Python API: Segfault with nodes() and rels() when using get_as_arrow()

[cmdlineluser]

import kuzu
import pathlib
import tempfile

with tempfile.TemporaryDirectory() as tmp:
    tmp = Path(tmp)
    
    users = tmp / "users.csv"
    users.write_bytes(b"A\nB\nC\n")
    
    follows = tmp / "follows.csv"
    follows.write_bytes(b"A,B\nB,C\n")

    db = kuzu.Database(tmp)
    c = kuzu.Connection(db)

    c.execute("CREATE NODE TABLE User(name STRING, PRIMARY KEY (name))")
    c.execute("CREATE REL TABLE Follows(FROM User TO User)")

    c.execute(f'COPY User FROM "{str(users)}"')
    c.execute(f'COPY Follows FROM "{str(follows)}"')

    df = c.execute("""
    MATCH (a:User)-[f:Follows* SHORTEST]->(b:User) 
    RETURN a.name, b.name, nodes(f) as nodes
    """).get_as_pl()
    
    print(df)
    # segfault

I guess it's an arrow issue as .get_as_arrow() also segfaults.

.get_as_df() returns a pandas DataFrame without error:

  a.name b.name                                              nodes
0      B      C                                                 []
1      A      B                                                 []
2      A      C  [{'_id': {'offset': 1, 'table': 0}, '_label': ...

properties(nodes(f), 'name') works as expected and returns a list[str] which suggests the {} dict/struct type may be the issue.

[prrao87]

Yep, that seems like it's to do with Arrow, so once that's addressed, the get_as_pl() would also work as intended.

[andyfengHKU]

I wonder if @cmdlineluser can double check if this fixes the issue.

[prrao87]

@cmdlineluser you can test this out on your sample DB with the latest pre release:

pip install --pre kuzu

Let us know!