Skip to content

PdfCraft: An open-source PDF reader and editor written by scratch in pure Rust

fix(fonts): subset added TrueType fonts to used glyphs on document save

Combiler0142b2a3ebdbae
diff --git a/crates/edit/src/text.rs b/crates/edit/src/text.rs
index 272a15ec0..0e8dba5af 100644
--- a/crates/edit/src/text.rs
+++ b/crates/edit/src/text.rs
@@ -10,11 +10,11 @@
//! otherwise it is set in Helvetica (WinAnsi), and the result says the font was substituted.
//! Reflowing a paragraph to a new width is not done here (see `edit.text-reflow`).
-use std::collections::HashMap;
+use std::collections::{BTreeSet, HashMap};
use std::rc::Rc;
use pdfcraft_content::{Matrix, Op, Pieces, parse, serialize_ops};
-use pdfcraft_cos::{DecodeBudget, Dict, Document, Object, PdfString, Stream};
+use pdfcraft_cos::{DecodeBudget, Dict, Document, ObjRef, Object, PdfString, Stream};
use pdfcraft_fonts::pdf::Metrics;
use pdfcraft_fonts::{CraftFont, GlyphError, japanese_glyph_from};
@@ -158,6 +158,270 @@ pub fn register_text_font(doc: &mut Document, page: usize, bytes: &[u8]) -> Resu
Ok(name)
}
+fn page_content_bytes_for_subset(doc: &Document, page: &pdfcraft_model::Page) -> Vec<u8> {
+ let Some(c) = page.dict.get(b"Contents") else { return Vec::new() };
+ match doc.resolve(c).as_ref() {
+ Object::Stream(s) => s.decoded_within(16 << 20).unwrap_or_default(),
+ Object::Array(a) => a
+ .iter()
+ .flat_map(|x| match doc.resolve(x).as_ref() {
+ Object::Stream(s) => s.decoded_within(16 << 20).unwrap_or_default(),
+ _ => Vec::new(),
+ })
+ .collect(),
+ _ => Vec::new(),
+ }
+}
+
+/// Subset any added TrueType fonts in `doc` (marked with `/PCFullFont`) to only the glyphs used,
+/// update their BaseFont/FontName with an `ABCDEF+` subset prefix, and remove `/PCFullFont`.
+pub fn subset_document_fonts(doc: &mut Document) -> Result<(), EditError> {
+ let mut added_programs: Vec<ObjRef> = Vec::new();
+ for (r, obj) in doc.scan_objects() {
+ if let Object::Stream(s) = obj.as_ref()
+ && s.dict.get(b"PCFullFont").is_some()
+ {
+ added_programs.push(r);
+ }
+ }
+ if added_programs.is_empty() {
+ return Ok(());
+ }
+
+ let mut desc_for_program: HashMap<ObjRef, ObjRef> = HashMap::new();
+ let mut cid_for_desc: HashMap<ObjRef, ObjRef> = HashMap::new();
+ let mut type0_for_cid: HashMap<ObjRef, Vec<ObjRef>> = HashMap::new();
+
+ for (r, obj) in doc.scan_objects() {
+ if let Object::Dict(d) = obj.as_ref() {
+ if d.name(b"Type") == Some(b"FontDescriptor") {
+ if let Some(Object::Ref(p_ref)) = d.get(b"FontFile2").or_else(|| d.get(b"FontFile3"))
+ && added_programs.contains(p_ref)
+ {
+ desc_for_program.insert(*p_ref, r);
+ }
+ } else if d.name(b"Type") == Some(b"Font") {
+ if d.name(b"Subtype") == Some(b"CIDFontType2") || d.name(b"Subtype") == Some(b"CIDFontType0") {
+ if let Some(Object::Ref(d_ref)) = d.get(b"FontDescriptor") {
+ cid_for_desc.insert(*d_ref, r);
+ }
+ } else if d.name(b"Subtype") == Some(b"Type0") {
+ let desc_opt = d.get(b"DescendantFonts").map(|o| doc.resolve(o));
+ if let Some(desc_arc) = desc_opt
+ && let Object::Array(descendants) = desc_arc.as_ref()
+ && let Some(Object::Ref(c_ref)) = descendants.first()
+ {
+ type0_for_cid.entry(*c_ref).or_default().push(r);
+ }
+ }
+ }
+ }
+ }
+
+ let mut type0_to_program: HashMap<ObjRef, ObjRef> = HashMap::new();
+ for (&prog, &desc) in &desc_for_program {
+ if let Some(&cid) = cid_for_desc.get(&desc)
+ && let Some(type0s) = type0_for_cid.get(&cid)
+ {
+ for &t0 in type0s {
+ type0_to_program.insert(t0, prog);
+ }
+ }
+ }
+
+ let pages = pdfcraft_model::pages(doc);
+ let mut used_cids_by_program: HashMap<ObjRef, BTreeSet<u16>> = HashMap::new();
+
+ for page in &pages {
