Fix vd/mc splitting by case before and after a family comma (#573) - #579
Merged
Merged
Conversation
Three uniform-case shapes split a word of both the particle and the
unambiguous suffix vocabulary, each because one stage acted on a
reading the next never took:
- before a family comma, the particle chain stopped before a trailing
run the family part never reads, and the per-piece read then peeled
VD from between two family words ('SMITH VD MA, JOHN');
- between credentials after a family comma, P6's attachment pulled VD
out of the middle of a run assign had read whole
('DOE, JANE PHD VD MA');
- behind a never-given particle opening the given part, S2's slot read
the member as a credential with words to spare that P1's fold then
removed ('SMITH, VD MA').
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
rules.md: P2's chain stop applies only where a trailing run is read; P6 gains a third exception (a word inside a credential run); S2's given slot names the never-given particle that leaves no word to spare. decisions.md records the three readings and what each declined, with the blast radius over a generated grid. The ledgers classify the new rules.md examples at every baseline, and the guard rosters record their reach and probes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…docs
Code review and the docs review, plus a widened grid (adding MA and
Prof. to the pool), found the first round wrong in five places:
- P2's fix lived in group's chain stop for the whole family part, and
a comma followed by no name word reads that part positionally, so
'Berg de MA, Prof.' lost its trailing credential and its report.
Moved to assign's per-piece family read, the site of the defect.
- P6's stand-down asked S2's company query, which counts only an
unambiguous credential in front: 'DOE, JANE MA VD PHD' and the
titled 'DOE, JANE PHD PROF. VD MA' still split. It now asks whether
assign read the run whole, titles transparent; an empty run (a 'do'
assign already read as the credential) declines.
- S2's never-given lead overrode mixed-case capitals ('Smith, de MA'
moved in silence) and missed a chained particle run ('SMITH, VD DE
LA MA'). Capitals now speak first; every token of a piece is asked.
- The lead test was a hand copy of P1's fold predicate; both now call
`_pieces.is_lone_never_given_particle`.
- decisions.md, the ledgers and the release note claimed mixed case
was already right; 2.2.0 and 2.3.0 split 'Smith vd Ma, John' too,
and 1,310 of the grid's movers are mixed case. Rewritten, with a
recipe that reproduces the grid exactly, and pointers from the P2
and S2 sections.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The given-slot lead test rescanned the particle run once per member, quadratic on 'SMITH, VD ' + 'DE '*k + 'MA '*k (29,670 -> 165,591 frames at k=256 against e56dd29). The run's end is kept once per title set, and _REREAD_SHAPES["lead"] guards it (3.38x kept, 7.53x rescanning, bound 6.0; checked failing on the rescanning code). - P6's third exception counts any post-nominal, a generation included, which the code did and the statement did not say. Decided and recorded: 'Berg, Jan PhD vd Jr.' reads suffix 'PhD vd Jr.'. - Its title skip BEHIND the word was dead (a title there leaves the word a name in assign) and is removed; the statement now says a title in front is transparent. - decisions.md: the false linear-cost claim corrected with the measurement, the residual positional split ('Smith vd Ma, Prof.') named as not decided. P2 says a lone letter behind the particle is a name word. Stale "company exception" label and comment drift fixed. Correction to 30e38ed's message: 'Smith, de MA' moved at ce4e1c5 but reported suffix-or-name; it did not move in silence. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ticle
P2 said a lone letter behind the particle is a name word; a lowercase
one is the numeral, as it is with no particle ('smith vd v, john' as
'smith v, john'). The _REREAD_SHAPES header and failure message now
record the lead row, its thin broken-side margin, and that counting
<genexpr> frames is not isolated from the rest of the parse.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #579 +/- ##
=======================================
Coverage 98.90% 98.91%
=======================================
Files 45 45
Lines 4118 4151 +33
=======================================
+ Hits 4073 4106 +33
Misses 45 45 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
credential_run failed on CI (3.11, 8.9x) and in three local full-suite runs, never alone. Frame counts on that input are identical to master and exactly linear (96,100 -> 384,100); the extra time was a full cyclic-GC pass over the heap earlier tests leave live (~339k objects), landing inside group's snapshot comprehension (34ms self time at x3200). Its cost is the heap's, not the parse's. Measured in the failing sequence: 6.33x with the collector on, 4.00-4.06x off. _best now collects once, then times with the collector disabled. A planted O(n)-per-piece scan still fails the row 3/3 at 16x. Co-Authored-By: Claude Opus 5.5 <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
1 participant
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.
Closes #573. The issue's fourth bullet (the silent mixed-case
Doe, Jane PhD vd Ma) moved to #578.What changes
A word in both the particle and the unambiguous suffix vocabulary (
vd,mc) split in three shapes. Each split was one stage acting on a reading the next stage never took.SMITH VD MA, JOHNSMITH MA, suffixVDSMITH VD MADOE, JANE PHD VD MAVD DOE, suffixPHD, MAPHD VD MASMITH, VD MASMITH VD, suffixMASMITH VD MASMITH VD, JOHN,SMITH VD JR, JOHN).Doe, Jane PhD vd→vd Doe). Two such words side by side are one particle run, which P6 takes whole.Smith, de MA), and an ambiguous particle keeps the given name (SMITH, VAN MA). P1's fold and the slot share one predicate,_pieces.is_lone_never_given_particle.Decided with the maintainer, including that a generation counts on either side:
Berg, Jan PhD vd Jr.reads suffixPhD vd Jr., where 2.2 and 2.3 read familyvd Berg.Docs
implemented:gains_assign.py.fix(#573)rules at every baseline, plus guard-roster enrollments, all reach verified name by name.Blast radius
Measured against
e56dd296:vd/mcleaving the family for the credential run (2,142). No list of ambiguity kinds changes, and 1,310 of the movers are mixed case.Review history
Three review rounds. Each round's findings were reproduced before being fixed:
Berg de MA, Prof.; it moved to assign. P6 asked S2's company query and missedDOE, JANE MA VD PHDand the titled shape. S2 overrode mixed-case capitals and missed chained particle runs. The docs claimed mixed case was already right, which 2.2 and 2.3 disprove._REREAD_SHAPES["lead"]guards it; the guard fails on the old code. P6's scope was stated, and a dead title skip removed.Commit
30e38ed5's message saysSmith, de MA"moved in silence"; it moved but reported. The correction is in43672bb9's message.Verification
uv run pytest: 10,975 passed, 327 skipped, 3 xfailed. mypy and ruff are clean.test_parse_cost_grows_no_worse_than_linearly[credential_run]failed on CI (3.11, 8.9x) and in full-suite runs, never alone. It was not load, as first assumed. On that input this branch and master run identical, exactly linear frame counts (96,100 → 384,100). The extra time was a full cyclic-GC pass over the heap earlier tests leave live, landing inside one parse.44a8fa8ctimes the scaling guards with the collector off after one collection. In the failing sequence the ratio was 6.33x with GC on and 4.00–4.06x with it off. A planted O(n)-per-piece scan still fails the row 3/3 at 16x.🤖 Generated with Claude Code