@stll/folio-core 0.32.1 → 0.33.0

package/dist/fonts/sfnt/parse.js

@@ -0,0 +1,525 @@
+import { SFNT_TABLE, SFNT_VERSION, fitsInView, readLocaOffsets, readSfntDirectory, readSfntVersion } from "./tables.js";
+import { Result, TaggedError } from "better-result";
+//#region src/fonts/sfnt/parse.ts
+/**
+* A self-contained TrueType/OpenType (sfnt) reader.
+*
+* It reads exactly what text measurement and PDF embedding need: the metrics
+* head/hhea/OS-2 carry, a character-to-glyph lookup, and advance widths. It
+* does not decode outlines (the subsetter copies `glyf` bytes verbatim) and it
+* does not hint, shape or rasterize.
+*
+* Robustness contract: every offset and length is checked against the buffer
+* before it is read, so a truncated or lying font yields an `Err` rather than
+* a throw, and no field is ever `NaN`.
+*/
+var SfntParseError = class extends TaggedError("SfntParseError") {};
+const fail = (message) => Result.err(new SfntParseError({ message }));
+/** Byte offsets inside `head`. */
+const HEAD = {
+	checkSumAdjustment: 8,
+	magicNumber: 12,
+	unitsPerEm: 18,
+	xMin: 36,
+	yMin: 38,
+	xMax: 40,
+	yMax: 42,
+	indexToLocFormat: 50,
+	size: 54
+};
+/** `head.magicNumber`, the one field that proves the table is really `head`. */
+const HEAD_MAGIC = 1594834165;
+const MIN_UNITS_PER_EM = 16;
+const MAX_UNITS_PER_EM = 16384;
+/** Byte offsets inside `hhea`. */
+const HHEA = {
+	ascender: 4,
+	descender: 6,
+	lineGap: 8,
+	numberOfHMetrics: 34,
+	size: 36
+};
+/** Byte offsets inside `maxp` (only the half both versions share). */
+const MAXP = {
+	numGlyphs: 4,
+	size: 6
+};
+/** Byte offsets inside `OS/2`. */
+const OS2 = {
+	version: 0,
+	fsType: 8,
+	sxHeight: 86,
+	sCapHeight: 88,
+	/** Version 2 is the first that has sxHeight/sCapHeight at all. */
+	minimumVersionWithHeights: 2,
+	sizeWithHeights: 96
+};
+/** Byte offsets inside `post`. */
+const POST = {
+	italicAngle: 4,
+	size: 32
+};
+/** 16.16 fixed-point, as `post.italicAngle` and table versions are stored. */
+const FIXED_POINT_DIVISOR = 65536;
+/** One hmtx `longHorMetric`: advanceWidth (uint16) then lsb (int16). */
+const HMTX_METRIC_SIZE = 4;
+/** Byte offsets inside a glyph description header. */
+const GLYPH_HEADER = {
+	numberOfContours: 0,
+	xMin: 2,
+	yMin: 4,
+	xMax: 6,
+	yMax: 8,
+	size: 10
+};
+/**
+* Cap and x-height fallbacks for fonts without OS/2 version 2 whose 'H' and
+* 'x' cannot be measured (no cmap, no outline, or CFF outlines this reader
+* does not decode). Ratios of the em, rounded, so measurement stays sane.