+ let Some(res) = page.dict.get(b"Resources").and_then(|r| doc.dict(r)) else { continue };
+ let mut font_res_to_program: HashMap<Vec<u8>, ObjRef> = HashMap::new();
+ if let Some(fonts) = res.get(b"Font").and_then(|f| doc.dict(f)) {
+ for (name, val) in fonts.iter() {
+ if let Object::Ref(r) = val
+ && let Some(&prog) = type0_to_program.get(r)
+ {
+ font_res_to_program.insert(name.clone(), prog);
+ }
+ }
+ }
+ if font_res_to_program.is_empty() {
+ continue;
+ }
+
+ let content = page_content_bytes_for_subset(doc, page);
+ let parsed = parse(&content);
+ let mut current_font: Option<Vec<u8>> = None;
+ for op in parsed.ops {
+ if op.is("Tf") {
+ current_font = op.operands.first().and_then(|o| o.as_name()).map(|n| n.to_vec());
+ } else if let Some(cur) = current_font.as_ref()
+ && let Some(&prog) = font_res_to_program.get(cur)
+ {
+ let cids = used_cids_by_program.entry(prog).or_default();
+ if op.is("Tj") || op.is("'") {
+ if let Some(s) = op.operands.first().and_then(|o| o.as_string()) {
+ for chunk in s.bytes.as_chunks::<2>().0 {
+ cids.insert(u16::from_be_bytes(*chunk));
+ }
+ }
+ } else if op.is("\"") {
+ if let Some(s) = op.operands.get(2).and_then(|o| o.as_string()) {
+ for chunk in s.bytes.as_chunks::<2>().0 {
+ cids.insert(u16::from_be_bytes(*chunk));
+ }
+ }
+ } else if op.is("TJ")
+ && let Some(arr) = op.operands.first().and_then(|o| o.as_array())
+ {
+ for item in arr {
+ if let Some(s) = item.as_string() {
+ for chunk in s.bytes.as_chunks::<2>().0 {
+ cids.insert(u16::from_be_bytes(*chunk));
+ }
+ }
+ }
+ }
+ }
+ }
+ }
+
+ for prog_ref in added_programs {
+ let Some(&desc_ref) = desc_for_program.get(&prog_ref) else { continue };
+ let Some(&cid_ref) = cid_for_desc.get(&desc_ref) else { continue };
+ let empty_set = BTreeSet::new();
+ let used_cids = used_cids_by_program.get(&prog_ref).unwrap_or(&empty_set);
+
+ let Some(cid_dict) = doc.dict(&Object::Ref(cid_ref)) else { continue };
+ let cid_to_gid_obj = cid_dict.get(b"CIDToGIDMap").cloned();
+ let mut map_bytes: Option<(ObjRef, Vec<u8>)> = None;
+ let mut used_gids = BTreeSet::new();
+ used_gids.insert(0u16);
+
+ if let Some(Object::Ref(map_ref)) = cid_to_gid_obj {
+ if let Object::Stream(map_stream) = doc.get(map_ref).as_ref()
+ && let Ok(bytes) = map_stream.decoded_within(1 << 17)
+ {
+ for &cid in used_cids {
+ let offset = usize::from(cid).saturating_mul(2);
+ if let Some(chunk) = bytes.get(offset..offset.saturating_add(2))
+ && chunk.len() == 2
+ {
+ let gid = u16::from_be_bytes([chunk[0], chunk[1]]);
+ used_gids.insert(gid);
+ }
+ }
+ map_bytes = Some((map_ref, bytes));
+ }
+ } else {
+ for &cid in used_cids {
+ used_gids.insert(cid);
+ }
+ }
+
+ let Object::Stream(prog_stream) = doc.get(prog_ref).as_ref().clone() else { continue };
+ let Ok(font_bytes) = prog_stream.decoded_within(16 << 20) else { continue };
+
+ let prefix = pdfcraft_fonts::embedding::subset_tag(u64::from(prog_ref.num));
+ let mut desc_dict = doc.dict(&Object::Ref(desc_ref)).unwrap_or_default();
+ let orig_name =
+ desc_dict.name(b"FontName").or_else(|| cid_dict.name(b"BaseFont")).and_then(|n| std::str::from_utf8(n).ok()).unwrap_or("Font");
+ let clean_name = orig_name.rsplit('+').next().unwrap_or(orig_name);
+ let subset_name = format!("{prefix}{clean_name}");
+
+ let used_gids_vec: Vec<u16> = used_gids.into_iter().collect();
+
+ if let Ok((subset_bytes, old_to_new)) = pdfcraft_fonts::embedding::subset_truetype(&font_bytes, &used_gids_vec, &subset_name) {
+ let mut new_prog_dict = prog_stream.dict.clone();
+ new_prog_dict.remove(b"PCFullFont");
+ new_prog_dict.set(b"Length1".to_vec(), Object::Int(subset_bytes.len() as i64));
