cannot be measured and
+/// counts as too large to keep.
+///
+/// The tree counts. Measured by its strings and streams alone, a page dictionary carrying
+/// InDesign's `/PieceInfo` -- thousands of small numeric arrays, parsed into tens of thousands of
+/// values -- weighed nothing, so the caches kept a thousand of them: a 35 MB file held 1.5 GB
+/// (#5873).
+pub(super) fn held_bytes(object: &Object, data: &Arc<Vec<u8>>, depth: usize) -> usize {
     if depth > OWNED_BYTES_DEPTH {
         return usize::MAX;
     }
+    let node = std::mem::size_of::<Object>();
+    // A key, its slot beside the value, and room in the index a large dictionary keeps.
+    let entry = |key: &[u8]| key.len().saturating_add(std::mem::size_of::<(crate::Name, Object)>()).saturating_add(16);
+    let dict =
+        |d: &crate::Dict, start: usize| d.iter().fold(start, |n, (k, v)| n.saturating_add(entry(k)).saturating_add(held_bytes(v, data, depth + 1)));
     match object {
-        Object::String(s) => s.bytes.len(),
-        Object::Stream(s) => {
-            let own = if s.raw.shares(data) { 0 } else { s.raw.len() };
-            s.dict.iter().fold(own, |n, (_, v)| n.saturating_add(owned_bytes(v, data, depth + 1)))
-        }
-        Object::Array(a) => a.iter().fold(0usize, |n, v| n.saturating_add(owned_bytes(v, data, depth + 1))),
-        Object::Dict(d) => d.iter().fold(0usize, |n, (_, v)| n.saturating_add(owned_bytes(v, data, depth + 1))),
-        _ => 0,
+        Object::String(s) => node.saturating_add(s.bytes.len()),
+        Object::Name(n) => node.saturating_add(n.len()),
+        Object::Stream(s) => dict(&s.dict, node.saturating_add(if s.raw.shares(data) { 0 } else { s.raw.len() })),
+        Object::Array(a) => a.iter().fold(node, |n, v| n.saturating_add(held_bytes(v, data, depth + 1))),
+        Object::Dict(d) => dict(d, node),
+        _ => node,
     }
 }
 

crates/cos/src/document/tests.rs

@@ -1109,7 +1109,8 @@ fn temporary_reader_keeps_file_views_and_evicts_data_of_its_own() {
     // Data of its own (decrypted streams, strings) counts: past the budget the cache is
     // cleared after the next read, including a failed one.
     let owned = Object::Stream(crate::Stream::from_raw(Dict::new(), vec![b'y'; ObjectReader::STREAM_BYTES + 1]));
-    reader.document.cache.lock().unwrap().insert(3, Arc::new(owned));
+    let size = held_bytes(&owned, &reader.document.data, 0);
+    reader.document.cache.lock().unwrap().insert(3, Arc::new(owned), size);
     assert_eq!(*reader.get(ObjRef::new(99, 0)), Object::Null);
     assert!(reader.document.cache.lock().unwrap().is_empty(), "owned data over budget is evicted");
 }
@@ -1192,13 +1193,14 @@ fn editor_cache_counts_deeply_nested_strings() {
     source.update_dict(source.root().unwrap(), |d| d.set(b"Extension".to_vec(), Object::Array(objects))).unwrap();
     let bytes = Arc::new(crate::write_full(&source, &crate::SaveOptions::default()).unwrap());
     let doc = Document::open(bytes).unwrap();
-    assert_eq!(owned_bytes(&nested, &doc.data, 0), 1 << 20);
+    let held = held_bytes(&nested, &doc.data, 0);
+    assert!((1 << 20..(1 << 20) + 4096).contains(&held), "the string and a little tree: {held}");
     let root = doc.get(doc.root().unwrap());
     for reference in root.as_dict().unwrap().get(b"Extension").unwrap().as_array().unwrap() {
         let Object::Ref(reference) = reference else { panic!("indirect object") };
         drop(doc.get(*reference));
         let cache = doc.cache.lock().unwrap();
-        let owned = cache.values().fold(0usize, |n, o| n.saturating_add(owned_bytes(o, &doc.data, 0)));
+        let owned = cache.values().fold(0usize, |n, o| n.saturating_add(held_bytes(o, &doc.data, 0)));
         let strings = cache.values().filter(|o| o.as_array().is_some()).count();
         assert!(strings << 20 <= Document::EDITOR_CACHE_BYTES + (1 << 20), "{strings} MiB strings cached, {owned} bytes counted");
     }
@@ -1240,8 +1242,8 @@ fn editor_cache_stays_bounded() {
 
 #[test]
 fn editor_cache_byte_bound_fires() {
-    // 40 × 1 MB strings: far below the count cap, above the 32 MB byte budget. Each
-    // newcomer exceeds EDITOR_LARGE_OBJECT, so every insert takes a full accounting pass.
+    // 40 × 1 MB strings: far below the count cap, above the 32 MB byte budget, which is
+    // checked after every insertion.
     let mut source = Document::new_empty();
     let objects: Vec<_> = (0..40)
         .map(|i| {
@@ -1727,3 +1729,50 @@ fn a_restored_revision_keeps_what_the_saved_file_held_in_object_streams() {
     let pages = reread.resolve(catalog.get(b"Pages").unwrap());
     assert_eq!(pages.as_dict().and_then(|d| d.name(b"Type")), Some(&b"Pages"[..]), "the pages survive: {pages:?}");
 }
+
+/// The cache keeps the sum of what its objects hold as they come and go, so neither reader has
+/// to measure the cache again on every read (#5873).
+#[test]
+fn the_object_cache_keeps_its_total_as_objects_come_and_go() {
+    let data = Arc::new(Vec::new());
+    let one = Arc::new(Object::String(crate::PdfString::literal(vec![b'x'; 100])));
+    let two = Arc::new(Object::Array(vec![Object::Int(1); 10]));
+    let (a, b) = (held_bytes(&one, &data, 0), held_bytes(&two, &data, 0));
+    let mut cache = super::state::ObjectCache::default();
+    cache.insert(1, one.clone(), a);
+    cache.insert(2, two, b);
+    assert_eq!(cache.owned(), a + b);
+    cache.insert(2, one, a);
+    assert_eq!(cache.owned(), a + a, "a replaced object no longer counts");
+    cache.retain(|n| *n == 1);
+    assert_eq!((cache.len(), cache.owned()), (1, a));
+    cache.clear();
+    assert_eq!((cache.len(), cache.owned()), (0, 0));
+}
+
+/// A dictionary of numbers owns no string or stream bytes, but parsed it is tens of thousands
+/// of values: InDesign's per-page `/PieceInfo`. Counting only strings and streams, the editing
+/// cache kept a thousand such pages and a 35 MB book held 1.5 GB (#5873). What the parsed tree
+/// holds counts now, so the byte budget bounds them.
+#[test]
+fn the_editor_cache_charges_a_dictionary_for_the_values_it_holds() {
+    let entries: String = (0..2000).map(|i| format!("/{i}[{i} 141 3 116.22 -375.993 1.0]")).collect();
+    let dict = format!("<< /PieceInfo << /InDesign << /Map << {entries} >> >> >> >>");
+    let count = 300;
+    let refs: String = (0..count).map(|i| format!("{} 0 R ", i + 2)).collect();
+    let mut bodies = vec![format!("<< /Type /Catalog /Extension [{refs}] >>")];
+    bodies.extend((0..count).map(|_| dict.clone()));
+    let bodies: Vec<&str> = bodies.iter().map(String::as_str).collect();
+    let doc = Document::open(Arc::new(build(&bodies, "/Root 1 0 R"))).unwrap();
+    let one = doc.get(crate::ObjRef::new(2, 0));
+    let held = held_bytes(&one, &doc.data, 0);
+    assert!(held > 2000 * std::mem::size_of::<Object>(), "the parsed tree counts: {held} bytes");
+    for i in 0..count {
+        drop(doc.get(crate::ObjRef::new(i as u32 + 2, 0)));
+    }
+    let cache = doc.cache.lock().unwrap();
+    assert!(cache.owned() <= Document::EDITOR_CACHE_BYTES, "{} bytes held", cache.owned());
+    assert!(cache.len() < count, "the byte budget evicted some, below the count cap: {} kept", cache.len());
+    let sum = cache.values().fold(0usize, |n, o| n.saturating_add(held_bytes(o, &doc.data, 0)));
+    assert_eq!(cache.owned(), sum, "the kept total is the cache's own");
+}

crates/cos/src/parser.rs

@@ -341,7 +341,11 @@ impl<'a> Lexer<'a> {
         let s = std::str::from_utf8(tok).map_err(|_| self.err("bad number"))?;
         if s.contains('.') {
             // Tolerate forms like "--5", "5-" or "1.2.3" by parsing the longest valid prefix.
-            let cleaned: String = s.trim_start_matches('+').replacen("--", "-", 1);
+            // Only a doubled sign needs a cleaned copy: a page of InDesign `/PieceInfo` is
+            // thousands of plain reals, and building a copy and a substring search for each
+            // was a large part of reading one (#5873).
+            let t = s.trim_start_matches('+');
+            let cleaned = if t.as_bytes().windows(2).any(|w| w == b"--") { std::borrow::Cow::Owned(t.replacen("--", "-", 1)) } else { t.into() };
             let v = cleaned.get(..real_prefix_len(cleaned.as_bytes())).and_then(|p| parse_counted::<f64>(p).ok());
             return Ok(Object::Real(v.unwrap_or(0.0)));
         }
@@ -707,4 +711,13 @@ mod tests {
         let mut lx = Lexer::new(two.as_bytes(), 0);
         assert!(lx.object().is_ok() && lx.object().is_ok());
     }
+
+    /// A plain real is read without a cleaned copy of its token; the forms that need one read
+    /// as they did (#5873).
+    #[test]
+    fn reals_read_the_same_with_and_without_a_doubled_sign() {
+        for (token, want) in [("1.5", 1.5), ("-375.993", -375.993), ("+2.25", 2.25), ("--5.5", -5.5), ("1.2.3", 1.2), ("-.5", -0.5), ("0.0", 0.0)] {
+            assert_eq!(Lexer::new(token.as_bytes(), 0).object().unwrap(), Object::Real(want), "{token}");
+        }
+    }
 }

crates/engine/src/compare.rs

@@ -140,6 +140,8 @@ pub enum OfficeFormat {
     Docx,
     Pptx,
     Html,
+    /// CSS-free HTML that preserves text and inline images but drops colours and layout styling.
+    HtmlSimple,
     Rtf,
     Markdown,
 }
@@ -150,6 +152,7 @@ impl OfficeFormat {
             OfficeFormat::Docx => "docx",
             OfficeFormat::Pptx => "pptx",
             OfficeFormat::Html => "html",
+            OfficeFormat::HtmlSimple => "html",
             OfficeFormat::Rtf => "rtf",
             OfficeFormat::Markdown => "md",
         }
@@ -301,6 +304,7 @@ impl crate::Document {
             OfficeFormat::Docx => pdfcraft_export::docx(&pages, &title),
             OfficeFormat::Pptx => pdfcraft_export::pptx(&pages, &title),
             OfficeFormat::Html => pdfcraft_export::html(&pages, &title).into_bytes(),
+            OfficeFormat::HtmlSimple => pdfcraft_export::html_simple(&pages, &title).into_bytes(),
             OfficeFormat::Rtf => pdfcraft_export::rtf(&pages).into_bytes(),
             OfficeFormat::Markdown => pdfcraft_export::markdown(&pages).into_bytes(),
         };

