Iconvex Platforms

iconvex_platforms adds 117 pure-Elixir, vendor- and platform-qualified codecs to Iconvex. The package contributes 354 unique registry names: 117 canonicals and 237 aliases.

There are no NIFs, ports, subprocesses, or runtime calls to iconv. Installing the OTP application registers the complete surface and its tables in one transaction; stopping it removes exactly those routes and providers.

Installation

def deps do
  [
    {:iconvex, "~> 0.1"},
    {:iconvex_platforms, "~> 0.1"}
  ]
end

A few fun conversions

IBM-1175 is the Turkish EBCDIC euro-and-lira revision:

iex> Iconvex.convert("TRY ₺€", "UTF-8", "IBM-1175")
{:ok, <<0xE3, 0xD9, 0xE8, 0x40, 0x9A, 0x9F>>}

GNU glibc's BRF map turns Braille ASCII cells into Unicode Braille:

iex> Iconvex.convert("AB", "BRF", "UTF-8")
{:ok, "⠁⠃"}

VietUnicode's four VNI transports are kept distinct. This is the 2002 ANSI Win/Unix profile:

iex> Iconvex.convert("Việt", "UTF-8", "VIETUNICODE-2002-VNI-ANSI-WIN-UNIX")
{:ok, <<0x56, 0x69, 0x65, 0xE4, 0x74>>}

Windows best-fit mappings are intentionally directional:

iex> Iconvex.convert("−", "UTF-8", "WINDOWS-BESTFIT-1252")
{:ok, "-"}
iex> Iconvex.convert("-", "WINDOWS-BESTFIT-1252", "UTF-8")
{:ok, "-"}

Java Modified UTF-8 rejects a literal zero byte and accepts the two-byte NUL:

iex> Iconvex.convert(<<0xC0, 0x80>>, "JAVA-MODIFIED-UTF-8", "UTF-8")
{:ok, <<0>>}

The stateful IBM-965 and IBM-17354 codecs emit designation and shift sequences, while UTF-8-MAC implements the HFS Plus decomposition profile:

iex> {:ok, encoded} = Iconvex.convert("A一", "UTF-8", "IBM-965")
iex> Iconvex.convert(encoded, "IBM-965", "UTF-8")
{:ok, "A一"}

iex> Iconvex.convert("é", "UTF-8", "UTF-8-MAC")
{:ok, "é"}

What is included

Use Iconvex.Platforms.encodings/0 for the canonical list and Iconvex.Platforms.registrations/0 for canonicals, aliases, modules, and source families. The immutable row-by-row contract is SURFACE_MANIFEST.tsv.

Archive-independent IBM tables

IBM-1175, IBM-965, and IBM-17354 were originally routed through ICU Archive provider IDs. IBM-934/938 and IBM-5052/5053/5055/958 also had seven hidden Archive table dependencies. Platforms copies all ten immutable tables under package-owned IDs. It neither depends on iconvex_icu_archive nor claims any Archive provider, so both optional packages can be installed without provider ownership collisions.

In total the package owns 84 tables: 73 declared manifest providers, ten relocated IBM dependencies, and the pinned glibc IBM423 table.

Verification

The test suite carries the original source fixtures and a 1,112,064-scalar UTF-32BE corpus. It verifies every row and canonical inverse of the generated mapping families, all 94×94 positions for IBM-965 and IBM-17354, every byte/inverse for the single-byte profiles, stateful boundaries, substitution and discard policies, malformed UTF-8 offsets, lifecycle rollback/restart, artifact selection, source digests, and license selection.

Representative source-bound benchmarks enforce a 30× native/reference ceiling and linear reduction scaling. See BENCHMARKS.md.

Extending Iconvex

Third-party packages can use Iconvex.Extension.register/2 to add codecs and table providers transactionally. Platforms itself is a leaf package: its only runtime dependency is Iconvex Core and its runtime contains no references to the former monolithic Specs namespace.

Original Iconvex code is LGPL-2.1-or-later. Retained mapping/source artifacts keep their upstream terms; see NOTICE and the shipped license files.