+ doc.set(prog_ref, Object::Stream(Stream::flate(new_prog_dict, &subset_bytes)));
+
+ if let Some((map_ref, mut m_bytes)) = map_bytes {
+ let max_cid = used_cids.iter().copied().next_back().unwrap_or(0);
+ let needed_len = usize::from(max_cid).saturating_add(1).saturating_mul(2);
+ if m_bytes.len() > needed_len {
+ m_bytes.truncate(needed_len);
+ }
+ for i in 0..m_bytes.len() / 2 {
+ let s_idx = i.saturating_mul(2);
+ if let Some(chunk) = m_bytes.get(s_idx..s_idx.saturating_add(2)) {
+ let old_gid = u16::from_be_bytes([chunk[0], chunk[1]]);
+ let new_gid = old_to_new.get(&old_gid).copied().unwrap_or(0);
+ if let Some(target) = m_bytes.get_mut(s_idx..s_idx.saturating_add(2)) {
+ target.copy_from_slice(&new_gid.to_be_bytes());
+ }
+ }
+ }
+ doc.set(map_ref, Object::Stream(Stream::flate(Dict::new(), &m_bytes)));
+ }
+ } else {
+ let mut new_prog_dict = prog_stream.dict.clone();
+ new_prog_dict.remove(b"PCFullFont");
+ doc.set(prog_ref, Object::Stream(Stream::flate(new_prog_dict, &font_bytes)));
+ }
+
+ desc_dict.set(b"FontName".to_vec(), Object::name(&subset_name));
+ doc.set(desc_ref, Object::Dict(desc_dict));
+
+ let mut new_cid_dict = cid_dict.clone();
+ new_cid_dict.set(b"BaseFont".to_vec(), Object::name(&subset_name));
+
+ let first_t0 = type0_for_cid.get(&cid_ref).and_then(|t0s| t0s.first()).and_then(|r| doc.dict(&Object::Ref(*r)));
+ let m_opt = first_t0.as_ref().map(|d| Metrics::from_dict(doc, d));
+
+ if let Some(ref m) = m_opt {
+ let mut w_array = Vec::new();
+ for &cid in used_cids {
+ let w = m.width(u32::from(cid)) * 1000.0;
+ w_array.push(Object::Int(i64::from(cid)));
+ w_array.push(Object::Array(vec![Object::Real(w)]));
+ }
+ if !w_array.is_empty() {
+ new_cid_dict.set(b"W".to_vec(), Object::Array(w_array));
+ }
+
+ let mut cmap = String::from(
+ "/CIDInit /ProcSet findresource begin\n12 dict begin\nbegincmap\n/CMapName /PCUnicode def\n/CMapType 2 def\n1 begincodespacerange\n<0000> <FFFF>\nendcodespacerange\n",
+ );
+ let used_cids_vec: Vec<u16> = used_cids.iter().copied().collect();
+ for chunk in used_cids_vec.chunks(100) {
+ use std::fmt::Write as _;
+ let _ = writeln!(cmap, "{} beginbfchar", chunk.len());
+ for &cid in chunk {
+ let text = m.decode(&cid.to_be_bytes());
+ let hex: String = if text.is_empty() { format!("{cid:04X}") } else { text.encode_utf16().map(|u| format!("{u:04X}")).collect() };
+ let _ = writeln!(cmap, "<{cid:04X}> <{hex}>");
+ }
+ cmap.push_str("endbfchar\n");
+ }
+ cmap.push_str("endcmap\nCMapName currentdict /CMap defineresource pop\nend\nend\n");
+
+ if let Some(type0s) = type0_for_cid.get(&cid_ref) {
+ for &t0_ref in type0s {
+ if let Some(t0_dict) = doc.dict(&Object::Ref(t0_ref))
+ && let Some(Object::Ref(to_uni_ref)) = t0_dict.get(b"ToUnicode")
+ {
+ doc.set(*to_uni_ref, Object::Stream(Stream::flate(Dict::new(), cmap.as_bytes())));
+ }
+ }
+ }
+ }
+
+ doc.set(cid_ref, Object::Dict(new_cid_dict));
+
+ if let Some(type0s) = type0_for_cid.get(&cid_ref) {
+ for &t0_ref in type0s {
+ if let Some(mut t0_dict) = doc.dict(&Object::Ref(t0_ref)) {
+ t0_dict.set(b"BaseFont".to_vec(), Object::name(&subset_name));
+ doc.set(t0_ref, Object::Dict(t0_dict));
+ }
+ }
+ }
+ }
+
+ Ok(())
+}
+
/// Read-only draft bounds, measured with the same PDF rewrite used by Apply/Save.
#[derive(Clone, Debug, PartialEq)]
pub struct TextEditLayout {
diff --git a/crates/engine/src/lib.rs b/crates/engine/src/lib.rs
index a8ace8b5d..21175bdb4 100644
--- a/crates/engine/src/lib.rs
+++ b/crates/engine/src/lib.rs
@@ -4484,12 +4484,14 @@ impl Session {
// file so that revision stays in the chain. A full rewrite (redaction, new encryption,
// a reconstructed file) still writes the current graph on its own.
