Issues

call and yield can't be a property key, a map key or a label, as Neo4j takes them

liberation-data/drivine4j#639 · opened

on Nov 6, 2026

Oracle: read a pattern's property through its index where the graph has one, from the store's catalog

liberation-data/drivine4j#638 · opened

on Nov 6, 2026

1

Run Neo4j schema DDL (CREATE/DROP INDEX and CONSTRAINT, db.awaitIndexes) through the schema machinery on the SQL stores

liberation-data/drivine4j#636 · opened

on Nov 6, 2026

Accept end, and the other reserved words Neo4j allows, as a variable, so an APOC export's relationship statements compile

liberation-data/drivine4j#635 · opened

on Nov 6, 2026

Say in the full-text docs that the SQL stores split dotted and hyphenated names into words, where Neo4j's default analyzer doesn't

liberation-data/drivine4j#633 · opened

on Nov 6, 2026

A supertype's private @NodeId: the runtime takes it as a subclass's node id, the model file says the subclass has none

liberation-data/drivine4j#628 · opened

on Nov 6, 2026

Postgres: a pooled session that built a schema index in parallel keeps its shared memory, so a later vector index can't be built in a 64 MB /dev/shm

liberation-data/drivine4j#626 · opened

on Nov 5, 2026

Full-text search on Oracle without Oracle Text finds nothing for a word with a combining mark (a virama, a decomposed accent), which Oracle's [[:alnum:]] holds

liberation-data/drivine4j#625 · opened

on Nov 5, 2026

Docs: a vector index on a large label can fail on Postgres in a container with Docker's default 64 MB /dev/shm

liberation-data/drivine4j#624 · opened

on Nov 5, 2026

2

Full-text search reads a * inside quotes, "of sou*", as a prefix on the SQL stores, where Neo4j's parser searches for the phrase "of sou"

liberation-data/drivine4j#622 · opened

on Nov 5, 2026

ModelFileGeneratorTest: the model file generated twice differed once in CI (the file is the same each time)

liberation-data/drivine4j#621 · opened

on Nov 5, 2026

Compiled Cypher: MATCH (n:L {key: row.x}) after UNWIND doesn't use the key's index on Postgres or Oracle, so a batch of relationships reads the label once per row (port of drivine#636)

liberation-data/drivine4j#617 · opened

on Nov 5, 2026

1

Full-text search reads a word's characters one way in the query (\p{L}\p{N}_) and another in the text ([[:alnum:]_]): ² and spacing marks fall between

liberation-data/drivine4j#616 · opened

on Nov 5, 2026

Full-text search on the SQL stores weighs a prefix term (kiw*) by BM25, where Neo4j scores every prefix hit the same

liberation-data/drivine4j#612 · opened

on Nov 5, 2026

A MATCH after a FOREACH that creates relationships is quadratic on Postgres

liberation-data/drivine4j#611 · opened

on Nov 5, 2026

A CALL body that creates a relationship and then MATCHes it is quadratic on Postgres

liberation-data/drivine4j#610 · opened

on Nov 5, 2026

Full-text search on the SQL stores undercounts a term repeated back to back: 'kiwi kiwi kiwi' counts 2

liberation-data/drivine4j#607 · opened

on Nov 5, 2026

Keep full-text search's label figures (node count, total length) between searches, so a rare word's BM25 search reads only its candidates

liberation-data/drivine4j#605 · opened

on Nov 5, 2026

Oracle: a relationship end bound in an outer part (EXISTS, WITH, OPTIONAL MATCH) isn't keyed as #581 keys one, and may read the label index once per edge

liberation-data/drivine4j#602 · opened

on Nov 5, 2026

1

Oracle: a text past 32 KB from an unwound parameter, returned in a list over two rows, breaks the session for a later statement

liberation-data/drivine4j#599 · opened

on Nov 5, 2026

Oracle: UNWIND $ids AS i MATCH (n) WHERE id(n) IN [i] compares each id with every node, as the IN list is a correlated JSON_TABLE

liberation-data/drivine4j#598 · opened

on Nov 5, 2026

UNWIND $ids ... DETACH DELETE n still gathers its rows before the set-based delete

liberation-data/drivine4j#594 · opened

on Nov 4, 2026

The Cypher parser refuses a variable named all, which Neo4j accepts

liberation-data/drivine4j#593 · opened

on Nov 4, 2026

A plain DELETE's still-has-relationships check runs per clause, not at commit as on Neo4j

liberation-data/drivine4j#592 · opened

on Nov 4, 2026

Reading a node later in the statement that deleted it differs from Neo4j on the SQL stores

liberation-data/drivine4j#591 · opened

on Nov 4, 2026

Navigation Menu

Repository navigation

Issues

call and yield can't be a property key, a map key or a label, as Neo4j takes them

Oracle: read a pattern's property through its index where the graph has one, from the store's catalog

Run Neo4j schema DDL (CREATE/DROP INDEX and CONSTRAINT, db.awaitIndexes) through the schema machinery on the SQL stores

Accept end, and the other reserved words Neo4j allows, as a variable, so an APOC export's relationship statements compile

Say in the full-text docs that the SQL stores split dotted and hyphenated names into words, where Neo4j's default analyzer doesn't

A supertype's private @NodeId: the runtime takes it as a subclass's node id, the model file says the subclass has none

Postgres: a pooled session that built a schema index in parallel keeps its shared memory, so a later vector index can't be built in a 64 MB /dev/shm

Full-text search on Oracle without Oracle Text finds nothing for a word with a combining mark (a virama, a decomposed accent), which Oracle's [[:alnum:]] holds

Docs: a vector index on a large label can fail on Postgres in a container with Docker's default 64 MB /dev/shm

Full-text search reads a * inside quotes, "of sou*", as a prefix on the SQL stores, where Neo4j's parser searches for the phrase "of sou"

ModelFileGeneratorTest: the model file generated twice differed once in CI (the file is the same each time)

Compiled Cypher: MATCH (n:L {key: row.x}) after UNWIND doesn't use the key's index on Postgres or Oracle, so a batch of relationships reads the label once per row (port of drivine#636)

Full-text search reads a word's characters one way in the query (\p{L}\p{N}_) and another in the text ([[:alnum:]_]): ² and spacing marks fall between

Full-text search on the SQL stores weighs a prefix term (kiw*) by BM25, where Neo4j scores every prefix hit the same

A MATCH after a FOREACH that creates relationships is quadratic on Postgres

A CALL body that creates a relationship and then MATCHes it is quadratic on Postgres

Full-text search on the SQL stores undercounts a term repeated back to back: 'kiwi kiwi kiwi' counts 2

Keep full-text search's label figures (node count, total length) between searches, so a rare word's BM25 search reads only its candidates

Oracle: a relationship end bound in an outer part (EXISTS, WITH, OPTIONAL MATCH) isn't keyed as #581 keys one, and may read the label index once per edge

Oracle: a text past 32 KB from an unwound parameter, returned in a list over two rows, breaks the session for a later statement

Oracle: UNWIND $ids AS i MATCH (n) WHERE id(n) IN [i] compares each id with every node, as the IN list is a correlated JSON_TABLE

UNWIND $ids ... DETACH DELETE n still gathers its rows before the set-based delete

The Cypher parser refuses a variable named all, which Neo4j accepts

A plain DELETE's still-has-relationships check runs per clause, not at commit as on Neo4j

Reading a node later in the statement that deleted it differs from Neo4j on the SQL stores