iconvex_specs

Iconvex Specs
Copyright 2026 Iconvex contributors

Original Iconvex Specs library code is licensed under GNU Lesser General Public
License, version 2.1 or (at your option) any later version, matching GNU
libiconv's libraries. See LICENSE.

Unicode-hosted Adobe and Apple mapping files and ICU-derived ISCII fixtures are
distributed under the Unicode License V3 in LICENSE.UNICODE.

Apache-licensed upstream fixtures and source material retain Apache License 2.0.
See LICENSE.APACHE-2.0 and each source manifest. Their inclusion does not change
license of original Iconvex implementation.

The verbatim CTAN TeX'n'ANSI encoding vector retains LPPL-1.0-or-later terms,
and the verbatim CTAN cmap 1.0j OT1 and OT1tt files retain
LPPL-1.3c-or-later terms. Complete upstream license texts are shipped in
`licenses/upstream/LPPL-1.0.txt` and `licenses/upstream/LPPL-1.3c.txt`.

The pinned ICU 78.3 LMBCS algorithm, converter registry, and UCM mapping files
retain the Unicode/ICU license recorded in `LICENSE.UNICODE` and their source
metadata. They are immutable provenance and executable-oracle evidence; the
native Elixir implementation remains LGPL-2.1-or-later.

The pinned CPython 3.14.6 Punycode module is a repository-only executable test
oracle under Python Software Foundation License Version 2. Its complete license
and byte digest are retained in `priv/sources/rfc3492`; it is not linked, loaded,
or called by the original LGPL-2.1-or-later runtime implementation.

UTF-5 (`draft-jseng-utf5-01`) is implemented from the retained Internet-Draft,
SHA-256 `12ae18367c110b5dcef9cc3f06b6ae40e60c8fde489fdd161f1bb98e3e5f2375`.
The draft's authorship and unlimited-distribution statement remain intact; they
are not replaced by the LGPL-2.1-or-later license of the independently written
Elixir implementation.

UTF-6 (`draft-ietf-idn-utf6-00`) is implemented from the retained Internet-Draft,
SHA-256 `80033b5e41bc9f2fd01bddf99a300827b837f06ba93ef303bc54bc53df3755ca`.
Its Internet Society copyright, all-rights-reserved notice, and unlimited-
distribution statement are retained verbatim. The native Elixir implementation
is separate original LGPL-2.1-or-later code.

OpenJDK source-qualified codecs and research use four source sets from exact revision
`6ae23a0d6574dc8139aea93ea3c562a7410fcb34`. Those repository-only files
retain GPL-2.0-only terms and the Classpath Exception where their headers state
it. Every source directory contains the exact upstream `LICENSE`,
`ADDITIONAL_LICENSE_INFO`, URLs, and digests; all are excluded from Hex.

The OpenJDK quarantine resolves the LGPL release blocker by removing all seven
source-informed codecs, their importers, and generated runtime assets.
`x-eucJP-Open`, `x-MS950-HKSCS-XP`, `x-windows-50220`,
`x-windows-50221`, `x-windows-iso2022jp`,
`x-ISO-2022-CN-GB`, and `x-ISO-2022-CN-CNS` are not registered and are not
shipped. Their four upstream snapshots remain repository-only provenance under
their recorded GPL-2.0-only-with-Classpath terms. This is a factual packaging
and provenance disposition, not legal advice.

Three separate Unicode signature profiles are shipped only under neutral
`ICONVEX-` identities. Their newly authored LGPL-2.1-or-later wrapper delegates
scalar validation and serialization to Iconvex Core and adds the documented
signature/default-endian rules; it contains no vendor implementation source or
mapping table. The profiles are custom transports, not Unicode-standard
encoding schemes. Exact Unicode specification URLs and the release disposition
ship in `priv/sources/iconvex-unicode-signature-profiles/SOURCE_METADATA.md`.

CPython ISO-2022-JP-EXT sources at `v3.14.6` are repository-only evidence under
Python Software Foundation License Version 2. They share the exact license text
retained for the CPython Punycode oracle and are neither loaded nor called by
the runtime.

.NET x-Europa and x-cp50227 evidence at revision
`dbb2178288bb4e1e8f1fde3958be3bd75573c459` retains Microsoft's MIT terms.
The upstream code and binary table are repository-only; generated mapping facts
and the independently written native codecs do not change ownership of those
sources.

glibc IBM423 evidence retains its upstream ownership. `ibm423.c` expressly uses
LGPL-2.1-or-later and ships with the exact matching `COPYING.LIB`; the headerless
glibc charmap is conservatively classified as repository-only evidence rather
than being blanket-relicensed.

UTF-8-MAC / HFS Plus evidence combines separately owned material. Apple's
`libiconv_test.c` retains APSL-1.0 and the complete exact terms are vendored;
`citrus_utf8mac.c` retains BSD-2-Clause; archived TN1150 remains Copyright 2018
Apple Inc. All Rights Reserved; and Unicode 3.2 data retains Unicode License V3.
Those upstream files are repository-only and excluded from Hex. The generated
`priv/utf8_mac_manifest.etf` is shipped and contains decomposition keys and
precomposition pairs derived from the BSD-2-Clause C tables plus Unicode 3.2
data. Its complete attribution is shipped in `LICENSE.BSD-2-CLAUSE`. The APSL
test oracle and its license text do not enter the Hex artifact.

