Fix #811: unnamed glyphs of an installed font stay in their type

In a PDF (or the PDF part of an .ai file), a glyph of a subset font that
no ToUnicode map, CID table or character map of the font program names
was drawn as outlines, so one text object came in split: "RADA" as the
type "ADA" plus the R's outline.

Such a glyph is now looked for in the installed face of the font's name:
the first character (of those single-byte fonts encode: ASCII, Latin-1,
Latin Extended-A, punctuation, the euro and trade mark signs, fi and fl)
whose glyph draws the same outline, point for point within 2/1000 em,
names it. A subset carries the font's own outlines, so an installed copy
of the same font finds every glyph. Other versions of the font, empty
glyphs and glyphs that are no character still keep their outlines, with
the warning as before. Results are cached per glyph.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

crates/pdf/src/import.rs

@@ -503,6 +503,46 @@ fn matching_face(
     best.map(|(_, f)| f)
 }
 
+/// The characters a glyph no mapping names is looked for among (#811): those single-byte fonts
+/// encode (ASCII, Latin-1, Latin Extended-A, general punctuation, the euro and trade mark signs,
+/// the fi and fl ligatures).
+const IDENTIFIABLE: [std::ops::RangeInclusive<u32>; 6] =
+    [0x21..=0x7E, 0xA1..=0x17F, 0x2010..=0x2044, 0x20AC..=0x20AC, 0x2122..=0x2122, 0xFB01..=0xFB02];
+
+/// The character whose glyph in installed `face` draws `embedded` (a glyph outline in thousandths
+/// of an em, y up): the first of [`IDENTIFIABLE`] with the same outline, point for point. A subset
+/// font embedded without a character map still carries the font's own outlines, so a glyph whose
+/// subset code names nothing is found this way. `None` when no glyph matches (another version of
+/// the font, or a glyph that is no character), and for empty glyphs, which every space draws.
+pub(crate) fn identify_glyph(face: &vectorcraft_text::FontFace, embedded: &BezPath) -> Option<char> {
+    if embedded.elements().is_empty() {
+        return None;
+    }
+    let k = 1000.0 / face.units_per_em();
+    // Installed outlines are y-down in font units.
+    let to_pdf = Affine::new([k, 0.0, 0.0, -k, 0.0, 0.0]);
+    let db = vectorcraft_text::FontDb::global();
+    IDENTIFIABLE.iter().cloned().flatten().filter_map(char::from_u32).find(|c| {
+        let gid = face.glyph_for(*c);
+        gid != 0 && same_shape(embedded, &(to_pdf * db.outline(face, gid).as_ref().clone()))
+    })
+}
+
+/// Do outlines `a` and `b` draw the same segments, every point within 2/1000 em (closing
+/// commands aside)?
+fn same_shape(a: &BezPath, b: &BezPath) -> bool {
+    let drawn = |p: &BezPath| p.elements().iter().filter(|e| !matches!(e, PathEl::ClosePath)).copied().collect::<Vec<_>>();
+    let (a, b) = (drawn(a), drawn(b));
+    let near = |p: kurbo::Point, q: kurbo::Point| (p - q).hypot() <= 2.0;
+    a.len() == b.len()
+        && a.iter().zip(&b).all(|(x, y)| match (x, y) {
+            (PathEl::MoveTo(p), PathEl::MoveTo(q)) | (PathEl::LineTo(p), PathEl::LineTo(q)) => near(*p, *q),
+            (PathEl::QuadTo(p1, p2), PathEl::QuadTo(q1, q2)) => near(*p1, *q1) && near(*p2, *q2),
+            (PathEl::CurveTo(p1, p2, p3), PathEl::CurveTo(q1, q2, q3)) => near(*p1, *q1) && near(*p2, *q2) && near(*p3, *q3),
+            _ => false,
+        })
+}
+
 /// The character of glyph `text` when its outline can be compared with an installed font's: a
 /// single character, and not CJK punctuation set vertically (its vertical form differs).
 fn comparable(text: &str, vertical: bool) -> Option<char> {
@@ -666,6 +706,9 @@ struct Builder<'p> {
     /// Font (cache key) → the installed face that draws its glyphs, decided from its first glyph
     /// that can be compared ([`matching_face`]); `None`: its glyphs differ, so it stays outlines.
     matched: HashMap<u128, Option<Arc<vectorcraft_text::FontFace>>>,
+    /// (Font, glyph) → the character its outline matches in the installed face, for glyphs no
+    /// mapping names ([`Builder::identified_char`]).
+    identified: HashMap<(u128, u32), Option<char>>,
     /// Per font: the installed version its type names, when it isn't the one its family and style
     /// resolve to (see `CharStyle::font_version`).
     versions: HashMap<u128, Option<String>>,
@@ -817,6 +860,7 @@ impl<'p> Builder<'p> {
             cid_text: HashMap::new(),
             font_chars: HashMap::new(),
             matched: HashMap::new(),
+            identified: HashMap::new(),
             versions: HashMap::new(),
             mask: None,
             masked: vec![],
@@ -1286,6 +1330,17 @@ impl<'p> Builder<'p> {
 
