Description

Currently, in pl.concat() with how="align", we can combine multiple DataFrames by auto-determining the common key columns. According to the documentation, this operation always performs a full outer join, which can be relatively slow for large datasets.

I noticed that in the internal function pl.align_frames(), it is possible to set how="left" for alignment, which speeds up the process significantly.

Proposal

It would be useful to introduce a how="align_left" option in pl.concat() that performs alignment by always using the keys from the first DataFrame for left joins. This would be a faster alternative to the current how="align" behavior.

Benefits

• Performance Improvement: Aligning frames with a left join is generally faster than performing a full outer join.
• Flexibility: Provides users with more control over how frames are aligned when combining them.
• Consistency: Leverages existing functionality in pl.align_frames() to enhance pl.concat().

Suggested Implementation

• Add a new how="align_left" value to pl.concat().
• Under the hood, use pl.align_frames() with how="left" to handle the alignment.

Example

df1 = pl.DataFrame({"a": [1, 2], "b": [3, 4]})
df2 = pl.DataFrame({"a": [1, 2], "c": [5, 6]})

# Current behavior
result = pl.concat([df1, df2], how="align")  # Full outer join on common columns

# Proposed behavior
result = pl.concat([df1, df2], how="align_left")  # Left join on keys from df1

Additional Notes

If this feature is feasible, updating the documentation and examples to clarify the difference between how="align" and how="align_left" would be essential.