let rewrite = editor.cos.full_save_required() || editor.cos.encryption_changed() || editor.cos.revisions().is_empty();
+ let mut cos = editor.cos.clone();
+ pdfcraft_edit::text::subset_document_fonts(&mut cos)?;
guard(|| {
if on_saved || rewrite {
- write_incremental(&editor.cos, &opts)
+ write_incremental(&cos, &opts)
} else {
let mut base = pdfcraft_cos::Document::open_with_password(editor.saved.clone(), editor.saved_keys.reopen.as_deref())?;
- base.overlay_changes_from(&editor.cos)?;
+ base.overlay_changes_from(&cos)?;
write_incremental(&base, &opts)
}
})
@@ -4506,7 +4508,9 @@ impl Session {
}
let editor = doc.editor.as_ref().ok_or_else(|| EditError::ReadOnly(doc.read_only_reason.clone().unwrap_or_default()))?;
let opts = SaveOptions { mod_date: self.now().map(pdfcraft_cos::pdf_date), ..SaveOptions::default() };
- guard(|| write_full(&editor.cos, &opts)).map_err(EditError::Write)?.map(Arc::new).map_err(|e| EditError::Write(e.to_string()))
+ let mut cos = editor.cos.clone();
+ pdfcraft_edit::text::subset_document_fonts(&mut cos)?;
+ guard(|| write_full(&cos, &opts)).map_err(EditError::Write)?.map(Arc::new).map_err(|e| EditError::Write(e.to_string()))
}
/// Check that `bytes`, from [`Self::save_bytes`] or [`Self::save_full_bytes`], reopen as this
diff --git a/crates/engine/tests/it/document_fonts.rs b/crates/engine/tests/it/document_fonts.rs
index e4212f4a3..bf9415ce9 100644
--- a/crates/engine/tests/it/document_fonts.rs
+++ b/crates/engine/tests/it/document_fonts.rs
@@ -96,3 +96,50 @@ fn contextual_arabic_marks_ligatures_and_mixed_bidi_keep_logical_text_after_reop
}
exercise(font.bytes.to_vec(), "سَلَام 123", "NotoSansArabic-Regular");
}
+
+#[test]
+fn added_font_is_subset_to_used_glyphs_on_save() {
+ let font_bytes = std::fs::read(std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("../../assets/fonts/Inter-Regular.ttf")).unwrap();
+ let font_len = font_bytes.len();
+ assert!(font_len > 300_000, "source font is full sized");
+
+ let mut s = Session::new();
+ let id = s.open("source.pdf", None, document(), None).unwrap();
+ s.apply(id, Edit::RegisterTextFont { page: 0, bytes: Arc::new(font_bytes) }).unwrap();
+ let resource = s.get(id).unwrap().text_editor_fonts(0).into_iter().find(|f| f.face == "Inter-Regular").unwrap().resource;
+ let style = BlockStyle { resource: Some(resource), width: Some(240.0), ..Default::default() };
+ let text = "Three words here";
+ s.apply(id, Edit::EditTextBlock { page: 0, block: 0, text: text.into(), style }).unwrap();
+
+ let saved = s.save_bytes(id).unwrap();
+ assert!(saved.len() < 50_000, "saved PDF with 3 words must be subset and small, was {} bytes", saved.len());
+
+ let saved_doc = pdfcraft_cos::Document::open(saved.clone()).unwrap();
+ let mut subset_found = false;
+ for (_, obj) in saved_doc.scan_objects() {
+ if let pdfcraft_cos::Object::Dict(d) = obj.as_ref()
+ && d.name(b"Type") == Some(b"Font")
+ && d.name(b"Subtype") == Some(b"Type0")
+ && let Some(base) = d.name(b"BaseFont")
+ {
+ let base_str = String::from_utf8_lossy(base);
+ if base_str.ends_with("Inter-Regular") {
+ assert!(base_str.len() >= 7 && base_str.as_bytes()[6] == b'+', "font base name {base_str} must have subset prefix");
+ subset_found = true;
+ }
+ }
+ if let pdfcraft_cos::Object::Stream(stream) = obj.as_ref()
+ && stream.dict.get(b"PCFullFont").is_some()
+ {
+ panic!("saved PDF must not retain PCFullFont flag");
+ }
+ }
+ assert!(subset_found, "Type0 subset font must be present in saved document");
+
+ // Also verify reopening and rendering matches:
+ let re = s.open("saved.pdf", None, saved, None).unwrap();
+ assert_eq!(s.get(re).unwrap().text_blocks(0)[0].editor_text, text);
+ let re_pixels = pixels(s.get(re).unwrap().display.clone());
+ let orig_pixels = pixels(s.get(id).unwrap().display.clone());
+ assert_eq!(re_pixels, orig_pixels, "reopened subset font renders identically to original");
+}
diff --git a/crates/fonts/src/embedding.rs b/crates/fonts/src/embedding.rs
index ea0f74785..866ab7c0b 100644
--- a/crates/fonts/src/embedding.rs
+++ b/crates/fonts/src/embedding.rs
@@ -1,7 +1,7 @@
//! Unicode embedding of an explicitly supplied, embeddable TrueType face. Font files are
//! document inputs or the optional craft-fonts build input, never system-interface fallbacks.
-use std::collections::HashMap;
+use std::collections::{BTreeSet, HashMap};
use std::fmt::Write as _;
use pdfcraft_cos::{Dict, Document, ObjRef, Object, PdfString, Stream};
@@ -338,6 +338,438 @@ fn same_resource(doc: &Document, fonts: &Dict, name: &str, programs: &mut Progra
.map(|w| w.resource)
}
+/// Generate a 6-character subset prefix tag for a font name, e.g. "ABCDEF+".
