fix(rust): Validate column names in unique() for empty DataFrames

[Biswas-N]

This PR addresses an issue where the unique() function in Polars does not raise a ColumnNotFoundError when called on an empty DataFrame with an unknown subset of column names. The changes ensure that column names in the subset are validated before proceeding, thereby raising the appropriate exception.

Changes Made:

1. Rust:

   • Added validation logic in UniqueExec executor to check the subset of column names provided exists in an empty DataFrame.

2. Python Tests:

   • Introduced a new test method, test_unique_with_bad_subset in test_unique.py to handle scenarios where Subset column name(s) do not exist.
   • Ensured invalid subset(s) raise a ColumnNotFoundError with appropriate message.

Linked Issue:

Closes #20209

Checklist:

• Changes rebased against the latest main branch.
• All new and existing tests pass.
• Verified using pytest for Python tests.
• Code adheres to the repository's contribution guidelines.

[codecov]

Codecov Report

Attention: Patch coverage is 50.00000% with 1 line in your changes missing coverage. Please review.

Project coverage is 79.02%. Comparing base (62ebbe5) to head (c982905).
Report is 10 commits behind head on main.

Files with missing lines	Patch %	Lines
...ates/polars-plan/src/plans/conversion/dsl_to_ir.rs	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #20411      +/-   ##
==========================================
+ Coverage   78.97%   79.02%   +0.04%     
==========================================
  Files        1562     1562              
  Lines      220103   220171      +68     
  Branches     2486     2486              
==========================================
+ Hits       173821   173981     +160     
+ Misses      45709    45617      -92     
  Partials      573      573              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ritchie46]

These error should be raised during conversion to IR, not at the implementation level.

[Biswas-N]

@ritchie46 thanks for having a look. I am new to contributing to pola-rs, could you help me by pointing at some code that raises issues during conversion to IR. It could help me understand pola-rs way of doing things.

[ritchie46]

A bit nitpicky, but I think this can be written a little bit less verbose.

for c in &cols {
   let _ = input_schema.try_get(name)?
}

[Biswas-N]

Haha no problem. My understanding was, if the subset column(s) does not exist we need to raise polars.exceptions.ColumnNotFoundError. This approach would raise polars.exceptions.SchemaFieldNotFoundError. On a high level look both seem appropriate to me, but if SchemaFieldNotFoundError is correct, I can update this piece to less verbose like you suggested.

Something like this:

let cols = expand_selectors(s, input_schema.as_ref(), &[])?;

// Checking if subset columns exist in the dataframe
for col in cols.iter() {
    let _ = input_schema.try_get(col)?;
}

Ok::<_, PolarsError>(cols)