+*/
+const CAP_HEIGHT_EM_RATIO = .7;
+const X_HEIGHT_EM_RATIO = .5;
+const CAP_HEIGHT_SAMPLE = 72;
+const X_HEIGHT_SAMPLE = 120;
+const NAME = {
+	count: 2,
+	stringOffset: 4,
+	recordsStart: 6,
+	recordSize: 12,
+	size: 6
+};
+const NAME_RECORD = {
+	platformId: 0,
+	encodingId: 2,
+	languageId: 4,
+	nameId: 6,
+	length: 8,
+	offset: 10
+};
+/** name id 6 is the PostScript name. */
+const POSTSCRIPT_NAME_ID = 6;
+const PLATFORM = {
+	unicode: 0,
+	macintosh: 1,
+	windows: 3
+};
+const WINDOWS_ENCODING = {
+	symbol: 0,
+	bmp: 1,
+	full: 10
+};
+const CMAP = {
+	numTables: 2,
+	recordsStart: 4,
+	recordSize: 8
+};
+const CMAP_RECORD = {
+	platformId: 0,
+	encodingId: 2,
+	offset: 4
+};
+const CMAP_FORMAT = {
+	/** Segment mapping to delta values: the BMP workhorse. */
+	segmentMapping: 4,
+	/** Trimmed table mapping: a single contiguous run. */
+	trimmedTable: 6,
+	/** Segmented coverage: the full Unicode range, uint32 code points. */
+	segmentedCoverage: 12
+};
+/** Byte offsets inside a format 4 subtable. */
+const FORMAT_4 = {
+	segCountX2: 6,
+	endCodes: 14,
+	headerSize: 14
+};
+/** Byte offsets inside a format 6 subtable. */
+const FORMAT_6 = {
+	firstCode: 6,
+	entryCount: 8,
+	glyphIds: 10,
+	headerSize: 10
+};
+/** Byte offsets inside a format 12 subtable. */
+const FORMAT_12 = {
+	numGroups: 12,
+	groups: 16,
+	headerSize: 16,
+	groupSize: 12
+};
+/**
+* Symbol fonts (platform 3, encoding 0) map their glyphs into the private use
+* area, so a lookup for 'A' has to be retried as `0xF000 | 'A'`.
+*/
+const SYMBOL_CODEPOINT_BASE = 61440;
+/**
+* Ceiling on entries materialized from a format 6/12 subtable. Glyph ids are
+* uint16, so an honest font cannot map more code points than this; a font that
+* claims to is lying about its group ranges and must not be allowed to
+* allocate against it.
+*/
+const MAX_CMAP_MAPPINGS = 1 << 18;
+const MAX_UNICODE_CODE_POINT = 1114111;
+/** Reads `count` big-endian uint16s into a typed array. Bounds pre-checked. */
+const readUint16Array = (view, offset, count) => {
+	const values = new Uint16Array(count);
+	for (let index = 0; index < count; index++) values[index] = view.getUint16(offset + index * 2);
+	return values;
+};
+const readInt16Array = (view, offset, count) => {
+	const values = new Int16Array(count);
+	for (let index = 0; index < count; index++) values[index] = view.getInt16(offset + index * 2);
+	return values;
+};
+/**
+* Format 4: four parallel segment arrays plus a trailing glyph id array that
+* segments address by a byte offset from their own `idRangeOffset` slot. The
+* arrays are read once here; lookup is a binary search over `endCodes`.
+*/
+const buildFormat4Lookup = ({ view, start, end }) => {
+	if (!fitsInView(view, start, FORMAT_4.headerSize) || start + FORMAT_4.headerSize > end) return;
+	const segCountX2 = view.getUint16(start + FORMAT_4.segCountX2);
+	const segCount = segCountX2 >> 1;
+	if (segCount === 0) return;
+	const endCodesBase = start + FORMAT_4.endCodes;
+	const startCodesBase = endCodesBase + segCountX2 + 2;
+	const idDeltasBase = startCodesBase + segCountX2;
+	const idRangeOffsetsBase = idDeltasBase + segCountX2;
+	const arraysSize = segCountX2 * 4 + 2;
+	if (!fitsInView(view, endCodesBase, arraysSize) || endCodesBase + arraysSize > end) return;
+	const endCodes = readUint16Array(view, endCodesBase, segCount);
+	const startCodes = readUint16Array(view, startCodesBase, segCount);
+	const idDeltas = readInt16Array(view, idDeltasBase, segCount);
+	const idRangeOffsets = readUint16Array(view, idRangeOffsetsBase, segCount);
+	return (codePoint) => {
