int_ranges: PanicException and InvalidOperationError

[Julian-J-S]

Checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

(
    pl.DataFrame({
        'asc': [1, 2, 3],
        'desc': [3, 2, 1],
        'zero': [0, 0, 0],
        'one': [1, 1, 1],
        'two': [2, 2, 2],
    })
    .with_columns(
        zero_asc=pl.int_ranges(start='zero', end='asc'),
        zero_desc=pl.int_ranges(start='zero', end='desc'),

        asc_desc=pl.int_ranges(start='asc', end='desc'),
        desc_asc=pl.int_ranges(start='desc', end='asc'),

        asc_two=pl.int_ranges(start='asc', end='two'),

        # <<< InvalidOperationError: range must be decreasing if 'step' is negative >>>
        # asc_desc_down=pl.int_ranges(start='asc', end='desc', step=-1),
        # desc_asc_down=pl.int_ranges(start='desc', end='asc', step=-1),

        # <<< # PanicException: capacity overflow; [thread '<unnamed>' panicked at library/alloc/src/raw_vec.rs:534:5:] >>>
        # asc_zero_down=pl.int_ranges(start='asc', end='zero', step=-1),
        # asc_one=pl.int_ranges(start='asc', end='one'),
    )
)
┌─────┬──────┬──────┬─────┬───┬───────────┬───────────┬───────────┬───────────┐
│ asc ┆ desc ┆ zero ┆ one ┆ … ┆ zero_desc ┆ asc_desc  ┆ desc_asc  ┆ asc_two   │
│ --- ┆ ---  ┆ ---  ┆ --- ┆   ┆ ---       ┆ ---       ┆ ---       ┆ ---       │
│ i64 ┆ i64  ┆ i64  ┆ i64 ┆   ┆ list[i64] ┆ list[i64] ┆ list[i64] ┆ list[i64] │
╞═════╪══════╪══════╪═════╪═══╪═══════════╪═══════════╪═══════════╪═══════════╡
│ 1   ┆ 3    ┆ 0    ┆ 1   ┆ … ┆ [0, 1, 2] ┆ [1, 2]    ┆ []        ┆ [1]       │
│ 2   ┆ 2    ┆ 0    ┆ 1   ┆ … ┆ [0, 1]    ┆ []        ┆ []        ┆ []        │
│ 3   ┆ 1    ┆ 0    ┆ 1   ┆ … ┆ [0]       ┆ []        ┆ [1, 2]    ┆ []        │
└─────┴──────┴──────┴─────┴───┴───────────┴───────────┴───────────┴───────────┘

Log output

No response

Issue description

InvalidOperationError: range must be decreasing if 'step' is negative

if ranges 1->3 = [1, 2] and 3->1 = []
then with step=-1 should be: 1->3 = [] and 3->1 = [3, 2] and not error

PanicException: capacity overflow

going from 1/2/3 down to zero with step=-1 should work and not panic
1->0 = [1], 2->0 = [2, 1], 3->0 = [3, 2, 1]
going from 1/2/3 to 1 should work and not panic
1->1 = [], 2->1 = [], 3->1 = []

Expected behavior

All examples should work without exception or panic

Installed versions

Replace this line with the output of pl.show_versions(). Leave the backticks in place.

[mcrumiller]

Sorry, do you mean that we should return nulls in cases where the operation is invalid, and return value elsewhere?

[Julian-J-S]

Sorry, do you mean that we should return nulls in cases where the operation is invalid, and return value elsewhere?

no ;)

We discussed this here #11892 and we all agreed that <xxx>_ranges should behave like python/rust ranges like this:

• no exceptions / no nulls (given that data types are correct)
• for ascending ranges:
  • 3-6 = [3, 4, 5]
  • 6-3 = [] end is lower than start: no problem, there is no element in between so empty list
• for descending ranges
  • 3-6 = [] end is higher than start: no problem, there is no element in between so empty list
  • 6-3 = [6, 5, 4]

[alexander-beedie]

Closed/fixed by #12548 👌