tableCmap · @awacloud/fonts

Table cmap — Character to Glyph Index Mapping (OT §6.4.1).

Module tableCmap | Source packages/front/office/fonts/src/table/cmap.js | Deps fontErrors, fontReader, tableCmapFormats | Worker-safe yes

Parses the header + every encodingRecords entry and its subtable. Formats decoded into Map<codePoint, glyphID>:

Format 0 — 256-byte direct array (Mac Roman legacy).
Format 2 — high-byte mapping (legacy CJK).
Format 4 — segment mapping to delta values (Windows BMP).
Format 6 — trimmed table mapping.
Format 12 — segmented coverage (Unicode full repertoire, uint32).
Format 13 — many-to-one segmented coverage.
Format 14 — Unicode Variation Sequences (defaultUVS + nonDefaultUVS).

Other formats (including format 10) are recognised at metadata level only ({ format, parsed: false, length }) and left unparsed.

Parser hardening

fonts/cmap-range-bomb — Formats 12 and 13 declared startCharCode/endCharCode as uint32. A malicious font could declare (0, 0xFFFFFFFF) → a loop of 4 billion entries (OOM/hang). Each group is now bounded to the Unicode domain [0..0x10FFFF] and the materialised total is capped at 2 × 0x110000. Any violation throws ParseError('fonts/cmap-range-bomb').
Format 14 — independent sub-readers — parseFormat14 used to rely on r.peek(rr => { rr.seek(...); return rr; }) to obtain its sub-tables. peek restores the parent cursor on return, so every subsequent sub.readUint* call was reading from the pre-peek position. defaultUVS / nonDefaultUVS were therefore silently corrupted. The code now uses r.sub(rel, length), which returns an independent reader. Locked in by two regression tests in tests/fuzz.test.js.
fonts/cmap-too-many-subtables — cmap.numTables ≤ 64.

Resolve

const { parseCmap, pickUnicodeMap } = runtime.resolve('tableCmap');

API

Method	Signature	Returns
`parseCmap`	`(bytes: Uint8Array) => { version, encodings: Array }`	One `encodings[i]` entry per directory record.
`pickUnicodeMap`	`(cmap) => Map<number, number> \| null`	Selects the best Unicode subtable.

`pickUnicodeMap` preference order

Windows / Unicode full repertoire (3, 10)
Unicode platform / Unicode 2.0 full (0, 4)
Windows / Unicode BMP (3, 1)
Unicode platform / any
Any subtable exposing .map

Shape of a record

{
    platformID, encodingID, subtableOffset,
    subtable: { format, length, language, map: Map<cp, gid>, ... }
}

Examples

Parse + pick

const { parseCmap, pickUnicodeMap } = runtime.resolve('tableCmap');
const cmap = parseCmap(sfnt.tables.cmap.bytes);
const unicode = pickUnicodeMap(cmap);
const gidA = unicode.get(0x41);

Notes

An out-of-bounds subtable offset throws ParseError('fonts/cmap-bad-offset').
Format 4 implements the OT pointer arithmetic exactly (idRangeOffset[i]/2 + (c - startCode[i]) + &idRangeOffset[i]).
Format 12 uses uint32 throughout — supports codepoints beyond the BMP — always under the Unicode cap (see "Parser hardening").
See errors for the complete list of fonts/cmap-* error codes thrown.