crates/engine/src/tests.rs

@@ -10330,3 +10330,20 @@ fn image_transform_engine_refuses_singular_matrices_without_mutation() {
         assert_eq!(s.get(id).unwrap().bytes, before);
     }
 }
+
+#[test]
+fn fill_sign_multiline_text_survives_engine_save_reopen_and_render() {
+    let (mut session, id) = rotated_blank(0);
+    let text = "\nAda\n\nLovelace\n";
+    let shape = Shape::Typewriter { rect: text_box_at(0, [20.0, 250.0], 120.0, 66.0).unwrap(), font_size: 10.0 };
+    let style = Style::default_for(&shape);
+    session.apply(id, Edit::AddAnnotation(NewAnnotation { page: 0, shape, style, contents: text.into(), author: "Test".into() })).unwrap();
+    assert_eq!(session.get(id).unwrap().info.annotations[0].contents.as_deref(), Some(text), "engine keeps outer typewriter line breaks");
+    let before = render_page(&session, id);
+    assert!(before.rgba.as_chunks::<4>().0.iter().any(|p| p[..3] == [0, 0, 0]), "the appearance contains ink");
+    let bytes = session.save_bytes(id).unwrap();
+    let mut reopened = Session::new();
+    let again = reopened.open("multiline.pdf", None, bytes, None).unwrap();
+    assert_eq!(reopened.get(again).unwrap().info.annotations[0].contents.as_deref(), Some(text), "reopening preserves all blank lines");
+    assert_eq!(before.rgba, render_page(&reopened, again).rgba, "the saved appearance is identical");
+}

crates/export/src/as_html.rs

@@ -38,11 +38,23 @@ pub(crate) fn base64(data: &[u8]) -> String {
 /// One HTML file: headings and paragraphs, tables as real `<table>` elements, images inline, a
 /// rule between pages.
 pub fn html(pages: &[Page], title: &str) -> String {
+    html_with_mode(pages, title, false)
+}
+
+/// A simple HTML export with no CSS; page layout and colours are omitted.
+pub fn html_simple(pages: &[Page], title: &str) -> String {
+    html_with_mode(pages, title, true)
+}
+
+fn html_with_mode(pages: &[Page], title: &str, simple: bool) -> String {
     let mut s = format!(
-        "<!DOCTYPE html>\n<html>\n<head>\n<meta charset=\"utf-8\">\n<title>{}</title>\n<style>body{{max-width:46em;margin:2em auto;font-family:sans-serif;line-height:1.45}}\
-img{{max-width:100%}}hr{{border:0;border-top:1px solid #ccc;margin:2em 0}}\
-table{{border-collapse:collapse;margin:1em 0}}td,th{{border:1px solid #999;padding:2px 6px;vertical-align:top}}</style>\n</head>\n<body>\n",
-        esc(title)
+        "<!DOCTYPE html>\n<html>\n<head>\n<meta charset=\"utf-8\">\n<title>{}</title>\n{}\n</head>\n<body>\n",
+        esc(title),
+        if simple {
+            ""
+        } else {
+            "<style>body{max-width:46em;margin:2em auto;font-family:sans-serif;line-height:1.45}img{max-width:100%}hr{border:0;border-top:1px solid #ccc;margin:2em 0}table{border-collapse:collapse;margin:1em 0}td,th{border:1px solid #999;padding:2px 6px;vertical-align:top}</style>"
+        }
     );
     for it in items(pages) {
         match it {
@@ -59,13 +71,13 @@ table{{border-collapse:collapse;margin:1em 0}}td,th{{border:1px solid #999;paddi
                     }
                     // The paragraph has its first run's colour; a run in another colour says so.
                     let colour = rgb8(r.block.color);
-                    if colour != rgb8(first.color) {
+                    if !simple && colour != rgb8(first.color) {
                         let [red, green, blue] = colour;
                         run = format!("<span style=\"color:#{red:02X}{green:02X}{blue:02X}\">{run}</span>");
                     }
                     t.push_str(&run);
                 }
-                let style = hex(first.color).map_or(String::new(), |h| format!(" style=\"color:#{h}\""));
+                let style = if simple { String::new() } else { hex(first.color).map_or(String::new(), |h| format!(" style=\"color:#{h}\"")) };
                 match lvl {
                     0 => s.push_str(&format!("<p{style}>{t}</p>\n")),
                     l => s.push_str(&format!("<h{l}{style}>{t}</h{l}>\n")),
@@ -84,7 +96,7 @@ table{{border-collapse:collapse;margin:1em 0}}td,th{{border:1px solid #999;paddi
                             body = format!("<strong>{body}</strong>");
                         }
                         let span = if c.span > 1 { format!(" colspan=\"{}\"", c.span) } else { String::new() };
-                        let style = hex(c.color).map_or(String::new(), |h| format!(" style=\"color:#{h}\""));
+                        let style = if simple { String::new() } else { hex(c.color).map_or(String::new(), |h| format!(" style=\"color:#{h}\"")) };
                         s.push_str(&format!("<td{span}{style}>{body}</td>"));
                     }
                     s.push_str("</tr>\n");

crates/export/src/lib.rs

@@ -27,9 +27,9 @@ pub(crate) use as_docx::hex;
 #[cfg(test)]
 use as_docx::{MAX_MEDIA, docx_table};
 pub(crate) use as_html::esc;
-pub use as_html::html;
 #[cfg(test)]
 use as_html::{allowed, base64};
+pub use as_html::{html, html_simple};
 pub use as_markdown::markdown;
 #[cfg(test)]
 use as_markdown::{emphasis, paragraph};

crates/export/src/tests.rs

@@ -85,6 +85,24 @@ fn html_has_headings_paragraphs_and_images_in_order() {
     assert_eq!(base64(b"Ma"), "TWE=");
 }
 
+#[test]
+fn simple_html_has_no_css_or_layout_elements() {
+    let mut p = page();
+    let mut unicode = block("naïve 日本語.", 72.0, 300.0, 300.0, 311.0);
+    unicode.bold = true;
+    unicode.italic = true;
+    p.blocks.push(unicode);
+    let source = html_simple(&[p, table_page()], "Simple export");
+    for banned in ["<style", "style=", "<div", "<span"] {
+        assert!(!source.contains(banned), "{banned} in {source}");
+    }
+    for wanted in ["<h1", "<p>", "<strong>", "<em>", "<table", "naïve 日本語."] {
+        assert!(source.contains(wanted), "no {wanted} in {source}");
+    }
+    // The rich export of the same pages keeps its stylesheet: only simple mode drops it.
+    assert!(html(&[page()], "Rich").contains("<style"));
+}
+
 #[test]
 fn docx_is_a_word_package() {
     let d = docx(&[page()], "Report");

crates/render/src/inspect/core.rs

@@ -288,11 +288,23 @@ impl<'a> Inspector<'a> {
     }
 
     pub(super) fn text(&self, d: &Dictionary, key: &[u8]) -> Option<String> {
-        let s = trimmed_text(self.resolve(d.get(key).ok()?))?;
-        (!s.is_empty()).then_some(s)
+        // Typewriter outer whitespace belongs to the text (keep in step with annot's summaries).
+        let s = if key == b"Contents"
+            && self.name(d, b"Subtype").as_deref() == Some("FreeText")
+            && self.name(d, b"IT").as_deref() == Some("FreeTextTypeWriter")
+        {
+            match self.resolve(d.get(key).ok()?) {
+                Object::String(bytes, _) => text_string(bytes).trim_matches('\0').to_owned(),
+                Object::Name(bytes) => String::from_utf8_lossy(bytes).trim_matches('\0').to_owned(),
+                _ => return None,
+            }
+        } else {
+            trimmed_text(self.resolve(d.get(key).ok()?))?
+        };
+        (!s.trim().is_empty()).then_some(s)
     }
 
-    /// [`Self::text`], converted and spent from `budget` as [`spend`] does, so text that trims
+    /// Whitespace-trimmed metadata, spent from `budget` as [`spend`] does, so text that trims
     /// short or to nothing still spends its raw length.
     pub(super) fn budgeted_text(&self, d: &Dictionary, key: &[u8], budget: &mut usize) -> Option<String> {
         let o = self.resolve(d.get(key).ok()?);

crates/render/src/inspect/tests.rs

@@ -1870,3 +1870,45 @@ trailer << /Root 1 0 R >>
     assert!((p.width - 3874.96).abs() < 0.1, "width expected ~3874.96, got {}", p.width);
     assert!((p.height - 5493.54).abs() < 0.1, "height expected ~5493.54, got {}", p.height);
 }
+
+/// Inspection keeps every object it reads, converted, for as long as it runs, and nothing it
+/// reports comes from `/PieceInfo`. InDesign writes kilobytes of that private data into every
+/// page, and a 1,412-page book held over a gigabyte of it (#5873). It is left out of what
+/// inspection keeps, and what inspection reports is the same as for the page without it.
+#[test]
+fn inspection_keeps_no_private_page_data() {
+    use lopdf::dictionary;
+    let make = |private: bool| -> Vec<u8> {
+        let mut doc = Document::with_version("1.7");
+        let pages_id = doc.new_object_id();
+        let mut page = dictionary! {
+            "Type" => "Page",
+            "Parent" => pages_id,
+            "MediaBox" => vec![0.into(), 0.into(), 200.into(), 300.into()],
+            "Rotate" => 90,
+        };
+        if private {
+            let map: Dictionary = (0..50).map(|i| (i.to_string(), Object::Array(vec![Object::Integer(i), Object::Real(116.22)]))).collect();
+            page.set("PieceInfo", dictionary! { "InDesign" => dictionary! { "PageItemUIDToLocationDataMap" => map } });
+        }
+        let page_id = doc.add_object(page);
+        doc.objects.insert(pages_id, dictionary! { "Type" => "Pages", "Kids" => vec![page_id.into()], "Count" => 1 }.into());
+        let catalog = doc.add_object(dictionary! { "Type" => "Catalog", "Pages" => pages_id });
+        doc.trailer.set("Root", catalog);
+        let mut out = Vec::new();
+        doc.save_to(&mut out).unwrap();
+        out
+    };
+    let (with, without) = (make(true), make(false));
+    let lazy = LazyStructure::new(Arc::new(with.clone()), None).unwrap();
+    let page = lazy.get_object((2, 0)).unwrap().as_dict().unwrap();
+    assert!(!page.has(b"PieceInfo"), "the private data is not kept");
+    assert!(page.has(b"MediaBox") && page.has(b"Rotate") && page.has(b"Parent"), "the rest of the page is");
+
+    let report = |bytes: Vec<u8>| {
+        let mut info = inspect(Arc::new(bytes), None).unwrap();
+        info.file_size = 0;
+        format!("{info:?}")
+    };
+    assert_eq!(report(with), report(without), "what inspection reports does not change");
+}

crates/render/src/raster/tests.rs

@@ -8218,6 +8218,61 @@ fn japanese_substitute_spaces_keep_pdf_widths_and_unicode() {
     }
 }
 
