Handle `pd.NaT` values in lists passed to DataFrame constructor about polars

[ritchie46]

@stinodego or @MarcoGorelli could you take a look?. I am not sure which recent PR's influenced this one.

[MarcoGorelli]

thanks @d-reynol for the report

Did this ever work in previous versions? I just tried it out in 0.20.3 and it also raises there

[ritchie46]

Ok, then we never supported this properly. Sorry for the ping.

[stinodego]

We shouldn't panic here, but I'm not sure we should even support this.

You are passing a Python list with some values in it. We need to determine which type those values are, because a Python list isn't strongly typed (unlike a NumPy ndarray or a pandas Series).

We don't know what type those values are, so we do some checks for various possible values we expect. Looks like we don't currently check for pd.NaT. I guess we could. But you can imagine that checking for every possible type of object someone could throw in a list is not feasible.

My advice is to just use None instead, or create a pandas Series with your pd.NaT values and then convert that to Polars.

I will take a look at our constructors to see if we can support this.

[ritchie46]

We should not check for pd.Nat in conversion to AnyValue.