+pub fn subset_tag(seed: u64) -> String {
+ let mut tag = String::with_capacity(7);
+ let mut s = seed;
+ for _ in 0..6 {
+ let b = b'A'.saturating_add((s % 26) as u8);
+ tag.push(b as char);
+ s /= 26;
+ }
+ tag.push('+');
+ tag
+}
+
+fn calc_table_checksum(data: &[u8]) -> u32 {
+ let mut sum = 0u32;
+ for chunk in data.chunks(4) {
+ let mut b = [0u8; 4];
+ if let Some(target) = b.get_mut(..chunk.len()) {
+ target.copy_from_slice(chunk);
+ }
+ sum = sum.wrapping_add(u32::from_be_bytes(b));
+ }
+ sum
+}
+
+fn read_tables(bytes: &[u8]) -> Option<HashMap<[u8; 4], &[u8]>> {
+ let count_bytes: [u8; 2] = bytes.get(4..6)?.try_into().ok()?;
+ let num_tables = usize::from(u16::from_be_bytes(count_bytes));
+ if num_tables > 128 {
+ return None;
+ }
+ let mut map = HashMap::new();
+ for i in 0..num_tables {
+ let entry = 12usize.checked_add(i.checked_mul(16)?)?;
+ let tag: [u8; 4] = bytes.get(entry..entry.checked_add(4)?)?.try_into().ok()?;
+ let off_bytes: [u8; 4] = bytes.get(entry.checked_add(8)?..entry.checked_add(12)?)?.try_into().ok()?;
+ let offset = usize::try_from(u32::from_be_bytes(off_bytes)).ok()?;
+ let len_bytes: [u8; 4] = bytes.get(entry.checked_add(12)?..entry.checked_add(16)?)?.try_into().ok()?;
+ let length = usize::try_from(u32::from_be_bytes(len_bytes)).ok()?;
+ let table_data = bytes.get(offset..offset.checked_add(length)?)?;
+ map.insert(tag, table_data);
+ }
+ Some(map)
+}
+
+fn glyph_slice<'a>(loca: &'a [u8], glyf: &'a [u8], is_long: bool, gid: u16) -> Option<&'a [u8]> {
+ let (start, end) = if is_long {
+ let s_idx = usize::from(gid).checked_mul(4)?;
+ let e_idx = s_idx.checked_add(4)?;
+ let start_bytes: [u8; 4] = loca.get(s_idx..e_idx)?.try_into().ok()?;
+ let end_bytes: [u8; 4] = loca.get(e_idx..e_idx.checked_add(4)?)?.try_into().ok()?;
+ (usize::try_from(u32::from_be_bytes(start_bytes)).ok()?, usize::try_from(u32::from_be_bytes(end_bytes)).ok()?)
+ } else {
+ let s_idx = usize::from(gid).checked_mul(2)?;
+ let e_idx = s_idx.checked_add(2)?;
+ let start_bytes: [u8; 2] = loca.get(s_idx..e_idx)?.try_into().ok()?;
+ let end_bytes: [u8; 2] = loca.get(e_idx..e_idx.checked_add(2)?)?.try_into().ok()?;
+ (usize::from(u16::from_be_bytes(start_bytes)).checked_mul(2)?, usize::from(u16::from_be_bytes(end_bytes)).checked_mul(2)?)
+ };
+ if start <= end && end <= glyf.len() { glyf.get(start..end) } else { None }
+}
+
+fn composite_components(data: &[u8]) -> Vec<u16> {
+ let mut comps = Vec::new();
+ if data.len() < 10 {
+ return comps;
+ }
+ let Some(header) = data.get(0..2) else { return comps };
+ let num_contours = i16::from_be_bytes([header[0], header[1]]);
+ if num_contours != -1 {
+ return comps;
+ }
+ let mut offset = 10usize;
+ while let Some(flags_bytes) = data.get(offset..offset.saturating_add(2)) {
+ if flags_bytes.len() < 2 {
+ break;
+ }
+ let flags = u16::from_be_bytes([flags_bytes[0], flags_bytes[1]]);
+ let Some(gid_bytes) = data.get(offset.saturating_add(2)..offset.saturating_add(4)) else { break };
+ if gid_bytes.len() < 2 {
+ break;
+ }
+ let gid = u16::from_be_bytes([gid_bytes[0], gid_bytes[1]]);
+ comps.push(gid);
+ let Some(mut next_offset) = offset.checked_add(4) else { break };
+ if flags & 0x0001 != 0 {
+ next_offset = match next_offset.checked_add(4) {
+ Some(o) => o,
+ None => break,
+ };
+ } else {
+ next_offset = match next_offset.checked_add(2) {
+ Some(o) => o,
+ None => break,
+ };
+ }
+ if flags & 0x0008 != 0 {
+ next_offset = match next_offset.checked_add(2) {
+ Some(o) => o,
+ None => break,
+ };
+ } else if flags & 0x0040 != 0 {
+ next_offset = match next_offset.checked_add(4) {
+ Some(o) => o,
+ None => break,
+ };
+ } else if flags & 0x0080 != 0 {
+ next_offset = match next_offset.checked_add(8) {
