Improper conversion of Pandas-Arrow date to datetime fails

[s-banach]

The Issue

Consider this branch of code:

plotly.py/packages/python/plotly/_plotly_utils/basevalidators.py

Line 103 in 956ab2c

elif v.dtype.kind == "M":

It attempts to call v.dt.to_pydatetime() on a Series with any kind of date or time dtype.

This isn't correct, because a date dtype is different than a datetime dtype, see the comment from pandas maintainer @mroeschke on this issue: pandas-dev/pandas#52812. In other words, this is unsupported by pandas, and will continue to raise an error in the future.

You should probably just use v.to_numpy() instead.

Reproducing Code

The following code results in the error AttributeError: 'pyarrow.lib.DataType' object has no attribute 'unit', as a result of the above bug.

import datetime

import plotly.express as px
import polars as pl

df = pl.DataFrame(
    {
        "x": [datetime.date(2020, n, 1) for n in range(1, 13)],
        "y": [3 * n % 13 for n in range(1, 13)],
    }
).to_pandas(use_pyarrow_extension_array=True)

fig = px.bar(data_frame=df, x="x", y="y")
fig.show()

Note that the following code works, by properly calling .to_numpy() and letting pandas figure out how to handle it:

fig = px.bar(x=df["x"].to_numpy(), y=df["y"].to_numpy())

[MarcoGorelli]

@emilykl I think this can be closed now, as per discussion in #4836, if I run this code now with the latest plotly build then the figure displays correctly