a inner join sql return wrong results about polars

[ion-elgreco]

Similar thing happens to cross join: #11927

[l1t1]

seems left join also wrong

df = pl.DataFrame({
    "id":['1','2','3']
})

df2 = pl.DataFrame({
    "id":['4','5','6']
})
ctx = pl.SQLContext(df=df, df2=df2)

lp1=ctx.execute(
    """
    SELECT df.id as id1,df2.id as id2
    FROM df
        left JOIN df2
    on df.id=df2.id
    """,
    eager=False,
)
lp1.collect()
shape: (3, 2)
┌─────┬─────┐
│ id1 ┆ id2 │
│ --- ┆ --- │
│ str ┆ str │
╞═════╪═════╡
│ 1   ┆ 1   │
│ 2   ┆ 2   │
│ 3   ┆ 3   │
└─────┴─────┘

should returns

shape: (3, 2)
┌─────┬─────┐
│ id1 ┆ id2 │
│ --- ┆ --- │
│ str ┆ str │
╞═════╪═════╡
│ 1   ┆ null   │
│ 2   ┆ null   │
│ 3   ┆ null   │
└─────┴─────┘

[ritchie46]

@universalmind303 can you take a look here? It seems we create wrong plans somehow, as it only happens in SQL.

[alexander-beedie]

Triage:

• For the first query: we don't actually recognise/support implicit join syntax (at all) yet, so it gets parsed (incorrectly) as a simple filter op where A == B. I'll have to see if we can identify the implicit join syntax and either generate the equivalent inner join or raise an error.

• For the second query: it seems we aren't resolving the post-join column selection properly here. We're doing the join correctly, but then returning only the cols from the left hand table. Will look at this first 🤔

[l1t1]

SELECT t.A, t.fruits, t1.B, t1.cars FROM t, t1 WHERE t.A=t1.B still returns wrong result in version 10. 20.31

[alexander-beedie]

SELECT t.A, t.fruits, t1.B, t1.cars FROM t, t1 WHERE t.A=t1.B still returns wrong result in version 10. 20.31

Indeed - as mentioned earlier we don't support implicit join syntax at the moment; can you confirm that the other error is fixed and open an issue for implicit join syntax as a new feature request? (Easier for us to track open/closed issues if they are distinct) 😎👍

[l1t1]

confirmed the inner join syntax works now