+		if (codePoint < 0 || codePoint > 65535) return 0;
+		let low = 0;
+		let high = segCount - 1;
+		let segment = -1;
+		while (low <= high) {
+			const middle = low + high >> 1;
+			if ((endCodes[middle] ?? 0) < codePoint) low = middle + 1;
+			else {
+				segment = middle;
+				high = middle - 1;
+			}
+		}
+		if (segment < 0) return 0;
+		const segmentStart = startCodes[segment] ?? 0;
+		if (segmentStart > codePoint) return 0;
+		const delta = idDeltas[segment] ?? 0;
+		const rangeOffset = idRangeOffsets[segment] ?? 0;
+		if (rangeOffset === 0) return codePoint + delta & 65535;
+		const address = idRangeOffsetsBase + segment * 2 + rangeOffset + (codePoint - segmentStart) * 2;
+		if (!fitsInView(view, address, 2) || address + 2 > end) return 0;
+		const glyphId = view.getUint16(address);
+		return glyphId === 0 ? 0 : glyphId + delta & 65535;
+	};
+};
+/** Format 6: one contiguous run of glyph ids, materialized eagerly. */
+const buildFormat6Lookup = ({ view, start, end }) => {
+	if (!fitsInView(view, start, FORMAT_6.headerSize) || start + FORMAT_6.headerSize > end) return;
+	const firstCode = view.getUint16(start + FORMAT_6.firstCode);
+	const entryCount = view.getUint16(start + FORMAT_6.entryCount);
+	const glyphIdsBase = start + FORMAT_6.glyphIds;
+	if (!fitsInView(view, glyphIdsBase, entryCount * 2) || glyphIdsBase + entryCount * 2 > end) return;
+	const mappings = /* @__PURE__ */ new Map();
+	for (let index = 0; index < entryCount; index++) {
+		const glyphId = view.getUint16(glyphIdsBase + index * 2);
+		if (glyphId !== 0) mappings.set(firstCode + index, glyphId);
+	}
+	return (codePoint) => mappings.get(codePoint) ?? 0;
+};
+/**
+* Format 12: sorted `(startCharCode, endCharCode, startGlyphID)` groups. Every
+* group is expanded into the map up to {@link MAX_CMAP_MAPPINGS}, so lookup is
+* a hash hit rather than a per-character binary search.
+*/
+const buildFormat12Lookup = ({ view, start, end }) => {
+	if (!fitsInView(view, start, FORMAT_12.headerSize) || start + FORMAT_12.headerSize > end) return;
+	const numGroups = view.getUint32(start + FORMAT_12.numGroups);
+	const groupsBase = start + FORMAT_12.groups;
+	const groupsSize = numGroups * FORMAT_12.groupSize;
+	if (!fitsInView(view, groupsBase, groupsSize) || groupsBase + groupsSize > end) return;
+	const mappings = /* @__PURE__ */ new Map();
+	for (let index = 0; index < numGroups; index++) {
+		const group = groupsBase + index * FORMAT_12.groupSize;
+		const startCharCode = view.getUint32(group);
+		const endCharCode = view.getUint32(group + 4);
+		const startGlyphId = view.getUint32(group + 8);
+		if (endCharCode < startCharCode || endCharCode > MAX_UNICODE_CODE_POINT) return;
+		if (mappings.size + (endCharCode - startCharCode + 1) > MAX_CMAP_MAPPINGS) return;
+		for (let codePoint = startCharCode; codePoint <= endCharCode; codePoint++) {
+			const glyphId = startGlyphId + (codePoint - startCharCode);
+			if (glyphId !== 0) mappings.set(codePoint, glyphId);
+		}
+	}
+	return (codePoint) => mappings.get(codePoint) ?? 0;
+};
+const buildLookup = (subtable) => {
+	switch (subtable.format) {
+		case CMAP_FORMAT.segmentMapping: return buildFormat4Lookup(subtable);
+		case CMAP_FORMAT.trimmedTable: return buildFormat6Lookup(subtable);
+		case CMAP_FORMAT.segmentedCoverage: return buildFormat12Lookup(subtable);
+		default: return;
+	}
+};
+/**
+* Subtable preference, most specific first. Order is the whole contract here:
+* a full-coverage Windows table beats the BMP one, a Unicode-platform table
+* beats the symbol encoding, and the symbol encoding is the only one whose
+* lookup retries in the private use area.
