Filter is not pushed down through joins

[coastalwhite]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

left = pl.LazyFrame({ 'a': [1, 2, 3], 'b': [2, 2, 2] })
right = pl.LazyFrame({ 'x': [7, 1, 2] })

df = left.join(right, how = 'cross').filter(pl.col.a == pl.col.b)

print(df.explain())

This outputs:

FILTER [(col("a")) == (col("b"))] FROM
  CROSS JOIN:
  LEFT PLAN ON: []
    DF ["a", "b"]; PROJECT */2 COLUMNS; SELECTION: None
  RIGHT PLAN ON: []
    DF ["x"]; PROJECT */1 COLUMNS; SELECTION: None
  END CROSS JOIN

Even though it should be able to optimize it to:

CROSS JOIN:
LEFT PLAN ON: []
  DF ["a", "b"]; PROJECT */2 COLUMNS; SELECTION: [(col("a")) == (col("b"))]
RIGHT PLAN ON: []
  DF ["x"]; PROJECT */1 COLUMNS; SELECTION: None
END CROSS JOIN

Log output

No response

Issue description

Filters are not pushed down through joins, even when that is possible.

Expected behavior

Filters are pushed down.

Installed versions

Replace this line with the output of pl.show_versions(). Leave the backticks in place.

[ritchie46]

Ouch.. Can you take this one?

[mcrumiller]

This is true even for left/inner joins too, not just cross:

left = pl.LazyFrame({"a": [1, 2, 3], "b": [1, 2, 9]})
right = pl.LazyFrame({"a": [1, 2, 3], "c": [4, 5, 6]})
out = left.join(right, how="left", on="a").filter(col("a") == col("b"))
print(out.explain())

FILTER [(col("a")) == (col("b"))] FROM
  LEFT JOIN:
  LEFT PLAN ON: [col("a")]
    DF ["a", "b"]; PROJECT */2 COLUMNS; SELECTION: None
  RIGHT PLAN ON: [col("a")]
    DF ["a", "c"]; PROJECT */2 COLUMNS; SELECTION: None
  END LEFT JOIN