[derek73]

Closes #1008.

Note

3.0 is not released before 2.8.1, which warns of it. 2.8.1, built on the 2.x branch, carries the DeprecationWarning that announces this removal. It is raised where code reads a constant of nameparser.config.regexes (from nameparser.config.regexes import REGEXES included); importing the module alone stays quiet. Every 3.0 removal warns in a released version first (decisions/3-0-reevaluations' standing discipline). This PR may merge to master before 2.8.1 ships, so the 3.0 migration guide that 2.8.1's warnings link to is live on readthedocs.

What changes

• nameparser/config/regexes.py is deleted. from nameparser.config.regexes import REGEXES raises ImportError. Parsing is unchanged: snapshot.py reports no moves.

• tests/v2/test_regex_sync.py pins each hand copy to the pipeline's pattern, not to the config table. _SOURCES now maps each copy to the (module, attribute) it must equal:

  • _render._ROMAN → _vocab._ROMAN.
  • _render._INITIAL → _vocab._INITIAL. This is now a straight copy; before, it was compared to config's initial with one ? spliced back in.
  • The imports _pieces/_assemble._PERIOD_ABBREV, _pieces/_assign._ROMAN and _classify._INITIAL → their _vocab original.

  The test that the two initial copies agree, the dotted-initial test, the comma-set test and the completeness checks all stay.

• Docs:

  • migrate.rst: a removals row (copy the pattern you reused; bidi and emoji are Policy(strip_bidi=...)/Policy(strip_emoji=...)). "Finding the sites" now says upgrade to 2.8.1 and that 2.8.1 warns where a constant of regexes is read. It lists a bare import of the module among the code that passes the warnings check and still fails on 3.0, and adds config.regexes to the names to search for.
  • customize.rst: the freeze paragraph names CAPITALIZATION_EXCEPTIONS alone.
  • modules.rst: the automodule entry and its "compiled patterns in regexes" clause are gone.
  • release log: a Breaking Changes bullet saying 2.8.1 warns where a constant of the module is read.
  • mechanisms.md Field notes: before deleting a pin, mutate each pattern it pins (Remove nameparser.config.regexes in 3.0? The parser has not read REGEXES since 2.0 #1008's three unguarded patterns).

• decisions/3-0-reevaluations: a 2026-11-02 Remove nameparser.config.regexes in 3.0? The parser has not read REGEXES since 2.0 #1008 entry superseding the 2026-10-28 Remove the 1.x vocabulary names (PREFIXES, FIRST_NAME_TITLES, …) and their bridge #357 entry's listing of regexes among the nameparser.config modules that stay. The Standing line and the superseded entry's parenthetical are updated to match.

• Smaller changes:

  • test_contracts.py drops regexes.REGEXES from the freeze test's exempt mappings.
  • Provenance comments in _vocab.py, _render.py, _tokenize.py and _pieces.py no longer point at the deleted file.
  • Root and nameparser/ AGENTS.md no longer point at it either, and the mypy gotcha gets a current example.

Patterns whose only other copy was the config table

• _render._SPACES, _SPACE_BEFORE_COMMA, _MAC and _WORD, _vocab._PERIOD_NOT_AT_END and _tokenize._BIDI have nothing left to sync with. They are declared in _SOURCES with None.
• _vocab._MARKS and _tokenize._EMOJI_RANGES are not compiled patterns, so they were never in the roster, and their dedicated tests are removed.
• The four _vocab patronymic patterns keep their pin through test_locales.py::test_a_packs_marker_regexes_are_the_rules_its_patch_turns_on, which holds the RU and TR_AZ packs' copies to them.

Three of these patterns had no other tripwire. Dropping U+1AB0-1AFF from _MARKS, or narrowing _SPACES or _SPACE_BEFORE_COMMA to a literal space, passed the whole suite. Behavior tests now catch each one:

• test_a_mark_from_each_combining_block_stays_in_its_word in test_parser.py: a mark from each of the five combining blocks keeps Dr. inz◌. Janez Novak's abbreviation a title.
• test_collapse_reads_any_whitespace_as_a_space in test_render.py: a tab, no-break space or ideographic space in a spec collapses, before a comma too.

Verification

• Every two-sided pin was run against a drifted copy, once with the pattern changed and once with the flags changed, and fails both times. That covers 7 roster entries plus test_initial_copies_agree.
• The three new behavior tests were each run against their mutation and fail: 1, 3 and 3 cases.
• uv run pytest -q (34,784 passed), uv run pytest -q -n0 -m serial (42 passed), uv run mypy, uv run ruff check, sphinx-build -b doctest docs (0 failures) and -W -b html, python -m doctest README.rst, and tools/differential/snapshot.py ("matches the snapshot").
• One serial run failed three scaling-ratio rows under a machine load average of 16 to 25. They passed on rerun, and the only nameparser/ change since the last full pass is a comment.

🤖 Generated with Claude Code