Stop parsing from mutating the Constants it reads#222
Merged
Conversation
parse_pieces() and join_on_conjunctions() derived new lookup entries while parsing (period-joined titles/suffixes like "Lt.Gov.", conjunction-joined pieces like "Mr. and Mrs." and "von und zu") and add()ed them directly to self.C — by default the shared module-level CONSTANTS singleton. Parsing one name therefore permanently changed how every later name in the process parsed: results depended on input order, were not reproducible across runs, and concurrent parsing raced on the shared sets. It also churned the suffixes_prefixes_titles cache, which invalidated on every such add(). The derivations are only needed within the parse that produced them, so track them in per-instance _derived_* sets that the is_title / is_suffix / is_conjunction / is_prefix / is_rootname predicates consult alongside self.C, and reset them at the start of each parse_full_name() run. Parse results are unchanged (each parse re-derives its own entries); the config is now read-only during parsing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #222 +/- ##
==========================================
+ Coverage 97.25% 97.30% +0.04%
==========================================
Files 13 13
Lines 801 815 +14
==========================================
+ Hits 779 793 +14
Misses 22 22 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
- Fix stale parse_pieces docstring that still described adding derived parts "to the constant" - Document the derived-set branch in the is_prefix / is_suffix / is_conjunction docstrings, and reword is_leading_title's now-vacuous "does not mutate C.titles" contrast into the narrower load-bearing fact (the regex match is not registered anywhere, even per-parse) - Remove the redundant _derived_suffixes check from is_suffix_lenient: it falls through to is_suffix(), which already consults the overlay, and derived suffixes always contain an interior period so they can never trip the is_an_initial() veto that "lenient" exists to bypass - Add mutation-verified regression tests for the two coverage gaps: is_rootname's derived-title exclusion (its overlay checks were deletable without any test failing, yet dropping them misparses "Lt.Gov. juan e garcia" into an empty first name) and the __setstate__ backfill for pickles predating the derived sets (whose absence crashes capitalize() with AttributeError) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The shared-configuration warning in customize.rst now has a precise boundary worth stating: after the parse-time mutation fix, instances can only affect each other through explicit add/remove calls, so concurrent parsing against the shared CONSTANTS is safe. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Replace the hand-enumerated _WATCHED_ATTRS list with a structural snapshot derived from Constants.__getstate__(), the canonical listing of an instance's own config. A collection added to Constants later is watched automatically — there is no list to remember to update (unlike conftest's _COLLECTION_CONFIG_ATTRS, which remains the only manual one). A sanity assert fails loud if the discovery ever stops covering the five historically-mutated sets, and the snapshot now also covers TupleManager collections and scalar overrides. Update AGENTS.md: fix the two Architecture lines that still described parse_pieces()/join_on_conjunctions() as writing into the constants, and add a gotcha documenting the parsing-never-writes-config invariant and the _derived_* overlay recipe for future derived categories. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The existing self-staleness reminders live inside two extension-pattern recipes and the release checklist, so a behavior change that fits neither pattern reaches PR time with no check on this file — exactly how the parse-flow lines describing the old mutate-the-config behavior survived the PR #222 review round. The nudge states why grep is the wrong tool for it: AGENTS.md paraphrases behavior in its own words, so stale text rarely matches the diff's phrasing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
parse_pieces()andjoin_on_conjunctions()wrote back into the config while parsing a name: period-joined titles/suffixes ("Lt.Gov."→C.titles,"JD.CPA"→C.suffix_not_acronyms) and conjunction-joined pieces ("of the"→C.conjunctions,"Mr. and Mrs."→C.titles,"von und zu"→C.prefixes). Sinceself.Cdefaults to the shared module-levelCONSTANTSsingleton, parsing one name permanently changed how every subsequent name in the process parsed:suffixes_prefixes_titlescache,conftest.pyhad to snapshot/restore allCONSTANTScollections around every test just to stay order-independent.Fix
The derivations are load-bearing only within the parse that produced them (registered in
parse_pieces, consumed by lateris_title()/is_suffix()/… checks in the sameparse_full_name()run). So they now live in per-instance_derived_titles/_derived_suffixes/_derived_conjunctions/_derived_prefixessets that the lookup predicates (is_title,is_suffix,is_conjunction,is_prefix,is_rootname) consult alongsideself.C, reset at the start of each parse and backfilled in__setstate__so pickles from older versions keep working.self.Cis now read-only during parsing.Parse results are unchanged — each parse re-derives its own entries — verified against master for all trigger inputs, including the comma-format paths.
Tests
New
ParsingDoesNotMutateConfigTests(TDD: written first, failed on the leak assertions):Constants.__getstate__(), so collections added toConstantsin the future are watched automatically (no list to maintain),Constants(constants=None) isn't polluted either, and re-assigningfull_namere-derives from a clean slate.Review-driven regression tests (both mutation-verified — the guarded code was deletable with the full suite passing before these):
is_rootnamemust exclude derived titles from the rootname count ("Lt.Gov. juan e garcia"— overlay-blind counting empties the first name),HumanNamestate without the_derived_*attributes must not crashcapitalize().Docs
customize.rst: the shared-config section now states the new guarantee — parsing never modifies the configuration, so concurrent parsing against the shared instance is safe.AGENTS.md: parse-flow notes updated (they described the old mutate-the-config behavior) and a new gotcha documents the invariant plus the_derived_*recipe for future derived categories.Full suite: 1340 passed. ruff, mypy, sphinx doctests, and README doctest all clean.
🤖 Generated with Claude Code