Bug description

>>> '鬝'.encode('iso2022_jp_3')
b'\x1b$(P};\x1b(B'
>>> _.decode('iso2022_jp_3')
'鬜'
>>> hex(ord(_))
'0x9b1c'

JIS X 0213:2004 changed plane 2 row 93 cell 27 (0x7D3B) from U+9B1D to U+9B1C, and CPython emulates the 2000 mapping in the codecs named after that edition. euc_jisx0213 and shift_jisx0213 do so in both directions:

>>> '鬝'.encode('euc_jisx0213').decode('euc_jisx0213')
'鬝'
>>> '鬜'.encode('euc_jisx0213')
Traceback (most recent call last):
  File "<python-input-5>", line 1, in <module>
    '鬜'.encode('euc_jisx0213')
    ~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
UnicodeEncodeError: 'euc_jisx0213' codec can't encode character '鬜' in position 0: illegal multibyte sequence

iso2022_jp_3 only does it when encoding: it accepts U+9B1D and rejects U+9B1C like the other 2000 codecs, but decodes 0x7D3B as U+9B1C, as iso2022_jp_2004 does.

The cause is in jisx0213_2000_2_decoder, Modules/cjkcodecs/_codecs_iso2022.c lines 728-739. The EMULATE_JISX0213_2000_DECODE_PLANE2_CHAR macro expands to an if that sets u = 0x9B1D for this cell, but the table lookup that follows it is a plain if (TRYMAP_DEC_ST(jisx0213_2_bmp, ...)) rather than else if, so it runs regardless and overwrites u with the 2004 value. The plane 1 decoder just above uses else if after its emulation macro and is fine, as are the EUC and Shift_JIS codecs, which use the OUTCHAR form of the macro in _codecs_jp.c. So the fix is the missing else.

A scan of every code point through iso2022_jp_3 found no other character that comes back different, so it should be just this one cell.

Found by the Hypothesis tests in #160155, where it is pinned as RoundTripBugs.test_iso2022_jp_3_mapping under expectedFailure and kept out of the generated text otherwise.

CPython versions tested on

CPython main (6e1bd24)

Operating systems tested on

Linux

Linked PRs

• gh-160306: Fix iso2022_jp_3 decoding of the JIS X 0213:2000 plane 2 mapping #160702