From 1cc392e5b04ada01b130236c6d101636e8bb62da Mon Sep 17 00:00:00 2001 From: aiongg Date: Thu, 22 Oct 2026 15:33:28 +0000 Subject: [PATCH] fix(fonts): cut ToUnicode destinations to the CMap limit A sequential bfrange stores one destination and builds every covered glyph's text from all of it, so one huge destination and many repeats of its codes made text extraction copy megabytes per glyph. Destinations keep at most 256 UTF-16 units (the format's 512 bytes), and marked-text extraction stops asking for a glyph's text once its id has all the text it keeps. --- crates/a11y/src/headings.rs | 12 ++++++------ crates/fonts/src/pdf.rs | 22 ++++++++++++++++++++-- 2 files changed, 26 insertions(+), 8 deletions(-) diff --git a/crates/a11y/src/headings.rs b/crates/a11y/src/headings.rs index d3a478cb95..41c749ed3c 100644 --- a/crates/a11y/src/headings.rs +++ b/crates/a11y/src/headings.rs @@ -285,17 +285,17 @@ pub(crate) fn mcid_texts(doc: &Document, page: &Dict) -> HashMap { }; for (code, len) in m.codes(&s.bytes) { let trm = tm.then(&gs.ctm); + // A glyph's text is only built while its id still takes more. if let Some(id) = id + && out.get(&id).is_none_or(|t| t.len() < MAX_TITLE * 4) && let Some(glyph) = m.text_of(code) { let text = out.entry(id).or_default(); - if text.len() < MAX_TITLE * 4 { - let breaks = ends.get(&id).is_some_and(|end| new_word(&trm, *end, gs.size, gs.scale)); - if breaks && !text.is_empty() && !text.ends_with(' ') && !glyph.starts_with(' ') { - text.push(' '); - } - text.push_str(&glyph); + let breaks = ends.get(&id).is_some_and(|end| new_word(&trm, *end, gs.size, gs.scale)); + if breaks && !text.is_empty() && !text.ends_with(' ') && !glyph.starts_with(' ') { + text.push(' '); } + text.push_str(&glyph); } let advance = m.width(code) * gs.size + gs.char_spacing + if m.is_space(code, len) { gs.word_spacing } else { 0.0 }; tm = Matrix::translate(advance * gs.scale, 0.0).then(&tm); diff --git a/crates/fonts/src/pdf.rs b/crates/fonts/src/pdf.rs index 0aaed6939e..0badcfc3a9 100644 --- a/crates/fonts/src/pdf.rs +++ b/crates/fonts/src/pdf.rs @@ -905,9 +905,13 @@ impl Metrics { } } -/// The UTF-16 code units of CMap hex bytes. +/// Most UTF-16 units kept from one ToUnicode destination: the CMap format allows destinations of up +/// to 512 bytes, and every glyph a sequential range covers builds its text from the whole of one. +const MAX_DESTINATION_UNITS: usize = 256; + +/// The UTF-16 code units of CMap hex bytes, at most [`MAX_DESTINATION_UNITS`] of them. fn code_units(b: &[u8]) -> Vec { - b.chunks(2).map(|c| u16::from_be_bytes([c[0], *c.get(1).unwrap_or(&0)])).collect() + b.chunks(2).take(MAX_DESTINATION_UNITS).map(|c| u16::from_be_bytes([c[0], *c.get(1).unwrap_or(&0)])).collect() } /// UTF-16BE (with surrogates) from CMap hex bytes. @@ -1262,6 +1266,20 @@ mod tests { use super::*; + /// Every glyph of a sequential range builds its text from the destination, so a long one is cut + /// to the CMap limit once, when it is read. + #[test] + fn unicode_destinations_are_cut_to_the_cmap_limit() { + let long = "0041".repeat(4096); + let cmap = format!("2 beginbfrange <0010> <0020> <{long}> <0030> <0030> [<{long}>] endbfrange 1 beginbfchar <0040> <{long}> endbfchar"); + let mut u = Unicode::default(); + parse_to_unicode(cmap.as_bytes(), &mut u); + for code in [0x10, 0x20, 0x30, 0x40] { + let text = u.text(code).unwrap_or_default(); + assert!(!text.is_empty() && text.len() <= MAX_DESTINATION_UNITS * 3, "code {code:#x}: {} bytes", text.len()); + } + } + /// A font whose ranges cover a few low codes keeps tables that size, not a fixed 64k-entry one. #[test] fn low_code_tables_end_with_the_last_covered_code() {