apoc.meta.data() reports wrong property type information

[neo-technology-build-agent]

Issue by moxious
Thursday Oct 17, 2019 at 18:50 GMT
Originally opened as neo4j-contrib/neo4j-apoc-procedures#1319

---

Replication that worked for me:

create (:A { x: 1 }), (:A { x: 'Foo' });
create (:B { x: 'Foo' }), (:B { x: 'Foo' }), (:B { x: 1 });
CALL apoc.meta.data();

This setup is intentionally to confuse APOC. For both labels A and B, the property X is typed both string and long.

The APOC call:

This procedure uses a sampling configuration as specified by the config parameters, which is a MetaConfig object. The sampling appears respected, but the property type information stored in the MetaResult class is "last type in wins".

I would say this behavior is incorrect. There are two possible correct behaviors:

1. Return an array of different types the property could have, as observed in the sample ["Long", "String"]
2. Perform some type-coercion, and always report "String" as a single property type, because it is the widest of all observed types.

[neo-technology-build-agent]

Comment by moxious
Thursday Oct 17, 2019 at 18:53 GMT

---

I found this because the issue is related to a PR in progress. In that PR APOC will have different stored procs that return the array of encountered types (fix suggestion 1) . neo4j-contrib/neo4j-apoc-procedures#1318

[neo-technology-build-agent]

Comment by jexp
Tuesday Oct 29, 2019 at 12:13 GMT

---

for the type field I would probably coerce to string
and add another field types that has the full list either single element or multi

[neo-technology-build-agent]

Comment by jexp
Tuesday Oct 29, 2019 at 12:14 GMT

---

Probably also need to adapt in apoc.meta.schema

[hvub]

One could also report a dynamic union type of all observed types. That would be closer to how Cypher thinks about it.