     /// The family and style of font `key` (drawing glyph `o`): the installed face of its PostScript
     /// name, else the available family its name reads as; whether it is available.
+    /// The character of glyph `glyph` of font `key` that no mapping names, found by its outline in
+    /// the installed face of the font's name ([`identify_glyph`]); cached per glyph.
+    fn identified_char(&mut self, key: u128, glyph: u32, o: &hayro_interpret::font::OutlineGlyph) -> Option<char> {
+        if let Some(c) = self.identified.get(&(key, glyph)) {
+            return *c;
+        }
+        let c = self.font_name(key, o).face.and_then(|f| identify_glyph(&f, &o.outline()));
+        self.identified.insert((key, glyph), c);
+        c
+    }
+
     fn font_name(&mut self, key: u128, o: &hayro_interpret::font::OutlineGlyph) -> FontInfo {
         if let Some(n) = self.font_names.get(&key) {
             return n.clone();
@@ -1345,7 +1400,9 @@ impl<'p> Builder<'p> {
                             .font_chars
                             .entry(key)
                             .or_insert_with(|| o.font_data().and_then(|d| font_chars(d.data.as_ref().as_ref())).map(Arc::new));
-                        (chars.as_ref().and_then(|m| m.get(&glyph)).map(|c| BfString::Char(*c)), false)
+                        let named = chars.as_ref().and_then(|m| m.get(&glyph)).copied();
+                        // Still unnamed: the installed font may draw the same outline (#811).
+                        (named.or_else(|| self.identified_char(key, glyph, o)).map(BfString::Char), false)
                     }
                 }
             }

crates/pdf/src/tests_realtext.rs

@@ -248,17 +248,33 @@ fn digits_unnamed(text: &str) -> Vec<u8> {
     out
 }
 
-/// #708: a glyph whose ToUnicode value is U+FFFD isn't type showing "�". This subset font has no
-/// character map of its own, so the digits keep their outlines, drawn as the file draws them, and
-/// a warning names the font.
+/// #708, #811: a glyph whose ToUnicode value is U+FFFD isn't type showing "�". This subset font
+/// has no character map of its own, but the font is installed: each digit is found by its outline
+/// there, so the whole text stays one type object, with no outlines left behind.
 #[test]
-fn glyphs_mapped_to_the_replacement_character_keep_their_outlines() {
+fn unnamed_glyphs_of_an_installed_font_are_found_by_their_outline() {
     let bytes = digits_unnamed("Tel 02-12345");
     let report = import_with_report(&bytes, &ImportOptions { text_as: TextAs::Text, ..Default::default() }).unwrap();
-    let all = texts(&report.document).concat();
-    assert!(!all.contains('\u{FFFD}') && all.contains("Tel"), "no replacement characters: {all:?}");
-    assert!(ink(&report.document).width() > 50.0, "the digits are outlines");
-    assert!(report.warnings.iter().any(|w| w.contains("U+FFFD")), "{:?}", report.warnings);
+    assert_eq!(texts(&report.document), ["Tel 02-12345"]);
+    let mut paths = 0;
+    report.document.walk(|n| paths += usize::from(matches!(n.kind, NodeKind::Path { .. } | NodeKind::Compound { .. })));
+    assert_eq!(paths, 0, "no outlines left: {:?}", report.warnings);
+    assert!(!report.warnings.iter().any(|w| w.contains("U+FFFD")), "{:?}", report.warnings);
+}
+
+/// #811: an outline is named only by a glyph that draws it exactly; one moved off its place, or an
+/// empty glyph (which every space draws), stays unnamed and keeps its outlines.
+#[test]
+fn glyphs_are_named_only_by_an_exact_outline() {
+    let db = FontDb::global();
+    let face = db.face("Source Sans 3", "Regular").unwrap();
+    let k = 1000.0 / face.units_per_em();
+    let pdf_space = |c: char| Affine::new([k, 0.0, 0.0, -k, 0.0, 0.0]) * db.outline(&face, face.glyph_for(c)).as_ref().clone();
+    for c in ['R', 'a', '7', '&', 'é'] {
+        assert_eq!(crate::import::identify_glyph(&face, &pdf_space(c)), Some(c));
+    }
+    assert_eq!(crate::import::identify_glyph(&face, &(Affine::translate((100.0, 0.0)) * pdf_space('R'))), None);
+    assert_eq!(crate::import::identify_glyph(&face, &BezPath::new()), None);
 }
 
 /// #708: where the embedded font program has a character map, a glyph's character comes from it.