+/// One code, `<0010>`, in a Type 0 font named `base_font` that the PDF doesn't embed, with a
+/// `/ToUnicode` that maps it to the UTF-16 `unicode`. 0x10 is the hyphen's glyph index in Arial,
+/// and Identity-H with an identity `/CIDToGIDMap` makes the code that index, as Word writes it.
+fn soft_hyphen_pdf(base_font: &str, unicode: &str) -> Vec<u8> {
+    let content = "BT /F1 60 Tf 10 20 Td <0010> Tj ET";
+    let cmap = format!(
+        "/CIDInit /ProcSet findresource begin 12 dict begin begincmap
+/CIDSystemInfo << /Registry (Adobe) /Ordering (UCS) /Supplement 0 >> def
+/CMapName /Adobe-Identity-UCS def /CMapType 2 def
+1 begincodespacerange <0000> <FFFF> endcodespacerange
+1 beginbfchar <0010> <{unicode}> endbfchar
+endcmap CMapName currentdict /CMap defineresource pop end end"
+    );
+    format!(
+        "%PDF-1.7
+1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj
+2 0 obj << /Type /Pages /Kids [3 0 R] /Count 1 >> endobj
+3 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 60 60] /Contents 4 0 R /Resources << /Font << /F1 5 0 R >> >> >> endobj
+4 0 obj << /Length {} >> stream
+{content}
+endstream endobj
+5 0 obj << /Type /Font /Subtype /Type0 /BaseFont /{base_font} /Encoding /Identity-H /DescendantFonts [6 0 R] /ToUnicode 8 0 R >> endobj
+6 0 obj << /Type /Font /Subtype /CIDFontType2 /BaseFont /{base_font} /CIDSystemInfo << /Registry (Adobe) /Ordering (Identity) /Supplement 0 >>
+  /FontDescriptor 7 0 R /CIDToGIDMap /Identity /DW 1000 /W [16 [333]] >> endobj
+7 0 obj << /Type /FontDescriptor /FontName /{base_font} /Flags 32 /FontBBox [-665 -325 2000 1040] /ItalicAngle 0 /Ascent 728 /Descent -210 /CapHeight 716 /StemV 80 >> endobj
+8 0 obj << /Length {} >> stream
+{cmap}
+endstream endobj
+trailer << /Root 1 0 R >>
+%%EOF",
+        content.len(),
+        cmap.len()
+    )
+    .into_bytes()
+}
+
+/// #5888: Windows exporters map the hyphen glyph of an Arial they don't embed to U+00AD. The
+/// standard substitute has no soft hyphen, so code 0x0010 was taken as its glyph 16 and drew
+/// `Ë` (`!` in bold). It draws the substitute's hyphen now, and the text keeps the soft hyphen.
+#[test]
+fn a_soft_hyphen_a_substitute_lacks_draws_its_hyphen() {
+    for base_font in ["ArialMT", "Arial-BoldMT"] {
+        let soft = soft_hyphen_pdf(base_font, "00AD");
+        let out = render_first_page(soft.clone());
+        let expected = render_first_page(soft_hyphen_pdf(base_font, "002D"));
+        assert!(expected.rgba.as_chunks::<4>().0.iter().any(|pixel| pixel[0] < 128), "{base_font}: the hyphen control must draw ink");
+        assert!(out.rgba.as_ref() == expected.rgba.as_ref(), "{base_font}: a soft hyphen draws as the substitute's hyphen");
+        let mut renderer = PageRenderer::new(Arc::new(soft), RenderConfig::default());
+        let text = renderer.render(RenderRequest { page: 0, kind: RequestKind::Text, ..Default::default() });
+        assert!(text.error.is_none(), "{:?}", text.error);
+        let glyphs: String = text.text.unwrap().glyphs.iter().map(|glyph| glyph.text.as_str()).collect();
+        assert_eq!(glyphs, "\u{ad}", "{base_font}: drawing the hyphen must not rewrite Unicode");
+    }
+}
+
 /// Half-width katakana are common in Japanese documents set in a non-embedded Mincho font
 /// such as HeiseiMin-W3. The Mincho substitute must have those glyphs: one that lacks them
 /// (Shippori Mincho) draws them as .notdef.

crates/render/src/structure.rs

@@ -187,9 +187,20 @@ fn convert_dict(dict: &pdfcraft_cos::Dict, depth: usize) -> Result<Dictionary> {
     if depth > 100 {
         return Err(Error::RecursionLimit);
     }
-    dict.iter().map(|(key, value)| Ok((key.clone(), convert(value, depth + 1)?))).collect::<Result<Vec<_>>>().map(|items| items.into_iter().collect())
+    dict.iter()
+        .filter(|(key, _)| key.as_slice() != PRIVATE_DATA)
+        .map(|(key, value)| Ok((key.clone(), convert(value, depth + 1)?)))
+        .collect::<Result<Vec<_>>>()
+        .map(|items| items.into_iter().collect())
 }
 
+/// `/PieceInfo`: an application's private data about a page, form or document (ISO 32000-1
+/// §14.5), left out of what inspection keeps. Nothing here reads it, and every object read is
+/// kept converted for the inspection's lifetime: InDesign writes kilobytes of it into every page,
+/// and a 1,412-page book held over a gigabyte of it (#5873). The file and the editing document are
+/// untouched; this is the read-only view inspection walks.
+const PRIVATE_DATA: &[u8] = b"PieceInfo";
+
 fn convert(object: &CosObject, depth: usize) -> Result<Object> {
     if depth > 100 {
         return Err(Error::RecursionLimit);

crates/ui-egui/src/dialogs.rs

@@ -20,6 +20,9 @@ pub(super) mod protect;
 pub(super) mod standards_ui;
 pub(super) mod word_count_ui;
 
+#[cfg(test)]
+mod tests;
+
 pub(crate) use prompts::{PromptButton, save_prompt_key};
 use prompts::{full_screen_prompt, link_prompt, password, remote_link_prompt, save_prompt, unlock};
 
@@ -1371,20 +1374,26 @@ fn dialog_preferences(ui: &mut egui::Ui, app: &mut PdfCraftApp, req: &mut Pendin
     // (which only applies when the settings overflow it) opens the dialog at full size.
     // A scrollbar of its own: the floating one lies over the right edge of the theme
     // cards and other full-width rows (#3557).
+    preferences_scroll(ui, max, |ui| preferences::preferences_body(ui, app, &t));
+    req.close = preferences::preferences_buttons(ui);
+    false
+}
+
+fn preferences_scroll(ui: &mut egui::Ui, max: f32, add_contents: impl FnOnce(&mut egui::Ui)) {
     ui.scope(|ui| {
         ui.spacing_mut().scroll.floating = false;
         ui.spacing_mut().scroll.bar_inner_margin = 14.0;
         ui.spacing_mut().scroll.bar_outer_margin = 2.0;
-        ui.style_mut().scroll_animation = egui::style::ScrollAnimation::duration(0.25);
+        if ui.style().scroll_animation != egui::style::ScrollAnimation::none() {
+            ui.style_mut().scroll_animation = egui::style::ScrollAnimation::duration(0.25);
+        }
         egui::ScrollArea::vertical()
             .id_salt("preferences-dialog-scroll")
             .max_height(max)
             .min_scrolled_height(max)
             .animated(true)
-            .show(ui, |ui| preferences::preferences_body(ui, app, &t));
+            .show(ui, add_contents);
     });
-    req.close = preferences::preferences_buttons(ui);
-    false
 }
 
 /// `Dialog::NumberPages` content. Returns `false` when it has its own buttons and no footer is wanted.

crates/ui-egui/src/dialogs/tests.rs

@@ -0,0 +1,35 @@
+use std::cell::Cell;
+use std::rc::Rc;
+
+use egui_kittest::Harness;
+use egui_kittest::kittest::Queryable;
+
+#[test]
+fn preferences_scroll_preserves_disabled_animation() {
+    for animated in [false, true] {
+        let scroll = Rc::new(Cell::new(false));
+        let requested = scroll.clone();
+        let mut h = Harness::builder().with_size(egui::vec2(600.0, 400.0)).with_step_dt(1.0 / 60.0).build_ui(move |ui| {
+            super::preferences_scroll(ui, 200.0, |ui| {
+                ui.add_space(1000.0);
+                let target = ui.label("Scroll target");
+                if requested.replace(false) {
+                    ui.scroll_to_rect(target.rect, Some(egui::Align::BOTTOM));
+                }
+                ui.add_space(300.0);
+            });
+        });
+        h.ctx.all_styles_mut(|style| {
+            style.scroll_animation = if animated { egui::style::ScrollAnimation::default() } else { egui::style::ScrollAnimation::none() };
+        });
+        h.run_steps(4);
+        let before = h.get_by_label("Scroll target").rect().top();
+        scroll.set(true);
+        h.run_steps(4);
+        let early = h.get_by_label("Scroll target").rect().top();
+        h.run_steps(20);
+        let settled = h.get_by_label("Scroll target").rect().top();
+        assert!((before - settled).abs() > 50.0, "the target must scroll: {before} -> {settled}");
+        assert_eq!((early - settled).abs() > 1.0, animated, "animated: {animated}, early: {early}, settled: {settled}");
+    }
+}

crates/ui-egui/src/tools/comments/composer.rs

@@ -111,7 +111,10 @@ pub(crate) fn composer(ctx: &egui::Context, view: &mut DocView, info: &DocInfo,
         return None;
     }
     let c = view.comments.composer.take()?;
-    let text = c.text.trim_end().to_string();
+    let typewriter_edit = matches!(c.kind, ComposerKind::Edit(index) if info.annotations.iter().any(|a|
+        a.page == page && a.index == index && a.subtype == "FreeText" && a.intent.as_deref() == Some("FreeTextTypeWriter")
+    ));
+    let text = if typewriter_edit { c.text } else { c.text.trim_end().to_string() };
     match c.kind {
         ComposerKind::Edit(index) => Some(Edit::SetAnnotationContents { page, index, text }),
         ComposerKind::Note => {

crates/ui-egui/src/tools/fill_sign.rs

@@ -490,7 +490,7 @@ pub(crate) fn page_input(cx: PageToolContext<'_>, input: PageInput<'_>, view: &m
                 FillAction::Edit(Box::new(typed_with_color(
                     tb.page,
                     tb.at,
-                    tb.text.trim(),
+                    &tb.text,
                     tb.size,
                     crate::comments::page_rotation(info, tb.page),
                     author,
@@ -555,14 +555,20 @@ pub(crate) fn type_box(ctx: &egui::Context, view: &mut DocView, info: &DocInfo,
             let Some(t) = view.fill_text.as_mut() else { return };
             // The box is white in every theme, so the caret must not follow the theme (pale in dark mode) (#655).
             ui.visuals_mut().text_cursor.stroke.color = crate::edit_text_ui::caret_color_on(Color32::WHITE);
-            let width = ((t.text.chars().count().max(8) as f32) * t.size as f32 * 0.6 * zoom).clamp(60.0, 600.0);
+            let font = egui::FontId::proportional((t.size as f32 * zoom).max(8.0));
+            let natural = ui.fonts_mut(|f| f.layout_no_wrap(t.text.clone(), font.clone(), Color32::BLACK).size().x);
+            let margin = egui::Margin::symmetric(4, 2);
+            // Include the field padding and a pixel for layout rounding in the measured width.
+            let width = (natural + margin.sum().x + 1.0).max(8.0 * t.size as f32 * 0.6 * zoom).clamp(60.0, 600.0);
             // The area's max rect is last frame's size, which caps the edit's width; widen it so the
             // box grows as text is typed.
             ui.set_min_width(width);
             let r = ui
                 .add(
-                    egui::TextEdit::singleline(&mut t.text)
-                        .font(egui::FontId::proportional((t.size as f32 * zoom).max(8.0)))
+                    egui::TextEdit::multiline(&mut t.text)
+                        .margin(margin)
+                        .desired_rows(1)
+                        .font(font)
                         .desired_width(width)
                         .background_color(Color32::from_rgba_unmultiplied(255, 255, 255, 230))
                         .text_color(crate::comments::color32(t.color))
@@ -587,7 +593,7 @@ pub(crate) fn type_box(ctx: &egui::Context, view: &mut DocView, info: &DocInfo,
     if commit {
         let tb = view.fill_text.take()?;
         if !tb.text.trim().is_empty() {
-            return Some(typed_with_color(tb.page, tb.at, tb.text.trim(), tb.size, crate::comments::page_rotation(info, tb.page), author, tb.color));
+            return Some(typed_with_color(tb.page, tb.at, &tb.text, tb.size, crate::comments::page_rotation(info, tb.page), author, tb.color));
         }
     }
     None

