Part a trailing suffix part at a declared suffix delimiter as a comma would (#549)

[derek73]

Closes #549.

What changes

A delimiter declared through Policy(extra_suffix_delimiters=...) or HumanName(suffix_delimiter=...) now splits a trailing suffix part exactly where a typed comma would. group() cuts a tail segment at each delimiter before any grouping, groups each piece the way the comma version's own segment would be grouped, and drops the delimiter. No join, maiden walk or link search can reach across a delimiter, because none of them ever sees one.

This reverses #538's "skip" reading, where the connective search stepped over a delimiter as if it weren't written. The skip reading was the source of #549: a delimiter standing alone was dropped like a comma, while one next to a connective was merged into the joined phrase like a word.

Because delimiters never reach grouping, the machinery that stepped over them is deleted:

• the cores parameter on four functions;
• _maiden_take's seen index list;
• the skip arguments of peel_walk and trailing_start;
• the old Apply suffix_delimiter only at suffix-consumption sites #206 drop block that ran after the joins.

The result is one call per parse fewer (369 vs 370, tools/perf/call_count.py, py3.11).

Behavior, with suffix_delimiter=" - " (each checked against the released wheels)

input	1.4.0	2.0–2.3	this PR
Smith, John, PhD - and MD	suffix PhD, and MD	PhD - and MD	PhD, and MD
Smith, John, Puig - y Soler	Puig, y Soler	Puig - y Soler	Puig, y Soler
Smith, John, PhD née Puig - i Soler	(no maiden markers)	maiden Puig (2.3.0)	maiden Puig, suffix PhD, i Soler
Smith, John, MD - née Jones Smith	suffix MD - née Jones Smith	maiden Jones Smith	suffix MD, née Jones Smith, no maiden

The last row is an accepted cost (decided 2026-10-03). A maiden marker with the delimiter straight before or after it takes no clause, as it takes none after a typed comma. By count it is the larger move: every one of the 240 maiden movers in the generated set loses the maiden. Recorded as Accepted in rules.md#C1, in bold in the release note, and pinned by the case row suffix_delimiter_core_after_a_marker_takes_no_clause.

At the default policy nothing changes, since no delimiter is declared. Outside a trailing suffix part (before the first comma, or in the given part of Family, Given), the delimiter is still a word, as in 1.4.0.

Measurements (2026-10-03, branch vs 061f02d)

The full figures and the recompute recipe are in the new decisions.md#C1 entry.

• Generated set (19,972 texts):
  • - delimiter: 2,624 parses move (suffix only 2,384; suffix + maiden 240). No other field moves and no ambiguity report moves.
  • / delimiter: the same 2,624 parses.
  • and delimiter: 2,528 parses.
• Typed-comma comparison, role fields only. Before → after:
  • Smith, John, and Smith, John, Jr., heads: 752 of 2,100 disagree each → 0.
  • John Smith, head: 1,956 of 2,100, unchanged. That head is out of scope, because there a typed comma changes how the name is split into parts.
• Differential gate: exit 0 at all five baselines.

Tests and docs

• New invariant test_a_delimiter_core_in_a_tail_reads_as_its_comma_twin (129 texts). Its recorded negative control is 102 disagreements at 061f02d. It replaces Should a suffix delimiter inside a maiden clause count as a name word beside a link? Smith, John, PhD née Puig Mr. - i Soler keeps i Soler in the maiden name #538's "reads as if not written" invariant.
• The Should a suffix delimiter inside a maiden clause count as a name word beside a link? Smith, John, PhD née Puig Mr. - i Soler keeps i Soler in the maiden name #538 tests in test_group.py are rewritten for the comma reading, with both earlier answers recorded as negative controls. There's a new case row, Doe, John, PhD - and MD.
• rules.md:
  • C1: new sentence, examples, and the Accepted maiden-marker cost.
  • P3: the delimiter sentence is scoped to trailing suffix parts.
  • M2: the rule text defers to C1, and the example is flipped.
• decisions.md: a C1 entry, plus pointers marking Should a suffix delimiter inside a maiden clause count as a name word beside a link? Smith, John, PhD née Puig Mr. - i Soler keeps i Soler in the maiden name #538's entry and juan y garcia nee jones loses the family name — P3's word count includes the words the maiden name takes away #418's repair as superseded.
• release_log.rst: a new Should a declared suffix delimiter inside a joined connective run be dropped? Smith, John, Puig - i Soler keeps the - in the suffix #549 bullet, and the Josep Carod i Rovira reads family Rovira — Catalan and Polish link surnames with i, which is not a conjunction #397/Should a suffix delimiter inside a maiden clause count as a name word beside a link? Smith, John, PhD née Puig Mr. - i Soler keeps i Soler in the maiden name #538 bullet corrected.
• The rules corpus is regenerated and the 1.4.0 ledger rows are re-recorded, each with a dated note.

Review

Four passes: design docs, code, a second round on the first fix commit, then /pr-review-toolkit:review-pr (code, tests, comments). None found a correctness bug. Over 517k parses on both trees, every field that moved equals what the typed-comma version gives, and every recorded negative control reproduced.

• Fixed from the review rounds: a slower-than-linear split loop (now 4.7× time for 4× the input, against 8.0×), leftover loop scaffolding in _maiden_take, and doc errors (stale rule statements, a wrong figure, misquoted excerpts, "Should a suffix delimiter inside a maiden clause count as a name word beside a link? Smith, John, PhD née Puig Mr. - i Soler keeps i Soler in the maiden name #538 fix" wording in five ledgers).
• Tests added from the toolkit pass: a pin for a suffix part that is only its delimiter (a guard no test covered before or after this change), and a parse-level pin for PhD née - Jones. The comma-comparison grid now also compares reported ambiguity kinds; its negative control is unchanged at 102 of 129.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.98%. Comparing base (061f02d) to head (d180516).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #600   +/-   ##
=======================================
  Coverage   98.98%   98.98%           
=======================================
  Files          45       45           
  Lines        4235     4238    +3     
=======================================
+ Hits         4192     4195    +3     
  Misses         43       43           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.