Skip to content

fix(cinbase): stop the 倉頡 siblings offering 㤘 under 手 codes (#390) - #1023

Merged
PCMan merged 1 commit into
EasyIME:masterfrom
asdf80730:pr/cj-siblings-qnsm-390
Nov 6, 2026
Merged

PCMan merged 1 commit into
EasyIME:masterfrom
asdf80730:pr/cj-siblings-qnsm-390

Conversation

@asdf80730

@asdf80730 asdf80730 commented Oct 30, 2026 •

Copy link
Copy Markdown
Contributor

Follow-up to #390 and #1020: the 酷倉 sweep found the same inherited
qnsm 㤘 line still shipping in the tables of the same 第五代倉頡
lineage — 倉五, 泰瑞倉頡 and 亂倉打鳥. 㤘 (U+3918) sits under Unihan
radical 61 (心, +5 strokes) and the CNS official 倉頡 table
(cin/cnscj.cin) codes it 心弓尸一 (pnsm) only, so the 忄-radical
character does not belong under the 手 root key in any table.

  • cin/cj5.cin: move 㤘 from qnsm (its only code there) to pnsm,
    the CNS code 心弓尸一.
  • cin/thcj.cin, cin/newcj3.cin: drop the qnsm copy; both tables
    already carry 㤘 correctly at pnsm.
  • cin/simplex5.cin (簡五): move qm to pm — the two-key scheme
    files a character by its full code's first and last key, and 㤘's
    full code 心弓尸一 derives to pm, where its 忄 siblings
    (怛, 恒, 性…) sit.
  • json/{cj5,thcj,newcj3,simplex5}.json: regenerated with
    python/cinbase/tools/cintojson.py (unsorted path, so candidate
    order stays file order). cj5, thcj and newcj3 lose the qnsm key
    (their qnsm bucket held only 㤘); thcj and newcj3 counters drop by
    one (cjkExtA 7451→7450 / 4523→4522, totalchardefs 85865→85864 /
    140615→140614), while cj5's and simplex5's stay unchanged because
    the character keeps its one entry.

Sweep for other 忄/扌/月 confusables, per table: every single-glyph
chardef line whose code starts with p/q/b was checked against the
character's Unihan kRSUnicode radical (cj5 3123 of 32696 lines, thcj
8669 of 85864, newcj3 6981 of 140655, simplex5 3123 of 32696). 17 /
59 / 23 / 17 pairs flagged. Every flag but 㤘 is a 心/手-at-the-bottom
decomposition whose first key legitimately comes from the top
component — 愛 bbpe, 慧 qjsmp, 惷 qap… — each matching the codes
that gcin cj5.cin and/or cnscj.cin assign. The one flag neither
reference covers, thcj's 𪬇 qjph (U+2AB07), is composed ⿱⿰丰丰必:
丰 opens 手十 and 必 is 心竹, so qjph spells its own components and
the q first key is legitimate. 㤘 was the only genuine left-radical
mismapping; gcin's 第五代 cj5.cin carries qnsm only for it — the
字典 "Cangjie QNSM" special case cited on #390, contradicted by CNS
official codes and Unihan.

Validation:

  • tests/verify_windows.sh on the integration tree carrying these
    exact table files: 25 passed, 0 failed — including the
    committed-pair regeneration test (each committed json/*.json
    reproduces converter output byte-for-byte) on the shipped
    python/python3/python.exe.
  • Runtime-JSON pins for these four tables (the 手-key code offers no
    㤘, the 忄 code still does) failed 6 ways on the unfixed tree and
    pass with these files.

Architecture and design by asdf80730. Implementation generated with AI assistance from omp (tensorfold/Qwen3.8-Flash-Next).

)

The 酷倉 sweep of ticket 08 found the same zidian-inherited `qnsm 㤘`
line shipping in the tables of the same 第五代倉頡 lineage. 㤘 (U+3918)
sits under Unihan radical 61 (心, +5 strokes) and the CNS official
倉頡 table (cnscj.cin) codes it 心弓尸一 only, so the character does not
belong under the 手 root key in any table:

- cin/cj5.cin: move the character from `qnsm` (its only code there) to
  `pnsm`, the CNS code 心弓尸一.
- cin/thcj.cin, cin/newcj3.cin: drop the `qnsm` copy; both tables
  already carry 㤘 correctly at `pnsm`.
- cin/simplex5.cin: move `qm` to `pm` — the two-key scheme files a
  character by its full code's first and last key, and the full code
  `pnsm` derives to `pm`, where its 忄 siblings (怛, 恒, 性…) sit.
- json/{cj5,thcj,newcj3,simplex5}.json: regenerated with
  python/cinbase/tools/cintojson.py (unsorted path, so candidate order
  stays file order). thcj and newcj3 lose the `qnsm` key
  (cjkExtA 7451→7450 / 4523→4522, totalchardefs 85865→85864 /
  140615→140614); cj5 and simplex5 only relocate the character between
  buckets, counters unchanged.

Sweep of each table for the 忄/扌/月 confusable family (first key p/q/b
checked against the character's Unihan kRSUnicode radical): every flag
but 㤘 is a 心/手-at-the-bottom decomposition whose first key
legitimately comes from the top component, corroborated by gcin
cj5.cin and/or cnscj.cin — 愛 `bbpe`, 慧 `qjsmp`, 惷 `qap`…. The one
flag neither reference covers, thcj's 𪬇 `qjph` (U+2AB07), is composed
⿱⿰丰丰必 per the dictionary, and 丰 opens with its 五代 code 手十 —
the q first key is legitimate too.

Architecture and design by asdf80730. Implementation generated with AI
assistance from omp (tensorfold/Qwen3.8-Flash-Next).
@asdf80730 asdf80730 mentioned this pull request Oct 30, 2026
@PCMan
PCMan merged commit e5e249b into EasyIME:master Nov 6, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants