Horizontal cumulative sum + unnest bug in polars

When I use horizontal cumulative sum followed by unnest, a "literal" column is formed that stays in the schema even when dropped.

Here is an example:

import polars as pl

def test_literal_bug():
    print("Polars version:", pl.__version__)
    
    # Create simple test data
    df = pl.DataFrame({
        "A": [1, 2, 3],
        "T0": [0.1, 0.2, 0.3],
        "T1": [0.4, 0.5, 0.6],
        "T2": [0.7, 0.8, 0.9],
    })
    
    time_cols = ["T0", "T1", "T2"]
    print("Original columns:", df.columns)
    print("Time columns:", time_cols)
    
    lazy_df = df.lazy()

    print("Schema before cumsum:", lazy_df.collect_schema().names())
    
    result = (
        lazy_df.select(pl.cum_sum_horizontal(time_cols))
        .unnest("cum_sum")
        .rename({col: f"C{col}" for col in time_cols})
    )
    
    print("Schema after cumsum:", result.collect_schema().names())
    
    # This will fail with: ColumnNotFoundError: "literal" not found
    try:
        collected = result.collect()
        print("v1: No bug reproduced")
    except pl.exceptions.ColumnNotFoundError as e:
        print(f"v1: BUG REPRODUCED: {e}")

    result_2 = result.drop("literal")
    result_2 = pl.concat([pl.LazyFrame({"B": [1, 2, 3]}), result_2], how="horizontal")

    print("Schema after drop and concat:", result_2.collect_schema().names())

    try:
        collected_2 = result_2.collect()
        print("v2: No bug reproduced")
    except pl.exceptions.ColumnNotFoundError as e:
        print(f"v2: BUG REPRODUCED: {e}")

if __name__ == "__main__":
    test_literal_bug()

Output:

Polars version: 1.31.0
Original columns: ['A', 'T0', 'T1', 'T2']
Time columns: ['T0', 'T1', 'T2']
Schema before cumsum: ['A', 'T0', 'T1', 'T2']
Schema after cumsum: ['CT0', 'CT1', 'CT2', 'literal']
v1: BUG REPRODUCED: "literal" not found
Schema after drop and concat: ['B', 'CT0', 'CT1', 'CT2']
v2: BUG REPRODUCED: "literal" not found

What is going on? Am I doing something wrong or is it a bug?

It is a bug.

1.32.0 has been released which contains a fix: https://github.com/pola-rs/polars/pull/23686

However, I think a new "bug" may have been introduced.

lazy_df.select(pl.cum_sum_horizontal(time_cols)).collect()

# InvalidOperationError: cannot add columns: dtype was not list on all nesting levels: 
# (left: list[str], right: f64)

It seems you cannot pass a list of names now and must unpack them:

lazy_df.select(pl.cum_sum_horizontal(*time_cols)).collect()

shape: (3, 1)
┌───────────────┐
│ cum_sum       │
│ ---           │
│ struct[3]     │
╞═══════════════╡
│ {0.1,0.5,1.2} │
│ {0.2,0.7,1.5} │
│ {0.3,0.9,1.8} │
└───────────────┘

The cum_sum_horizontal source code shows it as a wrapper around cum_fold() which does work with a list on 1.32.0

(lazy_df
  .select(pl.cum_fold(0, lambda x, y: x + y, time_cols))
  .unnest(pl.all())
  .collect()
)

shape: (3, 3)
┌─────┬─────┬─────┐
│ T0  ┆ T1  ┆ T2  │
│ --- ┆ --- ┆ --- │
│ f64 ┆ f64 ┆ f64 │
╞═════╪═════╪═════╡
│ 0.1 ┆ 0.5 ┆ 1.2 │
│ 0.2 ┆ 0.7 ┆ 1.5 │
│ 0.3 ┆ 0.9 ┆ 1.8 │
└─────┴─────┴─────┘