[guocn]

Summary

Addresses the plural-possessive part of #58: the already checked-in expected failures for compare("dresses's", "dresses'") and compare_adjs(...) now pass as "p:p".

After removing the possessive suffix, both inputs have the same stem. _pl_check_plurals_N only handles different irregular plural forms, so it does not recognize this case. Accept identical stems when the existing singular_noun classifies the stem as plural. In compare(), check this possessive equivalence before the generic noun/verb paths can treat the extra s as an ordinary inflection.

The change deliberately does not add the undocumented "s:s" result, alter possessive generation, or fix the singular-name TODOs (dress's/dress', Jess's/Jess'). Consequently this PR does not close all of #58. It follows the existing plural-vs-singular classification rather than introducing a new linguistic rule: ambiguous names such as James remain subject to the existing singularizer's classification.

Verification

Linux x86_64, native isolated CPython 3.10.21 and 3.13.5:

• New public-API regression matrix: baseline 8 failed, 7 passed; patch 15 passed. Both argument orders are checked for the four alternative plural pairs.
• Existing dresses xfails promoted to ordinary assertions in both generic and adjective comparison tests; unrelated s:s xfails retained.
• Full pytest with Ruff/checkdocs/mypy/coverage enabled: 344 passed, 14 xfailed on each interpreter. Untouched baseline: 325 passed, 16 xfailed.
• Statement coverage 99% (2,372 statements, 14 missed including tests); not claiming branch coverage.
• Wheel and sdist build pass; Ruff check/format and git diff --check pass.
• Native Windows/macOS/PyPy/other interpreters and upstream CI not verified. Documentation build not rerun: no API signature/doc page changes; Towncrier fragment included.

Timing / draft

Eight fresh subprocess samples per version, interleaved baseline/head then head/baseline, 1,000 public calls per case. Microseconds/call (min / median / max):

case	baseline	patch
compare('cat','cats')	48.10 / 55.16 / 60.78	43.50 / 49.27 / 57.54
compare_adjs('my','our')	22.90 / 24.80 / 28.23	21.29 / 25.73 / 29.25
alternative plural possessives	172.08 / 184.82 / 235.44	164.56 / 172.49 / 198.93
unrelated plural possessives	156.04 / 164.95 / 198.33	149.53 / 162.90 / 232.52

Ranges overlap; this does not establish performance neutrality. Draft for maintainer review of recognizing existing plural stems before generic noun/verb comparison. Ordinary nonpossessive generic comparisons bypass the new helper call. No broader possessive-generation redesign is included.

AI disclosure

Prepared with AI assistance. Investigation, implementation, tests, local checks and this description are AI-assisted; no human review is claimed. A separate assistant will independently inspect the production diff and rerun focused regressions. The small fix restores an existing checked-in expected result rather than inventing a new comparison category.