+ Some(o) => o,
+ None => break,
+ };
+ }
+ offset = next_offset;
+ if flags & 0x0020 == 0 {
+ break;
+ }
+ }
+ comps
+}
+
+fn remap_composite_glyph(data: &[u8], old_to_new: &HashMap<u16, u16>) -> Vec<u8> {
+ let mut out = data.to_vec();
+ if out.len() < 10 {
+ return out;
+ }
+ let Some(header) = out.get(0..2) else { return out };
+ let num_contours = i16::from_be_bytes([header[0], header[1]]);
+ if num_contours != -1 {
+ return out;
+ }
+ let mut offset = 10usize;
+ while let Some(flags_bytes) = out.get(offset..offset.saturating_add(2)) {
+ if flags_bytes.len() < 2 {
+ break;
+ }
+ let flags = u16::from_be_bytes([flags_bytes[0], flags_bytes[1]]);
+ let Some(gid_bytes) = out.get(offset.saturating_add(2)..offset.saturating_add(4)) else { break };
+ if gid_bytes.len() < 2 {
+ break;
+ }
+ let old_gid = u16::from_be_bytes([gid_bytes[0], gid_bytes[1]]);
+ let new_gid = old_to_new.get(&old_gid).copied().unwrap_or(0);
+ if let Some(slot) = out.get_mut(offset.saturating_add(2)..offset.saturating_add(4)) {
+ slot.copy_from_slice(&new_gid.to_be_bytes());
+ }
+ let Some(mut next_offset) = offset.checked_add(4) else { break };
+ if flags & 0x0001 != 0 {
+ next_offset = match next_offset.checked_add(4) {
+ Some(o) => o,
+ None => break,
+ };
+ } else {
+ next_offset = match next_offset.checked_add(2) {
+ Some(o) => o,
+ None => break,
+ };
+ }
+ if flags & 0x0008 != 0 {
+ next_offset = match next_offset.checked_add(2) {
+ Some(o) => o,
+ None => break,
+ };
+ } else if flags & 0x0040 != 0 {
+ next_offset = match next_offset.checked_add(4) {
+ Some(o) => o,
+ None => break,
+ };
+ } else if flags & 0x0080 != 0 {
+ next_offset = match next_offset.checked_add(8) {
+ Some(o) => o,
+ None => break,
+ };
+ }
+ offset = next_offset;
+ if flags & 0x0020 == 0 {
+ break;
+ }
+ }
+ out
+}
+
+fn build_name_table(ps_name: &str) -> Vec<u8> {
+ let utf16_be: Vec<u8> = ps_name.encode_utf16().flat_map(u16::to_be_bytes).collect();
+ let str_len = u16::try_from(utf16_be.len()).unwrap_or(0);
+ let mut out = Vec::new();
+ out.extend_from_slice(&0u16.to_be_bytes());
+ out.extend_from_slice(&1u16.to_be_bytes());
+ out.extend_from_slice(&18u16.to_be_bytes());
+ out.extend_from_slice(&3u16.to_be_bytes());
+ out.extend_from_slice(&1u16.to_be_bytes());
+ out.extend_from_slice(&0x0409u16.to_be_bytes());
+ out.extend_from_slice(&6u16.to_be_bytes());
+ out.extend_from_slice(&str_len.to_be_bytes());
+ out.extend_from_slice(&0u16.to_be_bytes());
+ out.extend_from_slice(&utf16_be);
+ out
+}
+
+fn build_cmap_table(char_to_new_gid: &[(u32, u32)]) -> Vec<u8> {
+ let mut sorted = char_to_new_gid.to_vec();
+ if sorted.is_empty() {
+ sorted.push((0, 0));
+ }
+ sorted.sort_by_key(|&(c, _)| c);
+ sorted.dedup_by_key(|&mut (c, _)| c);
+ let n_groups = u32::try_from(sorted.len()).unwrap_or(1);
+ let subtable_len = 16u32.saturating_add(n_groups.saturating_mul(12));
+ let mut out = Vec::new();
+ out.extend_from_slice(&0u16.to_be_bytes());
+ out.extend_from_slice(&1u16.to_be_bytes());
+ out.extend_from_slice(&3u16.to_be_bytes());
+ out.extend_from_slice(&10u16.to_be_bytes());
+ out.extend_from_slice(&12u32.to_be_bytes());
+ out.extend_from_slice(&12u16.to_be_bytes());
+ out.extend_from_slice(&0u16.to_be_bytes());
+ out.extend_from_slice(&subtable_len.to_be_bytes());
+ out.extend_from_slice(&0u32.to_be_bytes());
+ out.extend_from_slice(&n_groups.to_be_bytes());
+ for &(c, gid) in &sorted {
+ out.extend_from_slice(&c.to_be_bytes());
+ out.extend_from_slice(&c.to_be_bytes());
+ out.extend_from_slice(&gid.to_be_bytes());
+ }
+ out
+}
+
+/// Subset a TrueType font to only the specified glyph IDs.
+/// Returns the new sfnt font binary and a mapping from old GID to new GID.