Microsoft Windows Best Fit source files retain the Unicode data-file terms
recorded in `LICENSE.UNICODE`. They are repository-only generator evidence;
the package contains generated mapping tables, not the original text files.

IANA registry-derived families retain the ownership of their IANA submissions,
RFC, Unicode data, and Artifex source evidence. The exact Amiga-1251 registry
submission, historical ISO-10646 inputs, and normalized PCL evidence are
repository-only; no absent upstream software license is inferred.

The IBM Unicode CCSID reference page remains IBM-copyrighted, repository-only
evidence and is excluded from Hex. No redistribution grant or LGPL coverage is
inferred for that page.

The Wang WISCII chart in document 800-1149-01 remains copyrighted vendor
documentation. No redistribution license was identified, the PDF is not
bundled, and no CC BY-SA or LGPL coverage is inferred for it. The separate
Wikipedia revision used to bind chart glyphs to Unicode is attributed to that
revision's contributors under CC BY-SA 4.0; its prose and wikitext are not
bundled.

The C-DAC GIST PASCII Version 1.0 PDF remains a copyrighted reference with no
redistribution grant and is excluded from the repository package and Hex
artifact. The package contains only an independently expressed factual byte
table, explicit Iconvex projection policies, digest-bearing metadata, and the
original LGPL-2.1-or-later native Elixir implementation. No ownership or
license is inferred for the source PDF.

These audited artifacts are not LGPL-covered upstream material. Their presence
for repeatable source review does not alter the LGPL-2.1-or-later license of
original Iconvex code; complete file-level dispositions are in `SOURCES.md` and
each source directory's `SOURCE_METADATA.md`.

The pinned IBM and Control Data punched-card manuals, University of Iowa
reconstruction page, and Unicode L2/15-083R submission retain their respective
copyright and submission terms. They are repository-only evidence and are not
packaged as LGPL-covered runtime code. The independently implemented mapping
facts and native Elixir engines remain LGPL-2.1-or-later.

The IBM corporate GCGID registry pages and Digital Equipment Corporation's
1994 printer programming manual used for CP310/907/1116/1117/1287/1288 retain
their original notices and are repository-only provenance. The pinned ICU UCM
join inputs retain the Unicode/ICU data-file terms. IBM/tnz's exact CP310 source
table retains Apache-2.0. Hex artifacts ship only independently generated
256-row mapping facts plus digest-bearing metadata; the pure Elixir runtime and
generator remain LGPL-2.1-or-later.

The generated historical ICU archive tables are distributed across three
transparent `iconvex_specs_icu_archive_*` provider packages. That storage split
does not alter mappings, public codec names, or upstream license terms.

MARC-8 code tables are published by the Library of Congress. TSCII and BRF
charmaps are sourced from the GNU C Library. Every source URL and byte digest is
recorded in SOURCES.md and the generated manifests under priv.

Archival ECMA, Digital Equipment Corporation, and Control Data Corporation
manual scans retain their respective copyright notices. They are included only
as pinned normative test fixtures for ECMA-1, DEC SIXBIT, DEC Special Graphic,
DEC Technical, SI 960, DEC Hebrew 8-bit, the three distinct DEC RADIX-50 word
formats, and CDC Display Code; exact URLs, cited pages, and SHA-256 digests are
recorded in SOURCES.md.

The Sperry Rand UP-7824 Rev. 1 and UP-7604 Rev. 1 manual scans used to define
the UNIVAC 1100 Series and 4009 Display Console FIELDATA profiles retain their
original 1971/1974 and 1968/1974 copyright notices and provide no permissive
redistribution grant. The scans are not redistributed. Hex artifacts contain
only independently transcribed factual tables, digest-bearing source metadata,
and the original LGPL-2.1-or-later native Elixir implementation.

The pinned Kermit `ckcuni.c` tables for DEC, Data General, HP, Siemens
Nixdorf, QNX, Mazovia, Bulgarian PC, and Hebrew-7 profiles are used as exact
source-versioned mappings or independent historical cross-checks and retain
their BSD-3-Clause terms in
`priv/sources/dec-terminal-character-sets/kermit/COPYING`. Its inclusion does
not change the LGPL-2.1-or-later license of the Iconvex implementation.

The complete KOI8 Unified Cyrillic 2.1 mapping used by `KOI8-F` is copyright
2008 Department of Mathematical Sciences, New Mexico State University, and is
distributed under its MIT-style permission notice. The notice is retained in
`priv/sources/koi8-f/KOI8UNI.TXT` and reproduced in `LICENSE.MIT-NMSU`. The
native Elixir codec implementation remains LGPL-2.1-or-later.

The compact Unihan 17.0.0 `kMainlandTelegraph`, `kTaiwanTelegraph`, and `kGB3`
mapping facts are derived from Unicode data and retain the Unicode License V3
in `LICENSE.UNICODE`. The original archive is not packaged. The packaged
`kGB3` CSV records the exact provisional property values; its decimal-token
mapping and raw GL row/cell projection are source-qualified rather than aliases
for a published GB 13131 transport. Iconvex's property-token engines, GL codec,
and explicit Taiwan reverse/lossless policies remain original
LGPL-2.1-or-later implementation code.