orneryd / NornicDB / 37216027258

89%

Build Type

push

github

Committed by orneryd

Commit Message

fix(storage,cypher): label and relationship-type names are case-sensitive (#862)

Storage lower-cased label and relationship-type names in the label index,
the relationship-type index, the relationship-between set and heads, the
label and type counts and the temporal index, so :Person, :person and
:PERSON shared index entries and counts (count(:Person) was 2 with one
:Person and one :person node) and :R / :r between the same nodes shared
one lookup head. Neo4j's names are case-sensitive.

Key builders, caches, transaction and pending overlays, the asynchronous
engine and read-path comparisons now use the name as written
(normalizeLabel, normalizeCountLabel, normalizeCountEdgeType removed;
the async engine uses hasLabel). Cypher's relationship-type and label
filters (APOC path algorithms, FastRP, count fast paths, call tails,
index hints, vector and EXPLAIN index matching) compare exactly, and
generated constraint and index names keep the label's and property's
case, so unnamed schema on :Person and :person no longer collide. Vector
space keys keep the label and property case.

Storage version 4: the V3-to-V4 upgrade (behind --upgrade-storage)
rebuilds the label, relationship-type and relationship-between indexes
from the stored bodies (one chunked scan helper now serves the label,
type and relationship-between rebuilds), moves deindex catalogs and
tombstones to the new keys, and clears the clean-shutdown marker so the
temporal index and MVCC heads are rebuilt; counts are rebuilt at open.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Coverage Stats

396 of 396 new or added lines in 34 files covered. (100.0%)

42 existing lines in 11 files now uncovered.

180485 of 207449 relevant lines covered (87.0%)

1.02 hits per line