fix: de-duplicate identical editions in all_editions (#317)

[grossir]

What

Fixes #317. get_citations() was returning the same Edition object twice in all_editions, e.g.:

c = get_citations("238 F.3d 273")[0]
len(c.all_editions)                     # 2 (was 1 pre-#305)
len(set(c.all_editions))                # 1  -> identical objects
c.all_editions[0] is c.all_editions[1]  # True

Root cause

The whitespace relaxation from #305 turns each period/space in a reporter regex into \s*. For a reporter with a whitespace-only variation like F. 3d, the variation regex becomes F\.\s*\s*3d, which also matches the zero-space text F.3d. So two extractors fire on the same span — the exact one and the variation one — and both register the same Edition object (one into exact_editions, one into variation_editions). CitationToken.merge only deduped within each list, never across them, so all_editions = exact + variation carried the duplicate. The S.W.2d custom regex hit the same bug independently (its regex omits $edition), predating #305.

Fix

• New _dedupe_editions() helper: order-preservingly dedupes each list and drops any variation edition already present as an exact match.
• Routed ResourceCitation.__post_init__, CitationToken.__post_init__, and CitationToken.merge through it.
• Genuinely distinct editions (real ambiguity) are preserved.

Impact

Fixes downstream false positives in callers that use len(citation.all_editions) to detect reporter ambiguity (e.g. CourtListener's update_casenames_wl_dataset, which was skipping every whitespace-variation reporter as "unable to disambiguate").

Hash note: this changes the hash of law and journal citations that previously carried a duplicate edition. Case citations are unaffected — CaseCitation.__hash__ never depended on all_editions. Resolution grouping within a run is unchanged; only absolute hash values (never persisted) differ across versions.

Tests

• New test_no_duplicate_editions — covers F.3d, U.S., the S.W.2d collapse, and B.R. (genuine ambiguity must survive dedup).
• Updated test_reporter_tokenizer assertion.
• Full suite: 54 tests pass.

🤖 Generated with Claude Code

[github-actions]

The Eyecite Report 👁️

Gains and Losses

There were 0 gains and 0 losses.

Total citations found: base 15988, PR 15988 (net +0).

Click here to see details.

id	Gain	Loss

Time Chart

Generated Files

Base (main) Output
PR Output
Full Output CSV