Incorrect rows with comma-separated `MATCH` pattern list

[ting668]

Reporter: ting668

Environment

• OS: CentOS Linux 7 (Core), inside Docker container
• CPU/Architecture: x86_64
• Docker image: tugraph/tugraph-runtime-centos7:latest
• TuGraph-DB Version: 4.5.2 (tugraph-4.5.2-1.x86_64)

Description

While testing TuGraph using a method based on attribute-constraint analysis, I found that a comma-separated MATCH pattern list can return rows that violate the label constraint.

In the example below, a is constrained to Teacher, but TuGraph returns an Actor node as a.

How to Reproduce and Expected Behavior

Note: The queries below are a minimized, simplified example reproducing the bug.

Original graph G:

CALL db.createVertexLabel('Teacher','name','name','STRING',false,'gender','STRING',true,'faculty','STRING',true);
CALL db.createVertexLabel('Student','name','name','STRING',false,'gender','STRING',true,'faculty','STRING',true);
CALL db.createVertexLabel('Actor','name','name','STRING',false,'gender','STRING',true);

CREATE (:Teacher {name:'Bob', gender:'Male', faculty:'computer'}),
       (:Actor {name:'Owen', gender:'Male'}),
       (:Student {name:'Mary', gender:'Female', faculty:'computer'}),
       (:Student {name:'Lily', gender:'Female', faculty:'computer'});

Query:

MATCH (a:Teacher {gender:'Male'}), (b:Student {gender:'Female'})
RETURN a.name, b.name;

Expected result on G:

Bob, Mary
Bob, Lily

Actual result on G:

Bob, Mary
Bob, Lily
Owen, Lily

Induced subgraph G1 using faculty = 'computer' removes the Actor node:

CALL db.createVertexLabel('Teacher','name','name','STRING',false,'gender','STRING',true,'faculty','STRING',true);
CALL db.createVertexLabel('Student','name','name','STRING',false,'gender','STRING',true,'faculty','STRING',true);

CREATE (:Teacher {name:'Bob', gender:'Male', faculty:'computer'}),
       (:Student {name:'Mary', gender:'Female', faculty:'computer'}),
       (:Student {name:'Lily', gender:'Female', faculty:'computer'});

The same query on G1 returns the expected two rows:

Bob, Mary
Bob, Lily

Equivalent query on G also returns the expected two rows:

MATCH (a:Teacher {gender:'Male'})
WITH a
MATCH (b:Student {gender:'Female'})
RETURN a.name, b.name;

Expected behavior: all equivalent formulations should return only Teacher-Student pairs.

Actual behavior: the comma-separated pattern query on G leaks the Actor node Owen into the result.