+*/
+const CMAP_PREFERENCES = [
+	{
+		matches: (subtable) => subtable.platformId === PLATFORM.windows && subtable.encodingId === WINDOWS_ENCODING.full && subtable.format === CMAP_FORMAT.segmentedCoverage,
+		symbol: false
+	},
+	{
+		matches: (subtable) => subtable.platformId === PLATFORM.windows && subtable.encodingId === WINDOWS_ENCODING.bmp && subtable.format === CMAP_FORMAT.segmentMapping,
+		symbol: false
+	},
+	{
+		matches: (subtable) => subtable.platformId === PLATFORM.unicode && (subtable.format === CMAP_FORMAT.segmentedCoverage || subtable.format === CMAP_FORMAT.segmentMapping),
+		symbol: false
+	},
+	{
+		matches: (subtable) => subtable.platformId === PLATFORM.windows && subtable.encodingId === WINDOWS_ENCODING.symbol && subtable.format === CMAP_FORMAT.segmentMapping,
+		symbol: true
+	},
+	{
+		matches: (subtable) => subtable.platformId === PLATFORM.macintosh || subtable.platformId === PLATFORM.unicode,
+		symbol: false
+	}
+];
+const listCmapSubtables = (view, cmap) => {
+	const end = cmap.offset + cmap.length;
+	if (!fitsInView(view, cmap.offset, CMAP.recordsStart)) return [];
+	const numTables = view.getUint16(cmap.offset + CMAP.numTables);
+	const subtables = [];
+	for (let index = 0; index < numTables; index++) {
+		const record = cmap.offset + CMAP.recordsStart + index * CMAP.recordSize;
+		if (record + CMAP.recordSize > end || !fitsInView(view, record, CMAP.recordSize)) break;
+		const start = cmap.offset + view.getUint32(record + CMAP_RECORD.offset);
+		if (!fitsInView(view, start, 2) || start + 2 > end) continue;
+		subtables.push({
+			platformId: view.getUint16(record + CMAP_RECORD.platformId),
+			encodingId: view.getUint16(record + CMAP_RECORD.encodingId),
+			format: view.getUint16(start),
+			start,
+			end,
+			view
+		});
+	}
+	return subtables;
+};
+/**
+* Picks the best subtable the preference order admits and builds its lookup
+* once. A `cmap` that exists but yields no usable subtable is an error: the
+* alternative is a font that silently maps every character to .notdef.
+*/
+const buildCmapLookup = (view, cmap) => {
+	const subtables = listCmapSubtables(view, cmap);
+	for (const preference of CMAP_PREFERENCES) for (const subtable of subtables) {
+		if (!preference.matches(subtable)) continue;
+		const lookup = buildLookup(subtable);
+		if (!lookup) continue;
+		if (!preference.symbol) return Result.ok(lookup);
+		return Result.ok((codePoint) => {
+			const glyphId = lookup(codePoint);
+			return glyphId === 0 ? lookup(SYMBOL_CODEPOINT_BASE | codePoint) : glyphId;
+		});
+	}
+	const formats = subtables.map((subtable) => subtable.format).join(", ");
+	return fail(`cmap has no subtable this reader can use (found ${subtables.length} subtable(s), format(s): ${formats || "none"})`);
+};
+/** Strips the delimiters a PDF name cannot carry and caps the spec's length. */
+const MAX_POSTSCRIPT_NAME_LENGTH = 63;
+const POSTSCRIPT_NAME_DELIMITERS = /* @__PURE__ */ new Set("()<>[]{}/% ");
+const sanitizePostScriptName = (raw) => {
+	let name = "";
+	for (const character of raw) {
+		const code = character.codePointAt(0) ?? 0;
+		if (code > 32 && code < 127 && !POSTSCRIPT_NAME_DELIMITERS.has(character)) name += character;
+	}
+	return name.slice(0, MAX_POSTSCRIPT_NAME_LENGTH);
+};
+const decodeUtf16Be = (view, offset, length) => {
+	let text = "";
+	for (let index = 0; index + 1 < length; index += 2) text += String.fromCharCode(view.getUint16(offset + index));
+	return text;
+};
+const decodeMacRomanAscii = (view, offset, length) => {
+	let text = "";
+	for (let index = 0; index < length; index++) {
+		const code = view.getUint8(offset + index);
+		text += code < 128 ? String.fromCharCode(code) : "";
+	}
+	return text;
+};
+/**
+* name id 6, preferring the Windows UTF-16BE record over the Macintosh one.
