[stevevanhooser]

Closes #237.

What changes

did.database/search_doc_ids stripped the ~ from a ~or and searched it as a plain or, so ~or(a, b) returned the documents matching a or b. It now remembers the negation and returns:

• the documents of the branch minus the OR's matches, or
• inside an AND, where candidate_ids is a cell array (the restricted path), the candidates minus the OR's matches.

That is what did.datastructures.fieldsearch and DID-python's SQLiteDB._search_doc_ids already do.

The un-negated OR is searched by calling search_doc_ids again with operation = 'or', so it still takes the base.id fast path (baseIdLookupValues / search_base_ids) and the batched path from #234 (search_or_operands). A ~or nested inside an OR was already handed to search_doc_ids as a single operand by search_or_operands, so it now gets the negation too. The AND ranking (rank) is unchanged. query_struct_to_sql_str is not touched: a ~or never reaches it, and #236 is changing that function.

Tests

New tests/+did/+unittest/TestNegatedOr.m compares db.search with fieldsearch applied to every document, both with and without the search indexes (QueryLogSqlitedb.noSearchIndexes). It covers:

• a top-level ~or, including one that matches nothing and one that matches every document;
• a ~or nested in an OR, including a ~or inside a ~or;
• a ~or inside an AND (restricted candidate path). The test also checks that this path never lists the whole branch;
• a ~or of base.id lookups, at top level, nested in an OR and inside ANDs.

did.query can't build a ~or, so the structures are built by hand. They use field comparisons instead of isa, because fieldsearch only understands isa after did.query/to_searchstructure has expanded it.

TestBatchedOr/testMixedOr needs no change. Its batched and per-operand searches both send the ~or operand through search_doc_ids, so they still agree, and its other checks still hold under the corrected result. The extra branch-listing query isn't counted, because its SQL doesn't start with SELECT DISTINCT docs.doc_id.

I haven't run the MATLAB code locally (no MATLAB in this environment). CI will be the first run.

Companion PR

DID-python: VH-Lab/DID-python#107. It adds the same tests and moves the bridge hash to this branch's database.m commit 2c9d8fa7. Please merge this PR first, with a merge commit, so that hash stays on main.

🤖 Generated with Claude Code

https://claude.ai/code/session_012MroTijyNKsCMqFHpXFbpc

[github-actions]

Test Results

515 tests  +4   515 ✅ +4   2m 45s ⏱️ +5s
 43 suites +1     0 💤 ±0 
  1 files   ±0     0 ❌ ±0 

Results for commit 2c9d8fa. ± Comparison against base commit 6c3224f.

[codecov]

Codecov Report

❌ Patch coverage is 93.75000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 67.33%. Comparing base (6c3224f) to head (2c9d8fa).

Files with missing lines	Patch %	Lines
src/did/+did/database.m	93.75%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #238      +/-   ##
==========================================
+ Coverage   67.24%   67.33%   +0.09%     
==========================================
  Files          67       67              
  Lines        4793     4807      +14     
==========================================
+ Hits         3223     3237      +14     
  Misses       1570     1570              

☔ 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.