+pub fn subset_truetype(bytes: &[u8], used_gids: &[u16], postscript_name: &str) -> Result<(Vec<u8>, HashMap<u16, u16>), String> {
+ validate(bytes)?;
+ let map = read_tables(bytes).ok_or("cannot parse font table directory")?;
+ let orig_glyf = map.get(b"glyf").copied().ok_or("the font has no TrueType glyf table")?;
+ let orig_loca = map.get(b"loca").copied().ok_or("missing loca table")?;
+ let orig_head = map.get(b"head").copied().ok_or("missing head table")?;
+ let orig_hhea = map.get(b"hhea").copied().ok_or("missing hhea table")?;
+ let orig_hmtx = map.get(b"hmtx").copied().ok_or("missing hmtx table")?;
+ let orig_maxp = map.get(b"maxp").copied().ok_or("missing maxp table")?;
+
+ let index_fmt_bytes: [u8; 2] = orig_head.get(50..52).ok_or("invalid head table")?.try_into().map_err(|_| "invalid head table")?;
+ let is_long = i16::from_be_bytes(index_fmt_bytes) != 0;
+
+ let mut needed_gids = BTreeSet::new();
+ needed_gids.insert(0u16);
+ for &g in used_gids {
+ needed_gids.insert(g);
+ }
+ let mut queue: Vec<u16> = needed_gids.iter().copied().collect();
+ while let Some(gid) = queue.pop() {
+ if let Some(g_data) = glyph_slice(orig_loca, orig_glyf, is_long, gid) {
+ for comp in composite_components(g_data) {
+ if needed_gids.insert(comp) {
+ queue.push(comp);
+ }
+ }
+ }
+ }
+
+ let mut subset_glyphs = Vec::with_capacity(needed_gids.len());
+ let mut old_to_new_gid = HashMap::new();
+ for (new_gid, &old_gid) in needed_gids.iter().enumerate() {
+ let new_gid_u16 = u16::try_from(new_gid).map_err(|_| "too many glyphs in subset")?;
+ old_to_new_gid.insert(old_gid, new_gid_u16);
+ subset_glyphs.push(old_gid);
+ }
+
+ let mut subset_glyf = Vec::new();
+ let mut subset_loca = Vec::with_capacity((subset_glyphs.len().saturating_add(1)).saturating_mul(4));
+ for &old_gid in &subset_glyphs {
+ let offset_u32 = u32::try_from(subset_glyf.len()).map_err(|_| "glyf table overflow")?;
+ subset_loca.extend_from_slice(&offset_u32.to_be_bytes());
+ if let Some(g_data) = glyph_slice(orig_loca, orig_glyf, is_long, old_gid)
+ && !g_data.is_empty()
+ {
+ let remapped = remap_composite_glyph(g_data, &old_to_new_gid);
+ subset_glyf.extend_from_slice(&remapped);
+ if subset_glyf.len() % 2 != 0 {
+ subset_glyf.push(0);
+ }
+ }
+ }
+ let final_offset_u32 = u32::try_from(subset_glyf.len()).map_err(|_| "glyf table overflow")?;
+ subset_loca.extend_from_slice(&final_offset_u32.to_be_bytes());
+
+ let hmetrics_bytes: [u8; 2] = orig_hhea.get(34..36).ok_or("invalid hhea")?.try_into().map_err(|_| "invalid hhea")?;
+ let orig_num_hmetrics = usize::from(u16::from_be_bytes(hmetrics_bytes));
+ let mut subset_hmtx = Vec::with_capacity(subset_glyphs.len().saturating_mul(4));
+ for &old_gid in &subset_glyphs {
+ let old_idx = usize::from(old_gid);
+ let (adv, lsb) = if old_idx < orig_num_hmetrics {
+ let adv_bytes: [u8; 2] = orig_hmtx.get(old_idx * 4..old_idx * 4 + 2).unwrap_or(&[0, 0]).try_into().unwrap_or([0, 0]);
+ let lsb_bytes: [u8; 2] = orig_hmtx.get(old_idx * 4 + 2..old_idx * 4 + 4).unwrap_or(&[0, 0]).try_into().unwrap_or([0, 0]);
+ (u16::from_be_bytes(adv_bytes), i16::from_be_bytes(lsb_bytes))
+ } else {
+ let last_adv_offset = orig_num_hmetrics.saturating_sub(1) * 4;
+ let adv_bytes: [u8; 2] = orig_hmtx.get(last_adv_offset..last_adv_offset + 2).unwrap_or(&[0, 0]).try_into().unwrap_or([0, 0]);
+ let lsb_offset = orig_num_hmetrics * 4 + (old_idx.saturating_sub(orig_num_hmetrics)) * 2;
+ let lsb_bytes: [u8; 2] = orig_hmtx.get(lsb_offset..lsb_offset + 2).unwrap_or(&[0, 0]).try_into().unwrap_or([0, 0]);
+ (u16::from_be_bytes(adv_bytes), i16::from_be_bytes(lsb_bytes))
+ };
+ subset_hmtx.extend_from_slice(&adv.to_be_bytes());
+ subset_hmtx.extend_from_slice(&lsb.to_be_bytes());
+ }
+
+ let mut subset_hhea = orig_hhea.to_vec();
+ if subset_hhea.len() >= 36 {
+ let num_metrics = u16::try_from(subset_glyphs.len()).map_err(|_| "too many glyphs")?;
+ if let Some(target) = subset_hhea.get_mut(34..36) {
+ target.copy_from_slice(&num_metrics.to_be_bytes());
+ }
+ }
+
+ let mut subset_maxp = orig_maxp.to_vec();
+ if subset_maxp.len() >= 6 {
+ let num_g = u16::try_from(subset_glyphs.len()).map_err(|_| "too many glyphs")?;
+ if let Some(target) = subset_maxp.get_mut(4..6) {
+ target.copy_from_slice(&num_g.to_be_bytes());
+ }
+ }
+
+ let mut subset_head = orig_head.to_vec();
+ if subset_head.len() >= 54 {
+ if let Some(target) = subset_head.get_mut(8..12) {
+ target.copy_from_slice(&[0, 0, 0, 0]);
+ }
+ if let Some(target) = subset_head.get_mut(50..52) {
