COLLECT Return bug (CYPHER25)

[joshcornejo]

This is a fragment of a query I am modifying (originally I had 1 MATCH but my graph has changed and I need to separate it into 2:

MATCH (P:Node { uuid: 'some_uuid' })
      -[a:someRelLabel1*0..1]->(Q:ALabel)
MATCH (P)-[a1:someOtherRelLabels*0..1]->(Q1:AnotherLabel)
   ... more statement
RETURN COLLECT (DISTINCT P) + (all other nodes) AS retNodes
       COLLECT (DISTINCT a) + (DISTINCT a1) + (all other rels) AS retRels

This is what retRels contains:

       
[[], <-- this seems to be a bug? 
[
{
identity: 22400,
start: 20093,
end: 20092,
type: "someRelLabel1", // <-- this is the label from "DISTINCT a"
properties: {},
elementId: "5:c6da763b-b380-48b5-86f8-cfa7cba1d4f4:22400",
startNodeElementId: "4:c6da763b-b380-48b5-86f8-cfa7cba1d4f4:20093",
endNodeElementId: "4:c6da763b-b380-48b5-86f8-cfa7cba1d4f4:20092"
}
], 
...

For context, this code:

foreach (var rel in cursor.Current["retRels"].As<List<List<IRelationship>>>())

now throws this exception (before Q/Q1 and a/a1 which were in the same line)

System.InvalidCastException: Unable to cast object of type `Neo4j.Driver.Internal.Types.Relationship` to type `System.Collections.Generic.List`1[Neo4j.Driver.IRelationship]`

[joshcornejo]

It gets even weirder, if I change my collect to:

COLLECT(DISTINCT a + a1 + b + c + d + e + f) AS retRels

• The old Neo4J desktop fetches me all the relationships
• The new Neo4J desktop misses any new relationship that wasn’t inserted before the change, but I can’t see the content - it doesn’t fetch me any details on ids
• The c# driver has missed all the new relationships as well,

[dana_canzano]

@joshcornejo

if you revert to the prior cypher version and preface your queries with

cypher 5 

do you get expected results or one in the same when using cypher 25. just trying to determine where the failure begins

[joshcornejo]

prefix with CYPHER 5 in new Desktop … still same results:

[[[:hasRule {type: ""}]], [:repercussion], [:target], [], [[:asset]], [], [], []]

Strange enough, the “count” of sets seems correct (the empty brackets)

[andrew_bowman]

So keep in mind that when you have a var-length pattern with 0 as the lower bound (as in your MATCH (P)-[a1:someOtherRelLabels*0..1]->(Q1:AnotherLabel) line) that if P and Q1 could possibly be the same node (if the label there is also present on P) then it will match to a path where P and Q1 are the same, and the list of relationships (a1) for that path is empty (a1 must be a list, as it’s part of var-length relationship; empty because the lower bound allows a pattern where no relationship exists and the end nodes are the same node).

As a simple example, to run on an empty db:

CREATE (:Start:Node {id:1})-[:REL {relId:1}]->(:End:Node {id:2});

Now match and collect rels:

MATCH (s:Start)-[a:REL*0..1]->(e:Node)
RETURN collect(DISTINCT a)

The result is:

[[], [[:REL {relId: 1}]]]

The first element of the empty list is an empty list, when there is no rel and the start and end nodes are the same.

[joshcornejo]

But I’m getting different results to the same query between the old Desktop and the new one?

Even before this query change, it looked like this:

MATCH (P:Node { uuid: 'some_uuid' })
       -[a:someRelLabel1|someOtherRelLabels*0..1]->(Q:ALabel) 
    ... more statement 
RETURN COLLECT (DISTINCT P) + (all other nodes) AS retNodes       
COLLECT (DISTINCT a) + (all other rels)

And I didn’t have this empty set result …

[andrew_bowman]

Was the data the same? The only way you would get an empty set result is if the labels are such that P and Q could be the same node. If it doesn’t have the labels necessary for that, then you won’t have that result.