crates/ui-egui/tests/it/comments.rs

@@ -2925,7 +2925,7 @@ fn moved_fill_and_sign_text_is_drawn_where_it_goes_before_release() {
     click_pt(&mut h, (40.0, 150.0));
     h.event(egui::Event::Text("WWWWWW".into()));
     h.run_steps(1);
-    h.key_press(egui::Key::Enter);
+    click(&mut h, pos2(350.0, 850.0)); // Outside the field: Enter now inserts a line break.
     h.run_steps(3);
     let a = comments(&h).into_iter().find(|a| a.contents.as_deref() == Some("WWWWWW")).expect("the typed text");
     assert!(a.fill_sign && a.subtype == "FreeText", "{a:?}");
@@ -3379,3 +3379,85 @@ fn hotbar_floating_picker_does_not_overlap_rendered_bar() {
         settings_rect
     );
 }
+
+fn edit_composer_preserves_only_typewriter_whitespace(kind: u8) {
+    let shots = std::env::var_os("PDFCRAFT_COMPOSER_SHOTS").is_some();
+    let _gpu = shots.then(crate::gpu);
+    use pdfcraft_engine::{Edit, NewAnnotation, NoteIcon, Shape, Style};
+    let mut h = harness(move |app| {
+        app.set_option("layout", "single").unwrap();
+        app.set_option("panel", "comments").unwrap();
+        app.quick_tool = QuickTool::Select;
+        // Both pages use index 0, with opposite intents. The lookup must use page AND index.
+        for page in 0..2 {
+            let typewriter = if page == 0 { kind != 0 } else { kind == 0 };
+            let shape = if typewriter {
+                Shape::Typewriter { rect: [40.0, 40.0, 180.0, 90.0], font_size: 10.0 }
+            } else if kind == 2 {
+                Shape::Note { at: [80.0, 80.0], icon: NoteIcon::Comment }
+            } else {
+                Shape::TextBox { rect: [40.0, 40.0, 180.0, 90.0], font_size: 10.0 }
+            };
+            let style = Style::default_for(&shape);
+            assert!(app.apply_edit(Edit::AddAnnotation(NewAnnotation { page, shape, style, contents: "Original".into(), author: "Tester".into() })));
+        }
+    });
+    h.state_mut().views[0].go_to_page(1);
+    h.run_steps(4);
+    support::settle(&mut h);
+    let point = {
+        let s = h.state();
+        let v = &s.views[0];
+        let doc = s.session.get(v.id).unwrap();
+        let a = doc.info.annotations.iter().find(|a| a.page == 1 && a.index == 0).unwrap();
+        let page = v.page_screen_rect(1).expect("the second page is visible");
+        let info = &doc.info.pages[1];
+        let p = info.user_to_view((a.rect[0] + a.rect[2]) / 2.0, (a.rect[1] + a.rect[3]) / 2.0);
+        page.min + egui::vec2(p[0], p[1]) * (page.width() / info.width)
+    };
+    support::input::double_click(&mut h, point);
+    h.run_steps(3);
+    let c = h.state().views[0].comments.composer.as_ref().expect("double-click opens the edit composer");
+    assert_eq!(c.page, 1);
+    assert_eq!(c.kind, pdfcraft_ui_egui::comments::ComposerKind::Edit(0));
+    h.input_mut().events.push(egui::Event::Text(" edited\n\n".into()));
+    let ctx = h.ctx.clone();
+    let mut raw = std::mem::take(h.input_mut());
+    h.state_mut().prepare_raw_input(&ctx, &mut raw);
+    *h.input_mut() = raw;
+    h.step();
+    h.run_steps(2);
+    assert_eq!(h.state().views[0].comments.composer.as_ref().unwrap().text, "Original edited\n\n");
+    if shots {
+        h.render().unwrap().save(format!("/tmp/5954-composer-{kind}.png")).unwrap();
+    }
+    h.get_all_by_label("Post").last().unwrap().click();
+    h.run_steps(3);
+    assert!(h.state().views[0].comments.composer.is_none());
+    let id = h.state().views[0].id;
+    let bytes = h.state().session.save_bytes(id).unwrap();
+    // Check the raw saved string: ordinary-comment reader trimming must not hide a bad post.
+    let cos = pdfcraft_cos::Document::open(bytes.clone()).unwrap();
+    let pages = pdfcraft_model::pages(&cos);
+    let annots = cos.resolve(pages[1].dict.get(b"Annots").unwrap());
+    let annot = cos.resolve(&annots.as_array().unwrap()[0]);
+    let contents = cos.resolve(annot.as_dict().unwrap().get(b"Contents").unwrap());
+    let expected = if kind == 0 { "Original edited\n\n" } else { "Original edited" };
+    assert_eq!(contents.as_string().unwrap().to_text(), expected, "only a typewriter edit preserves trailing blank lines");
+    let mut reopened = pdfcraft_engine::Session::new();
+    let again = reopened.open("edited.pdf", None, bytes, None).unwrap();
+    let saved = reopened.get(again).unwrap().info.annotations.iter().find(|a| a.page == 1 && a.index == 0).unwrap();
+    assert_eq!(saved.contents.as_deref(), Some(expected));
+}
+
+#[test]
+fn typewriter_edit_composer_keeps_trailing_blank_lines() {
+    edit_composer_preserves_only_typewriter_whitespace(0);
+}
+
+#[test]
+fn other_edit_composers_keep_trimming_trailing_whitespace() {
+    for kind in [1, 2] {
+        edit_composer_preserves_only_typewriter_whitespace(kind);
+    }
+}

crates/ui-egui/tests/it/fill_sign.rs

@@ -130,7 +130,7 @@ fn text_marks_and_date() {
     click(&mut h, 50.0, 250.0);
     h.event(egui::Event::Text("Ada Lovelace".into()));
     h.run_steps(1);
-    h.key_press(egui::Key::Enter);
+    commit_typed_text(&mut h);
     h.run_steps(4);
     let v = items(&h);
     assert_eq!(v.len(), 4, "{v:?}");
@@ -139,9 +139,9 @@ fn text_marks_and_date() {
     assert!(v.iter().any(|(t, _)| t == "Stamp"));
 }
 
-/// The open type box, found as the focused text input.
+/// The open type box, found by its accessible name for either input role.
 fn type_box_rect(h: &Harness<'static, PdfCraftApp>) -> egui::Rect {
-    h.query_all_by_role(egui::accesskit::Role::TextInput).find(|n| n.is_focused()).expect("an open type box").rect()
+    h.get_by_label("Text to place").rect()
 }
 
 #[test]
@@ -149,13 +149,13 @@ fn clicking_elsewhere_with_add_text_keeps_the_typed_text() {
     let mut h = harness();
     h.state_mut().execute("sign.fill.text");
     click(&mut h, 50.0, 250.0);
-    h.event(egui::Event::Text("Ada Lovelace".into()));
+    h.event(egui::Event::Text("\nAda\nLovelace\n".into()));
     h.run_steps(1);
     // Still on Add text: the next click starts a new box, and must keep the first one.
     click(&mut h, 50.0, 150.0);
     h.run_steps(2);
     let v = items(&h);
-    assert_eq!(v, vec![("FreeText".to_string(), Some("Ada Lovelace".to_string()))]);
+    assert_eq!(v, vec![("FreeText".to_string(), Some("\nAda\nLovelace\n".to_string()))]);
     assert!(h.state().views[0].fill_text.as_ref().is_some_and(|t| t.text.is_empty()), "a new, empty box is open");
 }
 
@@ -170,7 +170,7 @@ fn fill_text_size_controls_new_text_and_survives_settings_restore() {
     assert_eq!(h.state().views[0].fill_text.as_ref().map(|t| t.size), Some(24.0));
     h.event(egui::Event::Text("Larger text".into()));
     h.run_steps(1);
-    h.key_press(egui::Key::Enter);
+    commit_typed_text(&mut h);
     h.run_steps(3);
     let annotation =
         h.state().session.get(h.state().views[0].id).unwrap().info.annotations.iter().find(|a| a.contents.as_deref() == Some("Larger text")).unwrap();
@@ -192,7 +192,7 @@ fn fill_text_size_resizes_the_selected_text() {
     click(&mut h, 50.0, 250.0);
     h.event(egui::Event::Text("Resize me".into()));
     h.run_steps(1);
-    h.key_press(egui::Key::Enter);
+    commit_typed_text(&mut h);
     h.run_steps(3);
     let height = |h: &Harness<'static, PdfCraftApp>| {
         let s = h.state();
@@ -297,7 +297,7 @@ fn fill_text_colour_survives_preview_commit_settings_and_save() {
     assert_eq!(h.state().views[0].fill_text.as_ref().unwrap().color, color, "the editor uses the selected colour");
     h.event(egui::Event::Text("Blue form text".into()));
     h.run_steps(2);
-    h.key_press(egui::Key::Enter);
+    commit_typed_text(&mut h);
     h.run_steps(3);
     let id = h.state().views[0].id;
     let expected = color.map(|v| v as f32);
@@ -1812,3 +1812,79 @@ fn mode_tabs_keep_the_hand_over_a_type_box_scrolled_under_them() {
     h.run_steps(3);
     assert_eq!(h.state().mode, pdfcraft_ui_egui::Mode::Convert, "the tab takes the click, not the text box");
 }
+
+fn commit_typed_text(h: &mut Harness<'static, PdfCraftApp>) {
+    let p = pos2(350.0, 850.0); // Blank canvas outside the page and the type box.
+    h.hover_at(p);
+    h.drag_at(p);
+    h.run_steps(1);
+    h.drop_at(p);
+    h.run_steps(3);
+}
+
+fn fill_input(h: &mut Harness<'static, PdfCraftApp>, events: impl IntoIterator<Item = egui::Event>) {
+    h.input_mut().events.extend(events);
+    let ctx = h.ctx.clone();
+    let mut raw = std::mem::take(h.input_mut());
+    h.state_mut().prepare_raw_input(&ctx, &mut raw);
+    *h.input_mut() = raw;
+    h.step();
+    h.run_steps(2);
+}
+
+fn fill_key(h: &mut Harness<'static, PdfCraftApp>, key: egui::Key) {
+    fill_input(
+        h,
+        [true, false].map(|pressed| egui::Event::Key { key, physical_key: Some(key), pressed, repeat: false, modifiers: egui::Modifiers::NONE }),
+    );
+}
+
+#[test]
+fn fill_sign_enter_adds_lines_outside_click_commits_and_escape_cancels() {
+    let shots = std::env::var_os("PDFCRAFT_FILL_SIGN_SHOTS").is_some();
+    let _gpu = shots.then(gpu);
+    let mut h = harness();
+    assert!(h.state_mut().execute("sign.fill.text"));
+    click(&mut h, 50.0, 300.0);
+    let initial_height = type_box_rect(&h).height();
+    fill_key(&mut h, egui::Key::Enter);
+    assert_eq!(h.state().views[0].fill_text.as_ref().map(|t| t.text.as_str()), Some("\n"), "Enter inserts a newline and keeps the editor open");
+    fill_input(&mut h, [egui::Event::Text("Ada".into())]);
+    fill_key(&mut h, egui::Key::Enter);
+    fill_key(&mut h, egui::Key::Enter);
+    fill_input(&mut h, [egui::Event::Text("Lovelace".into())]);
+    fill_key(&mut h, egui::Key::Enter);
+    let text = "\nAda\n\nLovelace\n";
+    assert_eq!(h.state().views[0].fill_text.as_ref().unwrap().text, text);
+    assert!(items(&h).is_empty(), "Enter never places an annotation");
+    let galley = h
+        .output()
+        .shapes
+        .iter()
+        .find_map(|s| match &s.shape {
+            egui::Shape::Text(t) if t.galley.text() == text => Some(&t.galley),
+            _ => None,
+        })
+        .unwrap();
+    assert_eq!(galley.rows.len(), 5, "short typed lines fit without accidental soft wraps");
+    assert!(type_box_rect(&h).height() > initial_height * 3.0, "the editor grows with its lines");
+    if shots {
+        h.render().unwrap().save("/tmp/5850-multiline.png").unwrap();
+    }
+    commit_typed_text(&mut h);
+    assert!(h.state().views[0].fill_text.is_none());
+    assert_eq!(items(&h), [("FreeText".to_owned(), Some(text.to_owned()))]);
+    let id = h.state().views[0].id;
+    let saved = h.state().session.save_bytes(id).unwrap();
+    let mut session = pdfcraft_engine::Session::new();
+    let reopened = session.open("again.pdf", None, saved.clone(), None).unwrap();
+    let annot = &session.get(reopened).unwrap().info.annotations[0];
+    assert_eq!(annot.contents.as_deref(), Some(text));
+    assert!(annot.rect[3] - annot.rect[1] >= 5.0 * 10.0 * 1.2, "the saved box keeps all five lines");
+    click(&mut h, 50.0, 180.0);
+    fill_input(&mut h, [egui::Event::Text("Cancel me".into())]);
+    fill_key(&mut h, egui::Key::Enter);
+    fill_key(&mut h, egui::Key::Escape);
+    assert!(h.state().views[0].fill_text.is_none());
+    assert_eq!(h.state().session.save_bytes(id).unwrap(), saved, "Escape cancels without writing");
+}

