Isn't it enough to check the first position? I think mixed positions are not allowed anyway. Also, do we need to handle np.datetime64 arrays?

@hasanrashid You've marked this as resolved. Please comment how my above questions are addressed.

@timhoffm Yes, it is enough to check the first position. The position argument can't mix datetime with float. width can't mix timedelta and float either.

I was approaching this by converting entries into numeric equivalent until it was suggested that I provide clearer message instead.

Also, yes, datetime64 isn't needed either. I was thinking of all test cases, but I realize datetime64 isn't a supported type in violinplot.

Making the adjustments

Why is datetime64 not supported? Should we rather fix this? (Can be later, but already add the check here?)

@timhoffm

If I try a testcase like this:

def test_np_datetime64_positions_with_float_widths_behaves_like_numeric():
    """Test that np.datetime64 positions with float widths do not raise TypeError (treated as numeric)."""
    fig, ax = plt.subplots()
    try:
        vpstats = violin_plot_stats()
        positions = [np.datetime64('2020-01-01'), np.datetime64('2021-01-01')]
        widths = [0.5, 1.0]
        # Should not raise TypeError, but may raise downstream if not supported
        ax.violin(vpstats, positions=positions, widths=widths)
    finally:
        plt.close(fig)

This produces the following error:
numpy._core._exceptions._UFuncBinaryResolutionError: ufunc 'add' cannot use operands with types dtype('float64') and dtype('<M8[D]')

I took this to mean that datetime64 was not an expected type of datetime for biolinplots- thus unsupported.

If we need to investigate adding datetime64 support in violineplot, should it be another issue? We are updating the error message enforcing timedelta with datetimes; datetime64 support- to me- seems like it should be investigated in a separate issue (which I am happy to open and work on)