+* Returns `""` when the table is absent, unreadable, or carries no usable
+* PostScript name; the caller substitutes a synthetic name.
+*/
+const readPostScriptName = (view, name) => {
+	const end = name.offset + name.length;
+	if (!fitsInView(view, name.offset, NAME.size)) return "";
+	const count = view.getUint16(name.offset + NAME.count);
+	const stringsBase = name.offset + view.getUint16(name.offset + NAME.stringOffset);
+	let windows = "";
+	let macintosh = "";
+	for (let index = 0; index < count; index++) {
+		const record = name.offset + NAME.recordsStart + index * NAME.recordSize;
+		if (record + NAME.recordSize > end || !fitsInView(view, record, NAME.recordSize)) break;
+		if (view.getUint16(record + NAME_RECORD.nameId) !== POSTSCRIPT_NAME_ID) continue;
+		const platformId = view.getUint16(record + NAME_RECORD.platformId);
+		const length = view.getUint16(record + NAME_RECORD.length);
+		const offset = stringsBase + view.getUint16(record + NAME_RECORD.offset);
+		if (!fitsInView(view, offset, length) || offset + length > end) continue;
+		if (platformId === PLATFORM.windows && windows === "") windows = decodeUtf16Be(view, offset, length);
+		else if (platformId === PLATFORM.macintosh && macintosh === "") macintosh = decodeMacRomanAscii(view, offset, length);
+	}
+	return sanitizePostScriptName(windows) || sanitizePostScriptName(macintosh);
+};
+/**
+* A stable stand-in when the font carries no PostScript name. Derived from
+* `head.checkSumAdjustment`, which is a function of the whole file, so the
+* same bytes always produce the same name and two different faces do not
+* collide in a PDF resource dictionary.
+*/
+const syntheticPostScriptName = (checkSumAdjustment) => `SfntFont-${checkSumAdjustment.toString(16).toUpperCase().padStart(8, "0")}`;
+/**
+* Builds the `glyphBoundsFor` accessor.
+*
+* A composite glyph carries its own bounds in the same header fields as a
+* simple one, so no component recursion is needed. A CFF font returns null
+* for every glyph: charstring interpretation is out of scope, and the caller
+* has to fall back on the hhea metrics.
+*/
+const buildGlyphBoundsReader = ({ directory, numGlyphs, longLoca }) => {
+	const loca = directory.tables.get(SFNT_TABLE.loca);
+	const glyf = directory.tables.get(SFNT_TABLE.glyf);
+	const offsets = loca && glyf ? readLocaOffsets({
+		view: directory.view,
+		loca,
+		numGlyphs,
+		longFormat: longLoca
+	}) : void 0;
+	if (!glyf || !offsets) return () => null;
+	const { view } = directory;
+	return (glyphId) => {
+		if (!Number.isInteger(glyphId) || glyphId < 0 || glyphId >= numGlyphs) return null;
+		const start = offsets[glyphId];
+		const end = offsets[glyphId + 1];
+		if (start === void 0 || end === void 0 || end - start < GLYPH_HEADER.size) return null;
+		const at = glyf.offset + start;
+		if (!fitsInView(view, at, GLYPH_HEADER.size) || start + GLYPH_HEADER.size > glyf.length) return null;
+		return {
+			xMin: view.getInt16(at + GLYPH_HEADER.xMin),
+			yMin: view.getInt16(at + GLYPH_HEADER.yMin),
+			xMax: view.getInt16(at + GLYPH_HEADER.xMax),
+			yMax: view.getInt16(at + GLYPH_HEADER.yMax)
+		};
+	};
+};
+/** Requires a table to exist and to be at least `size` bytes long. */
+const requireTable = (directory, tag, size) => {