+ target.copy_from_slice(&1i16.to_be_bytes());
+ }
+ }
+
+ let mut subset_post = vec![0u8; 32];
+ if let Some(target) = subset_post.get_mut(0..4) {
+ target.copy_from_slice(&0x00030000u32.to_be_bytes());
+ }
+ if let Some(orig_post) = map.get(b"post")
+ && orig_post.len() >= 16
+ && let (Some(target), Some(src)) = (subset_post.get_mut(4..16), orig_post.get(4..16))
+ {
+ target.copy_from_slice(src);
+ }
+
+ let char_to_new: Vec<(u32, u32)> = FontRef::new(bytes)
+ .ok()
+ .map(|face| {
+ face.charmap().mappings().filter_map(|(cp, old)| old_to_new_gid.get(&(old.to_u32() as u16)).map(|&new| (cp, u32::from(new)))).collect()
+ })
+ .unwrap_or_default();
+
+ let subset_name = build_name_table(postscript_name);
+ let subset_cmap = build_cmap_table(&char_to_new);
+
+ let mut tables: Vec<([u8; 4], Vec<u8>)> = vec![
+ (*b"cmap", subset_cmap),
+ (*b"glyf", subset_glyf),
+ (*b"head", subset_head),
+ (*b"hhea", subset_hhea),
+ (*b"hmtx", subset_hmtx),
+ (*b"loca", subset_loca),
+ (*b"maxp", subset_maxp),
+ (*b"name", subset_name),
+ ];
+ if let Some(os2) = map.get(b"OS/2") {
+ tables.push((*b"OS/2", os2.to_vec()));
+ }
+ tables.push((*b"post", subset_post));
+
+ tables.sort_by_key(|t| t.0);
+
+ let num_tables = u16::try_from(tables.len()).map_err(|_| "too many tables")?;
+ let mut max_pow2 = 1u16;
+ let mut entry_selector = 0u16;
+ while max_pow2.saturating_mul(2) <= num_tables {
+ max_pow2 = max_pow2.saturating_mul(2);
+ entry_selector = entry_selector.saturating_add(1);
+ }
+ let search_range = max_pow2.saturating_mul(16);
+ let range_shift = num_tables.saturating_mul(16).saturating_sub(search_range);
+
+ let mut sfnt = Vec::new();
+ sfnt.extend_from_slice(&0x00010000u32.to_be_bytes());
+ sfnt.extend_from_slice(&num_tables.to_be_bytes());
+ sfnt.extend_from_slice(&search_range.to_be_bytes());
+ sfnt.extend_from_slice(&entry_selector.to_be_bytes());
+ sfnt.extend_from_slice(&range_shift.to_be_bytes());
+
+ let dir_size = 12usize.saturating_add(usize::from(num_tables).saturating_mul(16));
+ let mut current_offset = dir_size;
+
+ struct Rec {
+ tag: [u8; 4],
+ checksum: u32,
+ offset: u32,
+ length: u32,
+ }
+ let mut records = Vec::with_capacity(tables.len());
+ for (tag, data) in &tables {
+ let checksum = calc_table_checksum(data);
+ let offset = u32::try_from(current_offset).map_err(|_| "font offset overflow")?;
+ let length = u32::try_from(data.len()).map_err(|_| "font table overflow")?;
+ records.push(Rec { tag: *tag, checksum, offset, length });
+ let padded_len = (data.len().saturating_add(3)) & !3;
+ current_offset = current_offset.saturating_add(padded_len);
+ }
+
+ for r in &records {
+ sfnt.extend_from_slice(&r.tag);
+ sfnt.extend_from_slice(&r.checksum.to_be_bytes());
+ sfnt.extend_from_slice(&r.offset.to_be_bytes());
+ sfnt.extend_from_slice(&r.length.to_be_bytes());
+ }
+
+ let mut head_offset = None;
+ for (r, (_, data)) in records.iter().zip(&tables) {
+ if &r.tag == b"head" {
+ head_offset = Some(sfnt.len());
+ }
+ sfnt.extend_from_slice(data);
+ while sfnt.len() % 4 != 0 {
+ sfnt.push(0);
+ }
+ }
+
+ if let Some(h_off) = head_offset {
+ let file_checksum = calc_table_checksum(&sfnt);
+ let adjustment = 0xB1B0AFBAu32.wrapping_sub(file_checksum);
+ if let Some(slot) = sfnt.get_mut(h_off.saturating_add(8)..h_off.saturating_add(12)) {
+ slot.copy_from_slice(&adjustment.to_be_bytes());
+ }
+ }
+
+ Ok((sfnt, old_to_new_gid))
+}
+
#[cfg(test)]
mod tests {
use super::*;
@@ -488,4 +920,19 @@ mod tests {
assert_eq!(programs, 1, "one program for both pages");
assert!(doc.scan_objects().count() > objects, "the second page gets its own resource");
}
+
+ #[test]
+ fn subset_truetype_creates_valid_compact_font() {
+ let bytes = include_bytes!("../../../assets/fonts/Inter-Regular.ttf");
+ let face = FontRef::new(bytes).unwrap();
+ let gid_t = face.charmap().map('T').unwrap().to_u32() as u16;
+ let gid_e = face.charmap().map('e').unwrap().to_u32() as u16;
+ let (subset, mapping) = subset_truetype(bytes, &[gid_t, gid_e], "ABCDEF+Inter-Regular").unwrap();
+ assert!(subset.len() < 10_000, "subset should be tiny, was {}", subset.len());
+ assert!(mapping.contains_key(&gid_t));
+ assert!(mapping.contains_key(&gid_e));
+ let sub_face = FontRef::new(&subset).unwrap();
+ assert_eq!(sub_face.charmap().map('T').unwrap().to_u32(), u32::from(*mapping.get(&gid_t).unwrap()));
+ assert_eq!(sub_face.charmap().map('e').unwrap().to_u32(), u32::from(*mapping.get(&gid_e).unwrap()));
+ }
}