[joshcornejo]

P is not the same as Q - the query below works “fine” (but I don’t want any subsequent nodes from Q:ALabel (i had to make subsequent relationships wildcards) so that’s why there are now 2 matches:

Original:

MATCH (P:Node { uuid: 'some_uuid' })
        -[a:someRelLabel1|someOtherRelLabels*0..1]->(Q:ALabel|AnotherLabel)

Modified:

MATCH (P:Node { uuid: 'some_uuid' })
       -[a:someRelLabel1*0..1]->(Q:ALabel) 
MATCH (P)-[a1:someOtherRelLabels*0..1]->(Q1:AnotherLabel)

Data the same, database the same, below is the same query ran on 1.6.2 and 2.x":

You can see it doesn’t fetch relationship 23260 (even though is the same type as 22738), and the results as a table are all different:

Further, C#’s results are the same as Desktop2 - empty relationships without content at the front and back and giving a casting exception.

[joshcornejo]

Here, the top of the image is with 2 MATCH clauses / the bottom of the image is with 1 MATCH … 23260 is still missing.

[joshcornejo]

FYI, I’ve modelled the query with 1 MATCH and a WHERE type(b) <> ‘AnotherLabel’ and I get output discrepancies between Dv2 (top) and Dv1 (bottom) - where bottom returns what I want (top still misses a the bottom left relationship.

(Everything is up-to-date - desktop and db)

[joshcornejo]

My data stored looks like the desktop v1 (node 0.1 has 2 relationships 22740 and 22400, 20162 has 1 relationship to 20650)

MATCH (P:Node { uuid: 'some_uuid' })
      -[a:someRelLabel1*0..1]->(Q:ALabel|AnotherLabel)
      -[b]->(R)
      ... more statements
WHERE type(b) <> 'AnotherLabel'

If I don’t have the where clause - i get the same results on both desktops, if I have the where clause, then I get different results (so i assume v1 has that “bring relationships anyway”)

If I change to:

MATCH (P:Node { uuid: 'some_uuid' })
      -[a:someRelLabel1*0..1]->(Q:ALabel|AnotherLabel)
WITH P, Q, a
OPTIONAL MATCH (Q)-[b]->(R)
WHERE type(b) <> 'AnotherLabel'

I get the same results.

If I try:

MATCH (P:Node { uuid: 'some_uuid' })
      -[a:someRelLabel1*0..1]->(Q:ALabel|AnotherLabel)
      -[b:*0..1]->(R)
      ... more statements
WHERE type(b) <> 'AnotherLabel'

I get an error “Invalid input '*', …”.

But this works:

MATCH (P:Node { uuid: 'some_uuid' })
      -[a:someRelLabel1*0..1]->(Q:ALabel|AnotherLabel)
      -[*0..1]->(R)
      ... more statements

… but in this case I have lost relationship b and the ability to fetch it …

[joshcornejo]

I am sure there is a bug somewhere … after much tinkering

MATCH (P:Node { uuid: 'some_uuid' })       
      -[a:someRelLabel1*0..1]->(Q:ALabel|AnotherLabel)
      -[b*0..1]->(R)

and

MATCH (P:Node { uuid: 'some_uuid' })       
      -[a:someRelLabel1*0..1]->(Q:ALabel|AnotherLabel)
      -[b*0..1]->(R)

and

MATCH (P:Node { uuid: 'some_uuid' })       
      -[a:someRelLabel1*0..1]->(Q:ALabel|AnotherLabel)
      -[b*0..1]->(R)
...
WHERE NOT (Q)-[:hasRule]->(R)

should return the same results for 2 of my 3 graphs … but when added the WHERE clause it inserted an empty []… which generated the:

@joshcornejo

System.InvalidCastException: Unable to cast object of type Neo4j.Driver.Internal.Types.Relationshipto typeSystem.Collections.Generic.List`1[Neo4j.Driver.IRelationship]``

I introduced code:

var tmpList = cursor.Current["retRels"].As<List<List<object>>>().Where(innerList => innerList.Count != 0).ToList();

to “clear” those empty lists from the result set in C#