orneryd / NornicDB / 36367670671

87%

Build Type

push

github

Committed by orneryd

Commit Message

fix(cypher): reject NOT IN, dangling list commas, doubled quotes, dangling UNWIND

#514 family, validator side: statements Neo4j rejects were accepted.

- `x NOT IN [list]` is now a SyntaxError on every route; the Cypher
  negation `NOT x IN [list]` keeps working, as does `NOT (x IN [list])`.
  Quoted/comment text never matches.
- A list literal with a trailing or leading comma (`[1, 2,]`, `[, 1]`,
  `[[1],]`) is a SyntaxError.
- Adjacent string literals (`'a''b'`) are a SyntaxError: Cypher's string
  escape is a backslash, not a doubled quote. `'a\'b'` and `''` remain
  valid.
- A statement whose last clause is UNWIND with nothing after it
  (`UNWIND [1] AS x`) is a SyntaxError. This is a whole-statement check:
  fabric fragments that legitimately end in UNWIND are not affected.
- Clause keywords used as names (aliases, property keys) are not mistaken
  for clauses; the scan uses a first-byte keyword index so it stays out
  of the hot path (CPU profile: validator scans gone from the top nodes
  after the optimization).

Statement-level tests that pinned the non-Cypher forms were updated to
the Neo4j forms. Coverage: TestNotInOperatorRejected,
TestListLiteralTrailingCommaRejected, TestAdjacentStringLiteralsRejected,
TestDanglingUnwindRejected. Full pkg/cypher suite green (42.3s),
pkg/server green (113s), TCK ratchet 7794/7794, new tests pass twice
under -race.

Benchmarks (M2 Max, BenchmarkStatementRouting simple_match_limit):
autocommit 17.6-17.9 us/op, 11246-11248 B, 161 allocs; explicit tx
11.45-11.53 us/op, 5679 B, 88 allocs — allocation-identical to the
recorded band (16.4-17.6 / 10.4-11.1 us, 160-163 / 87-90 allocs).

Coverage Stats

142 of 159 new or added lines in 3 files covered. (89.31%)

21 existing lines in 7 files now uncovered.

175282 of 201485 relevant lines covered (87.0%)

1.02 hits per line