After I update Polars from 26 to 29, I found that there is an odd result from the code I wrote before (using .list.sum()). The following are trying to demo this issue:

import polars as pl

x = pl.DataFrame(
    {
        'A': [1., None, 3.],
        'B': [4., 5., 6.],
        'C': [7., 8., None],
    }
)

x.with_columns(
    pl.when(pl.sum_horizontal('B', 'C') > 12)
    .then(pl.sum_horizontal(pl.col('A', 'B')))
    .otherwise(None)
    .alias('A+B when B+C>12 (expected)'),
    pl.when(pl.sum_horizontal('B', 'C') > 12)
    .then(pl.concat_list(pl.col('A', 'B')).list.sum())
    .otherwise(None)
    .alias('A+B when B+C>12 (odd)'),
    pl.concat_list(pl.exclude('C')).list.sum().alias('A+B'),
)

Run the above I got the following:

┌──────┬─────┬──────┬────────────────────────────┬───────────────────────┬─────┐
│ A    ┆ B   ┆ C    ┆ A+B when B+C>12 (expected) ┆ A+B when B+C>12 (odd) ┆ A+B │
│ ---  ┆ --- ┆ ---  ┆ ---                        ┆ ---                   ┆ --- │
│ f64  ┆ f64 ┆ f64  ┆ f64                        ┆ list[f64]             ┆ f64 │
╞══════╪═════╪══════╪════════════════════════════╪═══════════════════════╪═════╡
│ 1.0  ┆ 4.0 ┆ 7.0  ┆ null                       ┆ null                  ┆ 5.0 │
│ null ┆ 5.0 ┆ 8.0  ┆ 5.0                        ┆ [5.0]                 ┆ 5.0 │
│ 3.0  ┆ 6.0 ┆ null ┆ null                       ┆ null                  ┆ 9.0 │
└──────┴─────┴──────┴────────────────────────────┴───────────────────────┴─────┘

The last two columns have the same kind of calculations .list.sum(): one returns list[f64]: [5.0] (the old version Polars gives me f64: 5.0 which is expected), while the other returns f64: 5.0, which is expected.

Is it natural for the aggregation functions like .list.sum() to always return a single value other than a list?