[JLHwung]

Disclaimer: This PR is created with the help of LLM. The PR description is generated by LLM and edited by myself.

Summary

Several parsers omit implicit values, misinterpret historical formats, or generate incomplete property data. This PR corrects those behaviors across all 35 supported Unicode releases, from 1.1.5 to 18.0.0. It also adds the UCD files that older releases published but the resource list never included.

• Apply version-specific defaults for bidi classes, scripts, joining types, blocks, and line breaks.
• Generate algorithmic character names and Hangul decompositions, and restore omitted historical Han and private-use ranges.
• Preserve Indic_Conjunct_Break as an enumerated property, and exclude mapping and Quick_Check records from binary properties.
• Handle comments, whitespace, line endings, aliases, and legacy range formats correctly.
• Add missing historical resources, and preserve one-way bidi mirroring mappings.
• Correct the generated TypeScript declarations for mapping values and default exports.

Historical workarounds are restricted to the versions that require them, so parsing current releases skips them. Each commit is one self-contained fix with its own regression test. Reviewing commit by commit should be easier than reviewing the combined diff.

Changes to current output

Most fixes affect historical releases only, but some change the generated data for current versions too. I generated the Unicode 18.0.0 package from main and from this branch. These are the only differences:

Area	main	This branch
Bidi_Class	801,723 unlisted code points have no value	Unlisted code points get the @missing defaults from DerivedBidiClass.txt: L, R, AL, BN, or ET
Decomposition_Mapping/Canonical	2,081 mappings, with no Hangul syllables	13,253 mappings, adding the 11,172 algorithmic Hangul decompositions
Names	Range placeholders such as CJK Ideograph Extension A, plus surrogates and private use	Formal names such as CJK UNIFIED IDEOGRAPH-3400; code points without a Name are omitted
Block	No No_Block value	No_Block covers the gaps between blocks
Indic_Conjunct_Break	Generated as a Binary_Property/InCB set	Enumerated property with the values Consonant, Extend, Linker, and None
Binary_Property/NFKC_Simple_Casefold	Generated as a binary property	Excluded, because it's a mapping
index.d.mts	Declares named exports that index.mjs doesn't provide	Declares the default export that index.mjs has

Historical fixes

Defaults

• Before Unicode 5.0, Script defaults to Common, because Unknown didn't exist yet.
• Unicode 3.x applies the UAX #‌9 Bidi_Class defaults to unassigned code points.
• Releases with only ArabicShaping.txt derive implicit Transparent as Mn + Cf minus the joining controls. ZERO WIDTH JOINER counts as Join_Causing.
• Before Unicode 4.0, unlisted assigned code points get Line_Break AL, not XX.
• From Unicode 4.0 on, an empty simple titlecase mapping falls back to the uppercase mapping. Earlier releases keep the identity mapping of titlecase digraphs.

Aliases and names

• Historical spellings are canonicalized, such as White_space, STerm, and uppercase script names like LATIN.
• The Quick_Check records NFC_NO and NFC_MAYBE and the FNC alias from Unicode 3.1.0 through 4.0.0 are no longer binary properties.
• Two-field NameAliases.txt records from before Unicode 6.1 are treated as corrections.
• Trailing spaces in two Unicode 4.0.0 names, for U+02AE and U+0615, are trimmed.

Omitted ranges and legacy formats

• Unicode 1.1.5 lists U+4E00 only as a "representative" record and omits the compatibility ideographs. Following CJKXREF.TXT, 4E00..9FA5 and F900..FA2D are now assigned and named.
• Unicode 2.0 through 3.0.0 omit the plane 15 and 16 private-use characters that their private-use high surrogates encode. They're now restored as Private_Use, matching the Unicode 3.0.1 documentation.
• Old PropList files keep six-digit ranges and noncharacter records, and the 2.1.8 "Discrepancy" sections are skipped.
• Discontiguous legacy blocks are accumulated. Where the Unicode 2.0 blocks file overlaps at U+FEFF, the narrower Specials entry wins, as in the corrected 2.1.9 file.
• The Unicode 3.0 LineBreak.txt lists ranges as First and Last records, which are now expanded.

Bidi mirroring

The writer assumed every mapping had a matching reverse mapping. The 3.2 and 4.0 data contain two that don't, U+2A2D→U+2A2C and U+2A2E→U+2A2D, which it would have dropped. One-way mappings are now preserved. Output for symmetric data, including every current release, is byte-for-byte unchanged.

Missing resources added

A release that didn't republish a file reuses the previous release's copy, as 3.0.1 already did for Blocks.

Type	Versions added
derived-bidi-class	3.2.0 to 18.0.0
name-aliases	5.0.0 to 6.0.0
arabic-shaping	2.0.14 to 3.1.1
derived-joining-type	3.1.1 to 4.0.1
composition-exclusions	3.0.0 to 4.0.1
bidi-mirroring	3.0.1 to 4.0.1
line-break	3.0.0, 3.0.1
properties	2.1.5, reusing PropList-2.1.2

Unicode 3.1.1 now uses the published PropList-3.1.1.txt instead of the 3.1.0 file. That corrects Hex_Digit and adds ASCII_Hex_Digit.

Notes for reviewers

• Version guards. The new utils.isVersionBefore helper restricts each historical workaround to the releases that require it. Adding the guards left every parser's output identical for every release.
• CJKXREF.TXT is not vendored. Its license forbids redistribution, and this package publishes data/. The code links to it as the source for the 1.1.5 ranges instead.
• DerivedJoiningType-3.1.0 is not used. It omits Transparent and ZERO WIDTH JOINER, so 3.1.0 derives Joining_Type from ArabicShaping instead. From 3.1.1 to 4.0.1, the derived files and the ArabicShaping result match exactly.

[mathiasbynens]

Woah! Great set of fixes. Thanks!