docs/roadmap.md

@@ -1,5 +1,9 @@
 # PdfCraft roadmap: milestones and what's next
 
+> **Last reviewed:** 2026-11-03 | **Last updated:** 2026-11-03 | **Change:** patch (M5 #5850 review: indirect typewriter intent and composer edits preserve outer whitespace; movement regression commits by outside click; estimates unchanged)
+
+> **Last reviewed:** 2026-11-03 | **Last updated:** 2026-11-03 | **Change:** patch (M5 #5850: Enter adds Fill & Sign line breaks; outside clicks commit, Escape cancels, and saved multiline appearances retain their text and height; estimates unchanged)
+
 > **2026-11-03, M7 / #5835 review:** active image crops retain secondary clicks over overlapping images; crop controls share measured layout with hit detection. Rotation, handle and page-edge regressions added; focused tests, scoped Clippy, formatting, layering and asset checks passed locally. Estimates unchanged.
 
 > **Last reviewed:** 2026-11-01 | **Last updated:** 2026-11-01 | **Change:** patch (M7 #5657: header/footer, watermark and background operations commit all selected pages together; later-page errors preserve marks, content, shared/inherited resources and pending edits; estimates unchanged)
@@ -219,7 +223,7 @@ total matches the parity estimate.
 | M4 | Engine, history, save, organize | 100–180 | 95% | 8–15 | Done:<br>- command registry (menus, shortcuts and palette all use it);<br>- undo/redo; incremental, atomic and encrypted saves;<br>- autosave and crash recovery;<br>- organize, combine, extract, split and insert-from-file, with identical fonts and images stored once;<br>- CLI `edit/combine/extract/split`.<br>Done since: Organize opens directly into the page overview, with checkbox selection, an editable page-range selection (#1092), and per-page rotate/delete controls; Combine files in its own tab (sortable, resizable columns; folders; early warnings; unlocking protected files), bookmark editing, page labels (Number pages), CLI `run`, Set Page Boxes, Crop tool, Duplicate pages, Ctrl+A / Cmd+A to select all pages in Organize, Replace pages, bookmarks from tagged headings. Extract and Split preserve shared Type 3 glyph resource graphs, retaining conservative pruning limits. Missing: page transitions, recovery on the web M4: Insert-from-file preserves source bookmarks; integration tests retained with current main. |
 | M3 | Viewer app (native + web) | 80–150 | 90% | 10–20 | Acrobat-style shell, find, select, panels, tiles, web build, UI control channel for agents (opt-in). Missing: 60 fps test on a 500-page document, snapshot tests of every panel<br>Done since: fit visible, document title in the window; Linux middle-button scrolling; shared three-state theme preferences in the toolbar, View menu and control channel; numeric Linux portal events and idle wakeups completing #881, with native appearance inheritance retained (#446); late browser startup downloads preserve the current document and Save target (#171). Renderer-creation errors return to the existing OpenGL startup fallback; a real-GPU regression covers errors and device reuse. The underlying x86 DX12 shader compiler failure remains unresolved; see `windows-renderer-startup.md`. With Windows' Animation effects turned off, the wheel scrolls and zooms without easing and egui's animations take no time (`motion.rs`; macOS, Linux and the browser aren't asked yet); Preferences ▸ Animation effects (Use system setting, On, Off) decides it on every platform. |
 | M4 | Engine, history, save, organize | 100–180 | 95% | 8–15 | Done:<br>- command registry (menus, shortcuts and palette all use it);<br>- undo/redo; incremental, atomic and encrypted saves;<br>- autosave and crash recovery;<br>- organize, combine, extract, split and insert-from-file, with identical fonts and images stored once;<br>- CLI `edit/combine/extract/split`.<br>Done since: Combine files in its own tab (sortable, resizable columns; folders; early warnings; unlocking protected files), bookmark editing, page labels (Number pages), CLI `run`, Set Page Boxes, Crop tool, Duplicate pages, Ctrl+A / Cmd+A to select all pages in Organize, Replace pages, bookmarks from tagged headings. Missing: page transitions, recovery on the web |
-| M5 | Comments (all annotation types, XFDF) | 120–200 | 88% | 12–25 | Done: notes, highlight/underline/strikeout/squiggly, text boxes, ink, lines, arrows, rectangles, ovals, stamps (dynamic, Sign Here, Standard Business), with appearance streams; replies, status, checkmarks, locking, move/resize/restyle/delete; properties dialog; panel filter and sort; hide all; summaries (comments only); XFDF/FDF import and export (partial imports preserve existing reply and review-status links); flatten; Fill & Sign with adjustable text/date size; agent tools.<br>Done since: replies thread by their `/IRT` reference rather than the parent's `/NM`, so files that repeat names per page (PyMuPDF) thread correctly, and the Comments header counts comments without their replies, as Acrobat does, with the reply count under it; the Comments panel groups replies and counts all page roots once per frame (#2669, item 5); each reply in the Comments panel has its own "…" (Edit, Copy text, Delete): Delete takes that reply and the replies to it, the comment and its other replies stay, and the card stays selected; stamps placed on a rotated page read upright (#1398); polygons, connected lines, clouds, callouts, inserted text, colour and checkmark filters; image signature pointer previews, selection handles and live image movement/resizing with proportional corners and independent edge handles; Ctrl+Enter submission for highlight comment text and replies, with Shift+Enter newlines (reconciled with upstream main for PR #1457 on 2026-10-14). Missing: custom stamps, replace-text proposals, summary layouts with the page |
+| M5 | Comments (all annotation types, XFDF) | 120–200 | 88% | 12–25 | Done: notes, highlight/underline/strikeout/squiggly, text boxes, ink, lines, arrows, rectangles, ovals, stamps (dynamic, Sign Here, Standard Business), with appearance streams; replies, status, checkmarks, locking, move/resize/restyle/delete; properties dialog; panel filter and sort; hide all; summaries (comments only); XFDF/FDF import and export (partial imports preserve existing reply and review-status links); flatten; Fill & Sign with adjustable text/date size; Enter adds line breaks in a growing typewriter editor, outside clicks commit, Escape cancels, and saved/headless appearances keep multiline text (#5850); indirect typewriter intent resolves before whitespace decisions and Fill & Sign classification, with comment listing/property lookup retaining direct Subtype, and composer edits retain typewriter text as written while other comments retain trimming; agent tools.<br>Done since: replies thread by their `/IRT` reference rather than the parent's `/NM`, so files that repeat names per page (PyMuPDF) thread correctly, and the Comments header counts comments without their replies, as Acrobat does, with the reply count under it; the Comments panel groups replies and counts all page roots once per frame (#2669, item 5); each reply in the Comments panel has its own "…" (Edit, Copy text, Delete): Delete takes that reply and the replies to it, the comment and its other replies stay, and the card stays selected; stamps placed on a rotated page read upright (#1398); polygons, connected lines, clouds, callouts, inserted text, colour and checkmark filters; image signature pointer previews, selection handles and live image movement/resizing with proportional corners and independent edge handles; Ctrl+Enter submission for highlight comment text and replies, with Shift+Enter newlines (reconciled with upstream main for PR #1457 on 2026-10-14). Missing: custom stamps, replace-text proposals, summary layouts with the page |
 | M4 | Engine, history, save, organize | 100–180 | 95% | 8–15 | Done:<br>- command registry (menus, shortcuts and palette all use it);<br>- undo/redo; incremental, atomic and encrypted saves;<br>- autosave and crash recovery;<br>- organize, combine, extract, split and insert-from-file, with identical fonts and images stored once;<br>- CLI `edit/combine/extract/split`.<br>Done since: Combine files in its own tab (sortable, resizable columns; folders; early warnings; unlocking protected files), bookmark editing, page labels (Number pages), CLI `run`, Set Page Boxes, Crop tool, Duplicate pages, Ctrl+A / Cmd+A to select all pages in Organize, Replace pages, bookmarks from tagged headings. Missing: page transitions, recovery on the web M4: Insert-from-file preserves source bookmarks; integration tests retained with current main. |
 | M5 | Comments (all annotation types, XFDF) | 120–200 | 88% | 12–25 | Done: notes, highlight/underline/strikeout/squiggly, text boxes, ink, lines, arrows, rectangles, ovals, stamps (dynamic, Sign Here, Standard Business), with appearance streams; replies, status, checkmarks, locking, move/resize/restyle/delete; properties dialog; panel filter and sort; hide all; summaries (comments only); XFDF/FDF import and export (partial imports preserve existing reply and review-status links); flatten; Fill & Sign with adjustable text/date size; agent tools.<br>Done since: stamps placed on a rotated page read upright (#1398); polygons, connected lines, clouds, callouts, inserted text, colour and checkmark filters; image signature pointer previews, selection handles and live image movement/resizing with proportional corners and independent edge handles. Missing: custom stamps, replace-text proposals, summary layouts with the page |
 | M5 | Comments (all annotation types, XFDF) | 120–200 | 88% | 12–25 | Done: notes, highlight/underline/strikeout/squiggly, text boxes, ink, lines, arrows, rectangles, ovals, stamps (dynamic, Sign Here, Standard Business), with appearance streams; replies, status, checkmarks, locking, move/resize/restyle/delete; properties dialog; panel filter and sort; hide all; summaries (comments only); XFDF/FDF import and export (partial imports preserve existing reply and review-status links; nested FDF export uses lazy array owners and annotation positions, with retained entries budgeted before cloning under copied-graph limits and source `/Page` and `/Pages` dictionaries replaced by null at every copy depth); flatten; Fill & Sign with adjustable text/date size; agent tools.<br>Done since: polygons, connected lines, clouds, callouts, inserted text, colour and checkmark filters; image signature pointer previews, selection handles and live image movement/resizing with proportional corners and independent edge handles. A refused restyle or resize of a text box another program drew leaves that comment unchanged (#1327). Make Current Properties Default on an arrow updates arrows, not plain lines (#1302). Flattening keeps a visible comment it cannot draw (#1317). Missing: custom stamps, replace-text proposals, summary layouts with the page |
@@ -300,6 +304,8 @@ Estimates unchanged.
 