+	const record = directory.tables.get(tag);
+	if (!record) return fail(`font has no '${tag}' table`);
+	if (record.length < size || !fitsInView(directory.view, record.offset, size)) return fail(`'${tag}' table is ${record.length} bytes, needs at least ${size}`);
+	return Result.ok(record);
+};
+const rejectContainer = (version) => {
+	switch (version) {
+		case SFNT_VERSION.collection: return fail("font collections ('ttcf') are not supported; extract a single face first");
+		case SFNT_VERSION.woff: return fail("this is a WOFF container; decode it with decodeWoff before parsing");
+		case SFNT_VERSION.woff2: return fail("this is a WOFF2 container; it must be decoded before parsing");
+		case SFNT_VERSION.trueType:
+		case SFNT_VERSION.trueMac:
+		case SFNT_VERSION.openTypeCff: return Result.ok();
+		default: return fail(`unrecognized sfnt version 0x${(version >>> 0).toString(16).padStart(8, "0")}`);
+	}
+};
+const parseSfnt = (bytes) => {
+	const version = readSfntVersion(bytes);
+	if (version === void 0) return fail(`buffer is ${bytes.byteLength} bytes, too short to be a font`);
+	const container = rejectContainer(version);
+	if (container.isErr()) return Result.err(container.error);
+	const directoryResult = readSfntDirectory(bytes).mapError((error) => new SfntParseError({ message: error.message }));
+	if (directoryResult.isErr()) return Result.err(directoryResult.error);
+	const directory = directoryResult.value;
+	const { view } = directory;
+	const headResult = requireTable(directory, SFNT_TABLE.head, HEAD.size);
+	if (headResult.isErr()) return Result.err(headResult.error);
+	const head = headResult.value;
+	if (view.getUint32(head.offset + HEAD.magicNumber) !== HEAD_MAGIC) return fail("'head' table has the wrong magic number");
+	const unitsPerEm = view.getUint16(head.offset + HEAD.unitsPerEm);
+	if (unitsPerEm < MIN_UNITS_PER_EM || unitsPerEm > MAX_UNITS_PER_EM) return fail(`head.unitsPerEm is ${unitsPerEm}, outside the permitted 16..16384`);
+	const hheaResult = requireTable(directory, SFNT_TABLE.hhea, HHEA.size);
+	if (hheaResult.isErr()) return Result.err(hheaResult.error);
+	const hhea = hheaResult.value;
+	const maxpResult = requireTable(directory, SFNT_TABLE.maxp, MAXP.size);
+	if (maxpResult.isErr()) return Result.err(maxpResult.error);
+	const numGlyphs = view.getUint16(maxpResult.value.offset + MAXP.numGlyphs);
+	if (numGlyphs === 0) return fail("maxp.numGlyphs is 0");
+	const numberOfHMetrics = view.getUint16(hhea.offset + HHEA.numberOfHMetrics);
+	if (numberOfHMetrics === 0 || numberOfHMetrics > numGlyphs) return fail(`hhea.numberOfHMetrics is ${numberOfHMetrics}, which is not in 1..${numGlyphs} (maxp.numGlyphs)`);
+	const hmtxResult = requireTable(directory, SFNT_TABLE.hmtx, numberOfHMetrics * HMTX_METRIC_SIZE);
+	if (hmtxResult.isErr()) return Result.err(hmtxResult.error);
+	const hmtx = hmtxResult.value;
+	const cmap = directory.tables.get(SFNT_TABLE.cmap);
+	const unmappedLookup = () => 0;
+	const lookupResult = cmap ? buildCmapLookup(view, cmap) : Result.ok(unmappedLookup);
+	if (lookupResult.isErr()) return Result.err(lookupResult.error);
+	const lookup = lookupResult.value;
+	const glyphIdFor = (codePoint) => {
+		if (!Number.isInteger(codePoint) || codePoint < 0 || codePoint > MAX_UNICODE_CODE_POINT) return 0;
