From 8c3812499cc15ee0673159bf81c8a6c5ad64cd85 Mon Sep 17 00:00:00 2001
From: mtlcl410 <2368091+mtlcl410@users.noreply.github.com>
Date: Thu, 22 Oct 2026 18:26:21 +0800
Subject: [PATCH] M10: read a font's post names once for the 256 codes (#3962)

CodeRabbit on #3969 (nitpick): Post::glyph_name walks a version 2.0 table's name data
from its start for each glyph, and symbolic_glyph_texts asks for up to 256, so an
untrusted font program (up to 16 MiB) could make opening a document slow. The names the
table spells out are now read once (post_custom_names) and looked up by index
(post_name), giving what Post::glyph_name gives; a test checks that for every glyph of
Inter's version 2.0 table.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 crates/fonts/src/pdf.rs       | 30 ++++++++++++++++++++++++++++--
 crates/fonts/src/pdf/tests.rs | 16 ++++++++++++++++
 2 files changed, 44 insertions(+), 2 deletions(-)

diff --git a/crates/fonts/src/pdf.rs b/crates/fonts/src/pdf.rs
index a81f1b341e..5d42ca67d7 100644
--- a/crates/fonts/src/pdf.rs
+++ b/crates/fonts/src/pdf.rs
@@ -1260,6 +1260,7 @@ fn symbolic_glyph_texts(doc: &Document, font: &Dict) -> Vec<(u32, String)> {
         None => face.post(),
     };
     let Ok(post) = post else { return Vec::new() };
+    let custom = post_custom_names(&post);
     (0..=255u32)
         .filter_map(|code| {
             // The records in order, as the renderer's `map_code` reads a symbolic font: a Windows
@@ -1274,12 +1275,37 @@ fn symbolic_glyph_texts(doc: &Document, font: &Dict) -> Vec<(u32, String)> {
                 let subtable = record.subtable(cmap.offset_data()).ok()?;
                 offsets.iter().find_map(|o| subtable.map_codepoint(code + o).filter(|g| *g != GlyphId::NOTDEF))
             })?;
-            let glyph = skrifa::raw::types::GlyphId16::new(u16::try_from(glyph.to_u32()).ok()?);
-            Some((code, glyph_text(post.glyph_name(glyph)?)?))
+            let glyph = u16::try_from(glyph.to_u32()).ok()?;
+            Some((code, glyph_text(post_name(&post, &custom, glyph)?)?))
         })
         .collect()
 }
 
+/// The names a version 2.0 `post` table spells out, in order, read once. `Post::glyph_name` walks
+/// this data from its start for each glyph it is asked about, which over a font program of up to
+/// 16 MiB, asked about 256 codes, is work an untrusted file can make large. Reading stops at the
+/// first name that can't be read, as each lookup would.
+fn post_custom_names<'a>(post: &skrifa::raw::tables::post::Post<'a>) -> Vec<&'a str> {
+    post.string_data().map(|names| names.iter().map_while(|name| name.ok().map(|n| n.as_str())).collect()).unwrap_or_default()
+}
+
+/// `glyph`'s name in `post`, as `Post::glyph_name` gives it, with the table's own names from
+/// [`post_custom_names`].
+fn post_name<'a>(post: &skrifa::raw::tables::post::Post<'a>, custom: &[&'a str], glyph: u16) -> Option<&'a str> {
+    use skrifa::raw::{tables::post::DEFAULT_GLYPH_NAMES, types::Version16Dot16};
+    match post.version() {
+        Version16Dot16::VERSION_1_0 => DEFAULT_GLYPH_NAMES.get(usize::from(glyph)).copied(),
+        Version16Dot16::VERSION_2_0 => {
+            let index = usize::from(post.glyph_name_index()?.get(usize::from(glyph))?.get());
+            match index.checked_sub(DEFAULT_GLYPH_NAMES.len()) {
+                None => DEFAULT_GLYPH_NAMES.get(index).copied(),
+                Some(i) => custom.get(i).copied(),
+            }
+        }
+        _ => None,
+    }
+}
+
 /// Code → Unicode for a font: its ToUnicode CMap (for a composite font, also the name
 /// `/Identity-H` or `/Identity-V`, read as code = Unicode), else (simple fonts) its encoding.
 fn unicode_map(doc: &Document, font: &Dict, composite: bool) -> Unicode {
diff --git a/crates/fonts/src/pdf/tests.rs b/crates/fonts/src/pdf/tests.rs
index 83db09047a..d1cff9fa5c 100644
--- a/crates/fonts/src/pdf/tests.rs
+++ b/crates/fonts/src/pdf/tests.rs
@@ -190,6 +190,22 @@ fn mac_subset_cmap_uses_pdf_unicode_without_changing_source_program() {
     assert!(append_editor_cmap(&mut broken, &std::collections::BTreeMap::from([(65, 1)])).is_none());
 }
 
+/// The names read once give what `Post::glyph_name` gives, for every glyph of a font with its
+/// own names (Inter's version 2.0 table), and none past the last glyph.
+#[test]
+fn post_names_read_once_match_the_table() {
+    use skrifa::raw::TableProvider as _;
+    let face = skrifa::FontRef::new(include_bytes!("../../../../assets/fonts/Inter-Regular.ttf")).unwrap();
+    let post = face.post().unwrap();
+    let custom = post_custom_names(&post);
+    assert!(custom.len() > 100, "Inter names glyphs of its own: {}", custom.len());
+    let glyphs = face.maxp().unwrap().num_glyphs();
+    for g in 0..glyphs {
+        assert_eq!(post_name(&post, &custom, g), post.glyph_name(skrifa::raw::types::GlyphId16::new(g)), "glyph {g}");
+    }
+    assert_eq!(post_name(&post, &custom, glyphs), None);
+}
+
 /// #3962: a symbolic TrueType subset with no `/Encoding` whose codes are glyph positions in
 /// the Macintosh order, 29 below ASCII, mapped by a Mac (1,0) cmap. Its text is what the
 /// `post` table names the glyphs its codes select, and typing encodes back to those codes.