 | Date | Change | Summary |
 |---|---|---|
+| 2026-11-03 | patch | M5 #5850 review: indirect typewriter intent and composer edits preserve outer whitespace; movement regression commits by outside click; estimates unchanged. |
+| 2026-11-03 | patch | M5 #5850: Enter adds Fill & Sign line breaks; outside clicks commit, Escape cancels, and saved multiline appearances retain their text and height; estimates unchanged. |
 | 2026-11-02 | patch | M10 #5743: macOS print jobs carry the job name and owner some vendors' filters need; not yet checked on a physical printer |
 | 2026-11-01 | patch | M7 #5657: header/footer, watermark and background operations commit all selected pages together; later-page errors preserve marks, content, shared/inherited resources and pending edits; estimates unchanged. |
 | 2026-10-31 | patch | #1577: reconcile optional page shadows with current viewer/organizer and settings modules; CI validation pending. |

parity/acrobat-features.toml

@@ -794,8 +794,8 @@ milestone = "M2"
 status = "partial"
 commands = []
 tools = []
-evidence = ["crates/render/src/raster/tests.rs::huge_cid_width_ranges_terminate", "crates/render/src/raster/tests.rs::non_embedded_japanese_cid_fonts_draw_with_a_craft_fonts_face", "crates/render/src/raster/tests.rs::mincho_substitute_covers_half_width_katakana", "crates/render/src/raster/tests.rs::non_embedded_chinese_and_korean_cid_fonts_draw_with_an_installed_face", "crates/engine/src/tests.rs::installed_cjk_fonts_draw_in_the_viewer_but_not_in_an_export", "crates/render/src/text/tests.rs::identity_to_unicode_names_give_text", "crates/fonts/src/pdf/tests.rs::identity_to_unicode_names_read_codes_as_unicode"]
-notes = "CID-to-Unicode tables are M2.4; a non-embedded Adobe-Japan1 font is drawn in a craft-fonts Japanese face (Mincho or Gothic by name, else by the descendant font descriptor's Serif flag; bold Gothic for a Bold name or a `/FontWeight` of 600 or more; BIZ UDMincho for Mincho, which covers half-width katakana) when built with craft-fonts; the desktop viewer can fall back to an installed CJK face for GB1, CNS1, Korea1 and Japan1, including GBK-EUC-H TrueType CID fonts without ToUnicode. Windows preserves installed SimSun, SimHei, KaiTi and FangSong families, including subset-prefixed Founder aliases. Combine thumbnails and accessibility figure previews use the same on-screen fallback. System faces are display-only, disabled by PDFCRAFT_SYSTEM_FONTS=0, and not used by export, printing, OCR or automation; machines without a candidate and web builds remain unsupported. A `/ToUnicode` given as the name `/Identity-H` or `/Identity-V` (WSCAD) reads codes as Unicode for selection, search and editing, without control characters"
+evidence = ["crates/render/src/raster/tests.rs::huge_cid_width_ranges_terminate", "crates/render/src/raster/tests.rs::non_embedded_japanese_cid_fonts_draw_with_a_craft_fonts_face", "crates/render/src/raster/tests.rs::mincho_substitute_covers_half_width_katakana", "crates/render/src/raster/tests.rs::non_embedded_chinese_and_korean_cid_fonts_draw_with_an_installed_face", "crates/engine/src/tests.rs::installed_cjk_fonts_draw_in_the_viewer_but_not_in_an_export", "crates/render/src/text/tests.rs::identity_to_unicode_names_give_text", "crates/fonts/src/pdf/tests.rs::identity_to_unicode_names_read_codes_as_unicode", "crates/render/src/raster/tests.rs::a_soft_hyphen_a_substitute_lacks_draws_its_hyphen"]
+notes = "CID-to-Unicode tables are M2.4; a non-embedded Adobe-Japan1 font is drawn in a craft-fonts Japanese face (Mincho or Gothic by name, else by the descendant font descriptor's Serif flag; bold Gothic for a Bold name or a `/FontWeight` of 600 or more; BIZ UDMincho for Mincho, which covers half-width katakana) when built with craft-fonts; the desktop viewer can fall back to an installed CJK face for GB1, CNS1, Korea1 and Japan1, including GBK-EUC-H TrueType CID fonts without ToUnicode. Windows preserves installed SimSun, SimHei, KaiTi and FangSong families, including subset-prefixed Founder aliases. Combine thumbnails and accessibility figure previews use the same on-screen fallback. System faces are display-only, disabled by PDFCRAFT_SYSTEM_FONTS=0, and not used by export, printing, OCR or automation; machines without a candidate and web builds remain unsupported. A `/ToUnicode` given as the name `/Identity-H` or `/Identity-V` (WSCAD) reads codes as Unicode for selection, search and editing, without control characters. A non-embedded Identity font whose `/ToUnicode` gives a soft hyphen (U+00AD, as Windows exporters write Arial's hyphen) draws the standard substitute's hyphen (#5888). Any other character the substitute lacks still falls back to the CID taken as one of the substitute's glyph indexes, an unrelated glyph (#3529, #5726)"
 
 [[feature]]
 id = "view.render-image-codecs"
@@ -2607,11 +2607,11 @@ title = "Bates numbering across multiple files"
 area = "C"
 tier = "P1"
 milestone = "M7"
-status = "planned"
+status = "partial"
 commands = []
-tools = []
-evidence = []
-notes = ""
+tools = ["doc_bates_multi"]
+evidence = ["crates/automation/tests/it/automation.rs::bates_numbers_continue_across_files_and_report_preflight_and_permission_failures"]
+notes = "The headless batch tool allocates inclusive intervals in input order and reports per-file failures; it refuses existing output paths. Missing: a GUI batch dialog (Acrobat sets this up from Header & Footer ▸ Add Bates numbering with a file list), so it is reachable only through `doc_bates_multi`."
 
 [[feature]]
 id = "edit.bates-remove"
@@ -5399,8 +5399,8 @@ milestone = "M5"
 status = "shipped"
 commands = ["sign.fill.text"]
 tools = ["fill_sign_add"]
-evidence = ["crates/ui-egui/tests/it/fill_sign.rs::text_marks_and_date", "crates/annot/src/tests.rs::fill_and_sign_items_are_drawn", "crates/automation/tests/it/automation.rs::fill_and_sign_through_tools", "crates/engine/src/tests.rs::typed_text_reads_across_as_displayed_on_rotated_pages", "crates/ui-egui/tests/it/fill_sign.rs::fill_text_colour_survives_preview_commit_settings_and_save", "crates/automation/tests/it/automation.rs::fill_sign_text_and_date_colours_are_saved_and_validated"]
-notes = "typewriter FreeText, 10 pt Helvetica. Reads across as displayed on pages with /Rotate (sideways-stored scans): the box follows the displayed axes and the appearance is turned back (PdfCraft key /PCTextRotate); other apps that regenerate the appearance from /DA draw it unturned. A later page rotation does not re-turn it. Field detection does not work on scanned (image-only) forms yet; typed fractional font sizes commit once; Fill & Sign text/date colour is remembered and preserved in preview and saved appearances (#2444)"
+evidence = ["crates/annot/src/tests.rs::indirect_typewriter_names_keep_outer_whitespace_in_summaries", "crates/ui-egui/tests/it/comments.rs::typewriter_edit_composer_keeps_trailing_blank_lines", "crates/ui-egui/tests/it/comments.rs::other_edit_composers_keep_trimming_trailing_whitespace", "crates/ui-egui/tests/it/comments.rs::moved_fill_and_sign_text_is_drawn_where_it_goes_before_release", "crates/engine/src/tests.rs::fill_sign_multiline_text_survives_engine_save_reopen_and_render", "crates/ui-egui/tests/it/fill_sign.rs::fill_sign_enter_adds_lines_outside_click_commits_and_escape_cancels", "crates/automation/tests/it/automation.rs::multiline_fill_sign_engine_round_trip_preserves_lines_and_appearance", "crates/ui-egui/tests/it/fill_sign.rs::text_marks_and_date", "crates/annot/src/tests.rs::fill_and_sign_items_are_drawn", "crates/automation/tests/it/automation.rs::fill_and_sign_through_tools", "crates/engine/src/tests.rs::typed_text_reads_across_as_displayed_on_rotated_pages", "crates/ui-egui/tests/it/fill_sign.rs::fill_text_colour_survives_preview_commit_settings_and_save", "crates/automation/tests/it/automation.rs::fill_sign_text_and_date_colours_are_saved_and_validated"]
+notes = "Indirect IT is resolved when recognizing typewriter contents and Fill & Sign marks; comment listing and properties retain direct Subtype as before; composer edits preserve the typewriter text as written, while other comments retain their existing trimming. Enter inserts a newline; the editor grows with its lines, outside clicks commit and Escape cancels. Nonempty text preserves leading/internal/trailing line breaks, including after save/reopen; fill_sign_add sizes all wrapped rows for the saved appearance (#5850). typewriter FreeText, 10 pt Helvetica. Reads across as displayed on pages with /Rotate (sideways-stored scans): the box follows the displayed axes and the appearance is turned back (PdfCraft key /PCTextRotate); other apps that regenerate the appearance from /DA draw it unturned. A later page rotation does not re-turn it. Field detection does not work on scanned (image-only) forms yet; typed fractional font sizes commit once; Fill & Sign text/date colour is remembered and preserved in preview and saved appearances (#2444)"
 
 [[feature]]
 id = "form.fill-sign-marks"

vendor/README.md

@@ -39,7 +39,7 @@ Other resource categories retain their existing parent-chain lookup.
 
