IntelPython / dpnp / 32233785766

78%

push

github

Add proper exception handling for nested lists in `dpnp.repeat` (#3024)

This PR fixes #2972.

Passing a nested sequence as `repeats` to `dpnp.repeat` (e.g.
`np.repeat(x, [[4]])`) raised an unclear `TypeError` instead of a
meaningful error:

```python
>>> import dpnp as np
>>> x = np.array([3])
>>> np.repeat(x, [[4]])
TypeError: '<' not supported between instances of 'list' and 'int'
```

The sequence branch of `dpnp.tensor.repeat` special-cased a length-1
sequence by taking `repeats[0]` and comparing it to `0`, which fails
when that element is itself a list. It also had no dimensionality guard,
so a multi-element nested sequence could slip through as a 2-D array —
unlike the `usm_ndarray` branch, which already rejects `ndim > 1`.

The fix converts the sequence to an array up front and rejects a
dimensionality greater than 1, aligning with NumPy, which raises a
`ValueError` in this case:

```python
>>> np.repeat(x, [[4]])
ValueError: `repeats` sequence must be 0- or 1-dimensional, got 2 dimensions
```

The scalar fast path (`_repeat_by_scalar`) is preserved for the size-1
case.

1546 of 2898 branches covered (53.35%)

Branch coverage included in aggregate %.

7 of 7 new or added lines in 1 file covered. (100.0%)

26556 of 32920 relevant lines covered (80.67%)

8490.28 hits per line