Unexpected schema for polars Array type of dimensionality > 1

[hsuominen]

Describe the bug
Unexpected shape of polars Array type when dimensionality is greater than 1.

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of pandera.
• (optional) I have confirmed this bug exists on the main branch of pandera.

Code Sample, a copy-pastable example

import polars as pl
import pandera.polars as pa

class Schema(pa.DataFrameModel):
    array: pl.Array = pa.Field(dtype_kwargs={"inner": pl.Int64(), "shape": (2, 2)})

Schema.to_schema().columns["array"].dtype

resolves as DataType(Array(Int64, shape=(2, 2, 2)))

If shape is an int or tuple of size 1 behaviour is as expected. The size of the inferred DataType grows with increasing tuple length, e.g. a tuple with 3 elements yields a datatype with a length of 5.

Expected behavior

Polars evaluation matches the expected behaviour.

pl.Array(pl.Int64(), shape=(2, 2))

returns Array(Int64, shape=(2, 2))

Desktop (please complete the following information):

• OS: macOS Sequoia 15.3
• Python: 3.12
• Pandera version: 0.22.1
• Polars version: 1.22.0

[hsuominen]

A slightly more minimal repro that might point to the source of the issue:

from pandera.engines.polars_engine import Engine
from pandera.engines import engine
import polars as pl

engine.Engine.dtype(Engine, pl.Array(pl.Int64(), shape=(2, 2)))

and one step deeper:

from pandera.engines.polars_engine import Array
Array.from_parametrized_dtype(pl.Array(pl.Int64(), shape=(2, 2)))

which looks like it points to the root cause:

dt = pl.Array(pl.Int64(), shape=(2, 2))

dt.inner = Array(Int64, shape=(2,)) while at the same time dt.shape = (2, 2)

It looks like polars explictly handles tuples in inner differently