 | Crate | Version | Licence | Patches | Test |
 |---|---|---|---|---|
-| hayro-interpret | 0.7.0 | Apache-2.0 OR MIT | (1) select `/AP /N` appearance states via `/AS` (checkboxes, radios, stateful icons) and honour the NoView flag; (2) Type3 Unicode fallback to `/Encoding` glyph names, then printable ASCII; (3) `Type3Glyph::advance_width`; (4) `InterpreterSettings::ocg_overrides` for viewer layer toggles; (5) soft masks whose group lacks `/CS` were dropped (Chrome gradient text); (6) paint nesting cap `MAX_PAINT_NESTING` for tiling patterns and Type 3 glyphs (self-referencing content overflowed the stack); (7) CID width ranges in `/W` and `/W2` clamped to `MAX_CID` (a `0 4294967295 w` range hung); (8) `InterpreterSettings::hide_comments` skips markup annotations but keeps widgets and links (View ▸ Hide all comments); (9) `image_size_ok`: images and masks over 2^28 pixels are not decoded (a fuzzed `/W 4294967295` stencil mask allocated 4 GiB); (10) `MAX_NESTED_PAINTS`: nested paints (forms, Type 3 glyphs and tiling patterns inside one another) share a per-page budget of 50,000 (a glyph showing eight glyphs of its own font, sixteen deep, hung); (11) depends on skrifa 0.47 instead of 0.42 (upstream read-fonts fixed a Type 1 tokenizer that recursed through runs of numbers: a fuzzed font hung or overflowed the stack), with the one API change (`Charset::glyph_id` returns `Option`); (12) annotation appearances whose `/BBox` or `/Matrix` collapses to a line or a point are skipped (mapping them onto `/Rect` divided by zero and handed the device infinite or NaN matrices); (13) a Highlight annotation without a usable normal appearance is drawn from its `/QuadPoints`, `/C` and `/CA` with Multiply blending, as pdfcraft-annot draws highlights (it was not drawn at all); (14) simple-font `/Widths` (`read_widths`): entries only for codes 0–255 (one entry per code below `/FirstChar` was allocated: `4294967295` asked for 34 GB), an entry that isn't a number leaves its code missing instead of ending the array, and a `/LastChar` below `/FirstChar` gives no widths instead of discarding the font for a standard one; (15) the appearance box is mapped onto `/Rect` corner to corner: the translation now accounts for the scale (an appearance whose box didn't start at the origin and differed in size from `/Rect` was drawn shifted); (16) a clip set by `W`/`W*` takes effect after whichever path-painting operator ends the path (`S`, `f`, `B`, … as well as `n`; ISO 32000-2 §8.5.4): it was applied only on `n`, so `re W* S` clipped nothing and the clip stayed pending for a later, unrelated `n`; (17) `InterpreterSettings::cancelled`: once set, interpretation stops before the next content operator (a render pool sets it for every worker when dropped, after an edit or on closing, and for one worker when the watchdog gives up on its render, so renders whose answers nobody can receive stop instead of running to the end); (18) `MAX_PAGE_CONTENT`: the content a page holds at once (its contents, and the forms, tiling patterns and Type 3 glyph procedures being painted) stays within 512 MiB (`decode_content`; once it runs out, `InterpreterWarning::ContentTruncated` is reported (once per page by pdfcraft-render) so a truncated page is not a clean render; each level gives its bytes back when it finishes): forms nest fifty deep and each level kept its content while the next ran, so a form painting itself around a stream inflating to 256 MiB would have held over 12 GB; and Type 0 sampled functions take at most `MAX_SAMPLE_ENTRIES` (2^21) table points and `MAX_SAMPLE_VALUES` (2^23) samples, read no further than `/Size` asks (256 MiB of 1-bit samples asked for over 8 GB); (19) mesh shadings (types 4–7) are sampled only inside the area being drawn (`ShadingPattern::encode_within`, which the vendored hayro renderer calls with that area plus a pixel), within 8 visits per pixel of it plus 4 per triangle (`TriangleSampler`), and patches are triangulated one at a time: points far off the page were sampled over their whole bounding box, up to 65536 × 65536 entries (a fuzzed 147-byte patch stream; a synthetic page passed 6 GB in six seconds), ten thousand overlapping triangles ran for minutes, and every patch became 722 triangles before any limit; meshes hold at most `MAX_MESH_PATCHES` (2^16) patches and `MAX_MESH_TRIANGLES` (2^20) triangles, a lattice needs two to 2^19 vertices per row (`/VerticesPerRow 0` pushed empty rows for ever; one huge row read the whole stream), and a page samples at most `MAX_PAGE_MESH_TRIANGLES` (2^25) mesh triangles in all (a mesh painted a thousand times was triangulated a thousand times); each cut logs a warning; (20) PDF function construction is limited to 64 levels and 10,000 total nodes with cycle detection; calculator parsing and evaluation share procedure depth/work budgets, integer division/bitshift/rotation/index are panic-free, and fixed-size stack errors propagate; (21) a Type 0 font drawn with a substitute that isn't CID-keyed (it isn't embedded, or its program doesn't load) draws `.nLine truncated
+| hayro-interpret | 0.7.0 | Apache-2.0 OR MIT | (1) select `/AP /N` appearance states via `/AS` (checkboxes, radios, stateful icons) and honour the NoView flag; (2) Type3 Unicode fallback to `/Encoding` glyph names, then printable ASCII; (3) `Type3Glyph::advance_width`; (4) `InterpreterSettings::ocg_overrides` for viewer layer toggles; (5) soft masks whose group lacks `/CS` were dropped (Chrome gradient text); (6) paint nesting cap `MAX_PAINT_NESTING` for tiling patterns and Type 3 glyphs (self-referencing content overflowed the stack); (7) CID width ranges in `/W` and `/W2` clamped to `MAX_CID` (a `0 4294967295 w` range hung); (8) `InterpreterSettings::hide_comments` skips markup annotations but keeps widgets and links (View ▸ Hide all comments); (9) `image_size_ok`: images and masks over 2^28 pixels are not decoded (a fuzzed `/W 4294967295` stencil mask allocated 4 GiB); (10) `MAX_NESTED_PAINTS`: nested paints (forms, Type 3 glyphs and tiling patterns inside one another) share a per-page budget of 50,000 (a glyph showing eight glyphs of its own font, sixteen deep, hung); (11) depends on skrifa 0.47 instead of 0.42 (upstream read-fonts fixed a Type 1 tokenizer that recursed through runs of numbers: a fuzzed font hung or overflowed the stack), with the one API change (`Charset::glyph_id` returns `Option`); (12) annotation appearances whose `/BBox` or `/Matrix` collapses to a line or a point are skipped (mapping them onto `/Rect` divided by zero and handed the device infinite or NaN matrices); (13) a Highlight annotation without a usable normal appearance is drawn from its `/QuadPoints`, `/C` and `/CA` with Multiply blending, as pdfcraft-annot draws highlights (it was not drawn at all); (14) simple-font `/Widths` (`read_widths`): entries only for codes 0–255 (one entry per code below `/FirstChar` was allocated: `4294967295` asked for 34 GB), an entry that isn't a number leaves its code missing instead of ending the array, and a `/LastChar` below `/FirstChar` gives no widths instead of discarding the font for a standard one; (15) the appearance box is mapped onto `/Rect` corner to corner: the translation now accounts for the scale (an appearance whose box didn't start at the origin and differed in size from `/Rect` was drawn shifted); (16) a clip set by `W`/`W*` takes effect after whichever path-painting operator ends the path (`S`, `f`, `B`, … as well as `n`; ISO 32000-2 §8.5.4): it was applied only on `n`, so `re W* S` clipped nothing and the clip stayed pending for a later, unrelated `n`; (17) `InterpreterSettings::cancelled`: once set, interpretation stops before the next content operator (a render pool sets it for every worker when dropped, after an edit or on closing, and for one worker when the watchdog gives up on its render, so renders whose answers nobody can receive stop instead of running to the end); (18) `MAX_PAGE_CONTENT`: the content a page holds at once (its contents, and the forms, tiling patterns and Type 3 glyph procedures being painted) stays within 512 MiB (`decode_content`; once it runs out, `InterpreterWarning::ContentTruncated` is reported (once per page by pdfcraft-render) so a truncated page is not a clean render; each level gives its bytes back when it finishes): forms nest fifty deep and each level kept its content while the next ran, so a form painting itself around a stream inflating to 256 MiB would have held over 12 GB; and Type 0 sampled functions take at most `MAX_SAMPLE_ENTRIES` (2^21) table points and `MAX_SAMPLE_VALUES` (2^23) samples, read no further than `/Size` asks (256 MiB of 1-bit samples asked for over 8 GB); (19) mesh shadings (types 4–7) are sampled only inside the area being drawn (`ShadingPattern::encode_within`, which the vendored hayro renderer calls with that area plus a pixel), within 8 visits per pixel of it plus 4 per triangle (`TriangleSampler`), and patches are triangulated one at a time: points far off the page were sampled over their whole bounding box, up to 65536 × 65536 entries (a fuzzed 147-byte patch stream; a synthetic page passed 6 GB in six seconds), ten thousand overlapping triangles ran for minutes, and every patch became 722 triangles before any limit; meshes hold at most `MAX_MESH_PATCHES` (2^16) patches and `MAX_MESH_TRIANGLES` (2^20) triangles, a lattice needs two to 2^19 vertices per row (`/VerticesPerRow 0` pushed empty rows for ever; one huge row read the whole stream), and a page samples at most `MAX_PAGE_MESH_TRIANGLES` (2^25) mesh triangles in all (a mesh painted a thousand times was triangulated a thousand times); each cut logs a warning; (20) PDF function construction is limited to 64 levels and 10,000 total nodes with cycle detection; calculator parsing and evaluation share procedure depth/work budgets, integer division/bitshift/rotation/index are panic-free, and fixed-size stack errors propagate; (21) a Type 0 font drawn with a substitute that isn't CID-keyed (it isn't embedded, or its program doesn't load) draws `.nLine truncated
 | hayro | 0.7.1 | Apache-2.0 OR MIT | (1) `RenderSettings::{x_offset, y_offset}` to render one tile of a large page; (2) images over `MAX_IMAGE_PIXELS` (2^28) are skipped (a fuzzed `/W 4294967295` hung in resampling); (3) `tiling_cell_scale` keeps a tiling pattern's cell pixmap within 3000 px a side whatever its `/XStep`/`/YStep` (a fuzzed step far beyond `/BBox` allocated 2 GB); (4) stroke widths clamped to 4 × (canvas width + height) device pixels (a fuzzed `/LW 9223372036854775807` allocated 10 GB in stroke expansion); (5) `usable_dash_pattern`: a stroke whose dash pattern could cut it into more than `MAX_DASHES_PER_STROKE` (250,000) pieces, counting kurbo's restart of the pattern at every subpath and `ClosePath`, is drawn solid (a fuzzed `/D [[11] 0]` on a line from x = 92234775807 allocated over 5 GB, `[0.000001] 0 d` on a 20 pt line over 10 GB); (6) a dash pattern with a negative or non-finite entry, or a period that isn't positive, is drawn solid (`[-1 -1] 0 d` looped for ever in kurbo); (7) `render_size` and `render_into`: render into a buffer the caller allocates (PdfCraft renders into pixel-aligned memory that the GUI takes over as texture data); `render` is `render_into` on a new pixmap; (8) checked image source/target/RGBA, alpha-premultiplication scratch and intermediate sizes before resampling, finite transforms and exact source/mask buffers; sources wider than u16 may shrink before backend limits, with a 2^20 source-axis cap and conservative 128 MiB filter-planning estimate; an over-limit or failed resize keeps the original image resolution and geometry when the source fits the backend, otherwise skips it, with a logged warning; staging, resampling output/scratch, RGBA conversion and padding buffers use fallible reserves; (9) `may_paint_canvas`: a fill or stroke whose control-point box, grown by how far its paint reaches past the path (half the stroke width times the larger of the miter limit and √2) plus 0.5 user units and 2 pixels, misses the canvas is not