commit 7b8a7c85ad7456c5e6181bdae1920ddace8f9044
author Dewey Dunnington <dewey@dunnington.ca> Fri Dec 06 21:51:35 2024 +0000
committer GitHub <noreply@github.com> Fri Dec 06 15:51:35 2024 -0600
tree 7914feaa4db67270d0256ef121deb331a4e921db
parent e54b7df525fa1d310a96687bd99902823402b26c

feat(python): Implement extension type/mechanism in python package (#688)

This PR implements the first canonical extension type for
nanoarrow/Python: `nanoarrow.bool8()`. In doing so it also implements
some machinery for "extensions", which is intended to be internal and to
evolve with the requirements of some follow-up canonical extension
types. The extension points are:

- Parsing the metadata when `Schema.extension` is accessed to get
parameter access (and validate the metadata)
- Converting an extension array to Python objects (i.e.,
`array.to_pylist()`)
- Converting an extension array to a Python sequence (i.e.,
`array.to_pysequence()`)
- Constructing an extension array from Python objects (i.e.,
`na.Array([True, False, None], na.bool8())`
- Constructing an extension array from a Python buffer (i.e.,
`na.Array(np.array([True, False, False]), na.bool8())`

I am not sure the extension point implementations via the `Extension`
methods are the final way this should be done, but this PR at least
connects the wires. The tensor extensions are more complex and will
require some modification of this, but I'd like to do that in another
PR.

```python
import nanoarrow as na
import numpy as np

na.Array([True, False, None], na.bool8())
#> nanoarrow.Array<arrow.bool8{int8}>[3]
#> True
#> False
#> None

na.Array(np.array([True, False, True]), na.bool8())
#> nanoarrow.Array<arrow.bool8{int8}>[3]
#> True
#> False
#> True

na.Array([True, False, None], na.bool8()).to_pylist()
#> [True, False, None]

np.array(na.Array([True, False, True], na.bool8()).to_pysequence())
#> array([ True, False,  True])
```