fix(rust, python): fix rename + projection pushdown

[ritchie46]

fixes #10595

[reswqa]

Nice fix!

BTW, I have a little concern about the case without optimization.

polars/crates/polars-plan/src/logical_plan/functions/rename.rs

Lines 31 to 39 in a75495a

	for (old, new) in existing.iter().zip(new.iter()) {
	// the column might be removed due to projection pushdown
	// so we only update if we can find it.
	if let Some(dtype) = input_schema.get(old) {
	if new_schema.with_column(new.clone(), dtype.clone()).is_none() {
	new_schema.remove(old);
	}
	}
	}

Looks like rename_schema may trigger incorrect order when swapping fields. 🤔

That's to say the following code panicked unexpected.

  def test_without_opt() -> None:
    lf = pl.LazyFrame(schema=["a", "b"])
    assert lf.select("a", "b").rename({"b": "a", "a": "b"}).select(
        "a"
    ).collect(projection_pushdown=False).schema == {"a": pl.Float32}

[ritchie46]

Hmm... Is that panic dur to changes in this PR? Otherwise we have to fix that in a separate one.

Swapping renames bite me more often. :/

[reswqa]

Hmm... Is that panic dur to changes in this PR?

No, that panic is not introduced by this change. I am bringing it up here because the test case that can reproduce it is almost the same as in this PR, with the only difference being that optimization is turned off.

Otherwise we have to fix that in a separate one.

That's make sense! I will open a separate issue to cover it.

[ritchie46]

That's make sense! I will open a separate issue to cover it.

Thanks!