drawn, so a tile no longer strokes and fills every path of its page (item 5 of #623); `set_skip_offscreen_paths` turns this off for the regression test; (10) `/Interpolate` governs the pre-resize too (#624): a minified image that does not request interpolation is pre-resized with nearest-neighbour, planned without the Catmull-Rom filter limits (`image_resampling_size` takes the flag); a Type 3 glyph image is sampled bicubically only when interpolated; a minified image with a matching soft mask is nearest-sampled straight into premultiplied RGBA; (11) `may_paint_rect`: an image whose stream-size box, mapped to the device, misses the canvas is skipped before decoding (decoded data can only shrink against that box), so a tile of an image-heavy page no longer decodes, resamples and rasterizes every image on it; (12) preview renders (`InterpreterSettings::preview`: backdrops, thumbnails, grid cells, signature-drag backgrounds) skip images under `PREVIEW_MIN_IMAGE_AREA` device pixels, while final renders always draw them; (13) reading colours (#897): `RenderSettings::color_replacement` takes a `ColorReplacement` (text colour, background, `only_black`, `line_art`) that remaps solid text and line-art paint while rendering, glyph outlines counting as text and stencil-mask ink as line art; images, image alpha masks, soft masks and pattern cells keep their colours; `render_pdf` gains a `color_replacement: Option<ColorReplacement>` argument (pass `None` for true colours); the upstream-only `[dev-dependencies.log]` entry is removed from `Cargo.toml` (log is a normal dependency since the logging patch), so `cargo test -p hayro` runs this crate's unit tests; (14) a shrunk, interpolated image whose axes lie along the device's (upright, mirrored or a quarter turn, as on a page under `/Rotate`) is snapped to whole device pixels: shrunk once to its rounded device rectangle and drawn 1:1 from the rounded corner, instead of a ceil-sized raster resampled again at a fractional scale and offset (`snap_minified_image`; scans looked soft next to Acrobat, and until #5530 a turned one still did, up to 200 grey levels off the same page drawn upright). An image counts as axis-aligned when the terms the snap drops move its far edges by at most a hundredth of a device pixel (`SNAP_SKEW`), not by a tolerance on the coefficients, which let a slight skew of a very wide image through; and the snapped 1:1 transform is used only when the shrink produced the snapped size, so a failed resize keeps the full transform instead of drawing the source raster at its source size; (15) a soft mask of another size than its image is filtered when shrunk, whatever its `/Interpolate` (#1649): such a mask carries the shape on its own (Word saves text as a tiny black image with a glyph-bitmap mask), and nearest-neighbour dropped every stroke thinner than the shrink factor; (16) under a soft mask or a blend mode, axial and radial shadings are sampled instead of drawn as native gradients (#Line truncated
 | hayro-jbig2 | 0.3.0 | Apache-2.0 OR MIT | (1) at most 2^28 pixels per bitmap and page bitmap, and 2^25 cells per halftone grid and grayscale image (each side could be 65535: 4.3 billion pixels decoded one by one, and a 4-byte value per cell); (2) at most 2^26 symbol pixels per symbol dictionary (a fuzzed dictionary decoded for minutes); (3) a symbol dictionary's height classes are capped at one per new symbol and its export runs at `2 × symbols + 1`, each plus `EXTRA_EMPTY_STEPS` (1024): an empty height class or a zero-length export run doesn't advance, and past the end of its data the arithmetic decoder can decode them for ever (one changed byte hung the renderer); (4) a segment header's long-form referred-to segment count is capped at `u16::MAX` and a larger count is refused before its retention or referred-to bytes are read: the 29-bit count went straight to `Vec::with_capacity`, a reservation of about 2 GiB whose failure aborts the process (from storytold/craft-libs#176); (5) what a page's segments hold for later ones (intermediate regions, pattern and symbol dictionaries with their headers and retained contexts, and Huffman tables) is at most 64 MiB, two square bitmaps of 16,384 a side (`MAX_RETAINED_WORDS`), and a dictionary's new and exported symbols are counted against it as they are made: each was bounded on its own, but a few dozen bytes an intermediate region held 33 MB each, and 8 MiB symbols exported again and again killed the process, so a few kilobytes could exhaust memory; (6) a segment may refer to at most 2^20 symbols, checked before any are collected: references may repeat, and 65,535 of them to one dictionary, in a 78 KB stream, asked for 17 GB; (7) a pattern dictionary keeps at most 16 MiB of shifted copies, which only speed up drawing, and none that wouldn't fit what the page has left (`MAX_SHIFTED_WORDS`); (8) a custom Huffman table has at most 2^16 lines, and a stream with its globals at most 2^20 segments and 2^22 references, checked as they are read: both grew before anything counted them, and PDFs of a kilobyte or so reached 5 GB and more behind nested Flate; (9) a page past the bounds of (1) is refused when the stream is opened: a caller sizes its output from the header before the page bitmap is made, and an Indexed image in a 743-byte PDF asked for 137 GB, which aborts; (10) a stream may ask for at most 2^30 units of decoding work, four of the largest pages, about 6 s to 17 s (`MAX_WORK`): decoding a bitmap costs a pixel each and 4 a row, copying one a word each and 1 a row, and each integer an item reads costs 38, what one takes at worst, so one with no pixels pays too, before it is decoded: each bitmap was bounded, but nothing bounded how many there were, and a 795-byte PDF took 265 s; (11) `Image::decode_within` spends from what a run of images may still do, starting with its page's pixels, and `charge_reading` charges reading a stream and its globals before anything is parsed, so an operation over a document can bound what `MAX_WORK` bounds for one image, and images that fail or decode nothing add up too: a file of thousands of images made to reach that limit took as many times as long; (12) adding to a location, a reference offset, a symbol's corner or a Huffman range line near the end of an `i32`, and negating a decoded integer of 2^31, overflowed: a panic wherever overflow is checked. The refinement context, a text region's corners, the integer decoder and the Huffman table wrap, as a release build did, and `Bitmap::combine` saturates, which comes to the same, as a location that overflows is past the end of any page. What is drawn doesn't change | (1) `raster::tests::jbig2_regions_of_billions_of_pixels_are_refused`, `raster::tests::jbig2_halftone_grids_past_their_bound_are_refused`; (2) the nightly `cargo xtask fuzz` finding, which now finishes in about a second; (3) `raster::tests::jbig2_symbol_dictionaries_that_never_advance_terminate`; (4) `segment::tests::test_referred_to_count_is_capped` (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`); (5) `raster::tests::jbig2_regions_held_past_a_pages_bound_are_refused`, `jbig2_symbols_without_pixels_count_towards_a_pages_bound`, `jbig2_retained_contexts_count_towards_a_pages_bound`, `jbig2_huffman_tables_count_towards_a_pages_bound`, `jbig2_new_symbols_count_towards_a_pages_bound`, `jbig2_new_symbols_are_held_to_what_a_page_has_left`, `jbig2_pattern_dictionaries_count_towards_a_pages_bound`, and in the crate (`cargo test --manifest-path vendor/hayro-jbig2/Cargo.toml`) `decode::symbol::tests::exported_symbols_are_counted_before_they_are_copied`, `tests::the_room_left_shrinks_as_segments_are_stored`; (6) `raster::tests::jbig2_segments_referring_to_too_many_symbols_are_refused`; (7) `raster::tests::jbig2_pattern_dictionaries_too_large_to_shift_still_draw`, `jbig2_shifted_patterns_never_refuse_an_image`; (8) `raster::tests::jbig2_huffman_tables_of_too_many_lines_are_refused`, `jbig2_streams_of_too_many_segments_are_reLine truncated
 | lopdf | 0.45.0 | MIT | (1) PNG predictor rows longer than the data are refused before their buffers are allocated (a fuzzed `/Columns 4294967295` allocated two 4 GiB rows), and empty frames return without row buffers; (2) predictor row widths use saturating arithmetic or reject overflow; (3) sub-byte TIFF Predictor 2 scratch is bounded by the samples present in the input instead of the declared `/Colors`, aligned with [lopdf#589](https://github.com/J-F-Liu/lopdf/pull/589); (4) the object lookup keeps its cycle history through compressed objects, requires an ordinary in-file container, and caps the chain of indirect lookups at 16 (a chain of new objects, each a `/Length` compressed in a container whose own `/Length` names the next, or a stream whose `/Length` names the next stream, overflowed the stack: a dev build's 2 MiB thread within 800 hops); (5) object-stream members are parsed within their own extent (up to the next distinct offset), an entry whose offset repeats an earlier entry's is skipped (so a repeated object number can resolve to its earlier entry), and offsets are processed in index order; (6) `find_latest_trailer`: a rebuild takes a `/Type /XRef` stream dictionary as the trailer when the file has no `trailer` keyword (the latest in the file wins, as a later `trailer` would), so a concatenated or truncated file keeps its /Info and /ID instead of failing to load; a candidate whose /Root is the catalog beats one whose /Root is another dictionary, and an in-file /Root must match its generation (#2365) | (1) `inspect::tests::png_predictor_rows_longer_than_the_data_are_refused_before_allocating`, `inspect::tests::empty_png_predictor_frames_do_not_allocate_rows`; (2) `inspect::tests::tiff_subbyte_predictor_row_width_overflow_is_refused`; (3) `inspect::tests::tiff_subbyte_predictor_scratch_is_bounded_by_available_samples`, `inspect::tests::tiff_subbyte_predictors_keep_components_rows_and_padding`; (4) `inspect::tests::compressed_object_self_container_does_not_recurse`, `inspect::tests::compressed_object_mutual_containers_do_not_recurse`, `inspect::tests::object_stream_length_cycle_does_not_recurse`, `inspect::tests::compressed_object_container_chain_does_not_recurse`, `inspect::tests::compressed_length_chain_does_not_recurse`, `inspect::tests::plain_stream_length_chain_does_not_recurse`, `inspect::tests::compressed_objects_and_indirect_stream_length_still_load`, `inspect::tests::compressed_length_in_a_container_with_an_indirect_length_still_loads`; (5) `inspect::tests::object_stream_member_extents_bound_owned_strings`, `inspect::tests::object_stream_member_extents_preserve_index_order_and_whitespace`; (6) `inspect::tests::rebuilt_file_keeps_the_info_of_its_cross_reference_stream`, `inspect::tests::rebuilt_file_prefers_a_trailer_whose_root_is_the_catalog` |

vendor/hayro-interpret/src/font/cid.rs

@@ -287,13 +287,14 @@ impl Type0Font {
             }
             FontType::Cff(c) => {
                 // Map codepoint to glyph name via AFL, and then look it up.
-                if let Some(name) = glyph_names::get_reverse(character)
-                    && let Some(gid) = c.glyph_index_by_name(name)
-                {
-                    Some(gid)
-                } else {
-                    None
-                }
+                let by_name = |character: char| {
+                    glyph_names::get_reverse(character).and_then(|name| c.glyph_index_by_name(name))
+                };
+                // PdfCraft patch: a soft hyphen the substitute has no glyph for is its hyphen
+                // (#5888). Windows exporters map the hyphen glyph of a font they don't embed to
+                // U+00AD, which ISO 32000-2 Annex D.2 calls typographically the same as the
+                // hyphen; without this the CID, taken as a glyph index, drew an unrelated glyph.
+                by_name(character).or_else(|| (character == '\u{00ad}').then(|| by_name('-')).flatten())
             }
             FontType::Type1(_) => {
                 // We could in theory map to a glyph via AFL and then look up