+		const glyphId = lookup(codePoint);
+		return glyphId < numGlyphs ? glyphId : 0;
+	};
+	const advanceWidthFor = (glyphId) => {
+		if (!Number.isInteger(glyphId) || glyphId < 0) return 0;
+		const index = Math.min(glyphId, numberOfHMetrics - 1);
+		return view.getUint16(hmtx.offset + index * HMTX_METRIC_SIZE);
+	};
+	const longLoca = view.getInt16(head.offset + HEAD.indexToLocFormat) === 1;
+	const glyphBoundsFor = buildGlyphBoundsReader({
+		directory,
+		numGlyphs,
+		longLoca
+	});
+	const os2 = directory.tables.get(SFNT_TABLE.os2);
+	const os2HasHeights = os2 !== void 0 && os2.length >= OS2.sizeWithHeights && fitsInView(view, os2.offset, OS2.sizeWithHeights) && view.getUint16(os2.offset + OS2.version) >= OS2.minimumVersionWithHeights;
+	const capHeight = os2HasHeights ? view.getInt16(os2.offset + OS2.sCapHeight) : glyphBoundsFor(glyphIdFor(CAP_HEIGHT_SAMPLE))?.yMax ?? Math.round(unitsPerEm * CAP_HEIGHT_EM_RATIO);
+	const xHeight = os2HasHeights ? view.getInt16(os2.offset + OS2.sxHeight) : glyphBoundsFor(glyphIdFor(X_HEIGHT_SAMPLE))?.yMax ?? Math.round(unitsPerEm * X_HEIGHT_EM_RATIO);
+	const fsType = os2 !== void 0 && os2.length >= OS2.fsType + 2 && fitsInView(view, os2.offset, OS2.fsType + 2) ? view.getUint16(os2.offset + OS2.fsType) : 0;
+	const post = directory.tables.get(SFNT_TABLE.post);
+	const italicAngle = post !== void 0 && post.length >= POST.size && fitsInView(view, post.offset, POST.size) ? view.getInt32(post.offset + POST.italicAngle) / FIXED_POINT_DIVISOR : 0;
+	const name = directory.tables.get(SFNT_TABLE.name);
+	const postScriptName = (name ? readPostScriptName(view, name) : "") || syntheticPostScriptName(view.getUint32(head.offset + HEAD.checkSumAdjustment));
+	return Result.ok({
+		unitsPerEm,
+		ascender: view.getInt16(hhea.offset + HHEA.ascender),
+		descender: view.getInt16(hhea.offset + HHEA.descender),
+		lineGap: view.getInt16(hhea.offset + HHEA.lineGap),
+		capHeight,
+		xHeight,
+		italicAngle,
+		bbox: [
+			view.getInt16(head.offset + HEAD.xMin),
+			view.getInt16(head.offset + HEAD.yMin),
+			view.getInt16(head.offset + HEAD.xMax),
+			view.getInt16(head.offset + HEAD.yMax)
+		],
+		postScriptName,
+		isCff: directory.tables.has(SFNT_TABLE.cff),
+		numGlyphs,
+		fsType,
+		bytes,
+		glyphIdFor,
+		advanceWidthFor,
+		glyphBoundsFor
+	});
+};
+//#endregion
+export { SfntParseError, parseSfnt };

package/dist/fonts/sfnt/subset.d.ts

@@ -0,0 +1,20 @@
+import { SfntFont } from "./parse.js";
+import { Result } from "better-result";
+//#region src/fonts/sfnt/subset.d.ts
+declare const SubsetError_base: import("better-result").TaggedErrorClass<"SubsetError">;
+declare class SubsetError extends SubsetError_base<{
+  message: string;
+}> {}
+type TrueTypeSubset = {
+  /** Subset sfnt bytes. */
+  readonly bytes: Uint8Array;
+  /** Old glyph id -> new glyph id. Always maps 0 -> 0. */
+  readonly glyphIdMap: ReadonlyMap<number, number>;
+};
+/**
+ * Build a `glyf`-flavoured subset containing `glyphIds` plus the components
+ * they reference transitively, plus .notdef.
+ */
+declare const subsetTrueType: (font: SfntFont, glyphIds: ReadonlySet<number>) => Result<TrueTypeSubset, SubsetError>;
+//#endregion
+export { SubsetError, TrueTypeSubset, subsetTrueType };