t61codec 3.0.0

The codec now follows ITU-T T.61 (11/1988) to the letter. This is the
final feature release: the project is in maintenance mode (bug fixes,
support for new Python versions and security fixes only).

Breaking changes (spec conformance, #4)

Decoding:

• A diacritic (0xC1-0xCF) followed by a base letter now yields the
  precomposed character (b"Caf\xc2e" decodes to "Café"). Previously
  a standalone combining mark was emitted before the letter.
• Only the 155 combinations of Figure B-1 are valid. Anything else
  raises, for example C7 69, CB 67, C2 62 or a diacritic before a
  control.
• C2 67 decodes to ǵ (U+01F5).
• A diacritic followed by SPACE yields the spacing accent (b"\xc2 " is
  "´", b"\xc4 " is "~").
• A trailing diacritic or underline raises "unexpected end of data". It
  used to be emitted as a standalone mark.
• 0xCE is the ogonek (b"\xcea" is "ą"); it used to give U+032B.
• 0xCC (non-spacing underline) yields the base character plus U+0332. It
  works over accented letters and SPACE, with PLD/PLU allowed in
  between.
• 0xC9 is now decoded as an alias of 0xC8 (it used to be an error).
• 0xE0 decodes to U+03A9 (was U+2126) and 0xE2 to U+0110 (was U+00D0).
• 0x23 and 0x24 decode to # and ¤ (they used to be errors).
• 0xA0 (was NBSP) and 0x7F (was DEL) now raise.
• All C0/C1 controls except BS, LF, FF, CR, SUB, PLD, PLU and CSI now
  raise. This includes HT, NUL, VT, ESC and 0x80-0x9F apart from 0x8B,
  0x8C and 0x9B.
• Every decode error spans exactly one byte and decoding resumes at the
  next byte.
• The incremental decoder and StreamReader buffer a trailing prefix
  across chunks. StreamReader.read() raises for a prefix left at EOF.

Encoding:

• Input is NFC-normalised, so precomposed and decomposed text both
  encode to b"\xc2e". Previously decomposed input came out in the
  wrong order and precomposed input failed.
• Accented letters are encoded as diacritic + base letter. Base + U+0332
  is encoded with 0xCC first.
• U+03A9 and U+2126 both encode to 0xE0; U+0110 and U+00D0 both encode
  to 0xE2.
• # encodes to 0xA6 and ¤ to 0xA8. 0x23, 0x24 and 0xC9 are never
  emitted.
• These now raise: SUB, HT, NUL, DEL, NBSP, a leading or standalone
  combining mark and combinations not listed in Figure B-1 (such as
  "q\u0301").
• backslashreplace raises a clear error when encoding, since \ is
  not part of T.61.
• IncrementalEncoder and StreamWriter: a combining mark must arrive
  in the same chunk as its base letter.

Public API:

• t61codec.DECODING_TABLE and t61codec.ENCODING_TABLE were removed
  (the tables are now private, in t61codec._tables).
• StreamReader.decode gained final: bool = False.
• IncrementalDecoder now subclasses
  codecs.BufferedIncrementalDecoder.

Fixes

• search_function returns None for unknown codec names (#3). Once
  registered, it used to raise RecursionError (2.0.0) or
  AttributeError (1.0.1).

Packaging

• Requires Python 3.11 or newer (2.0.0 required 3.6).
• The package is typed: it ships py.typed and passes mypy --strict.
• PEP 639 license metadata. Development Status :: 6 - Mature: the
  project is in maintenance mode.
• src layout, uv, ruff and Sphinx documentation on Read the Docs.
• Releases are published through PyPI Trusted Publishing.

---

Upgrading from 2.x: decoding of accented text and the set of accepted bytes change; see the changelog above. Data produced by OpenSSL (T61String treated as Latin-1) may not be valid T.61 and will now raise; decode it with "latin-1" if needed.