BUG: Timestamp(pd.NA) raising TypeError

[lukemanley]

• closes BUG: Cannot convert pd.NA to Timestamp #45481
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/v2.1.0.rst file if fixing a bug or adding a new feature.

[MarcoGorelli]

can the

        if ts.value == NPY_NAT:
            return NaT

be removed from below now?

[lukemanley]

It doesn't look like it can be removed. Just tried it and this starts to raise if its removed:

pd.Timestamp(np.datetime64("NaT"))

[MarcoGorelli]

ah right, thanks, checknull checks for pandas NaT, not numpy NaT

does it work to do the check a bit deeper? i.e.

diff --git a/pandas/_libs/tslibs/conversion.pyx b/pandas/_libs/tslibs/conversion.pyx
index 57b7754b08..4d494d2282 100644
--- a/pandas/_libs/tslibs/conversion.pyx
+++ b/pandas/_libs/tslibs/conversion.pyx
@@ -1,4 +1,5 @@
 import numpy as np
+from pandas._libs.missing cimport checknull_with_nat_and_na
 
 cimport numpy as cnp
 from libc.math cimport log10
@@ -268,7 +269,7 @@ cdef _TSObject convert_to_tsobject(object ts, tzinfo tz, str unit,
     if isinstance(ts, str):
         return convert_str_to_tsobject(ts, tz, unit, dayfirst, yearfirst)
 
-    if ts is None or ts is NaT:
+    if checknull_with_nat_and_na(ts):
         obj.value = NPY_NAT
     elif is_datetime64_object(ts):
         reso = get_supported_reso(get_datetime64_unit(ts))
diff --git a/pandas/_libs/tslibs/timestamps.pyx b/pandas/_libs/tslibs/timestamps.pyx
index 80b7c2d2f5..59640a3ef4 100644
--- a/pandas/_libs/tslibs/timestamps.pyx
+++ b/pandas/_libs/tslibs/timestamps.pyx
@@ -1882,9 +1882,6 @@ class Timestamp(_Timestamp):
                                 hour or 0, minute or 0, second or 0, fold=fold or 0)
             unit = None
 
-        elif checknull_with_nat_and_na(ts_input):
-            return NaT
-
         if getattr(ts_input, "tzinfo", None) is not None and tz is not None:
             raise ValueError("Cannot pass a datetime or Timestamp with tzinfo with "
                              "the tz parameter. Use tz_convert instead.")

?

[lukemanley]

yes - updated. thanks

[jbrockmendel]

Is it obvious we want to allow this? If we consider pd.NA as meaningfully distinct from pd.NaT, then silently converting would be wrong

[MarcoGorelli]

I would've thought so - the docs say

For datetime64[ns] types, NaT represents missing values

and pd.NA is what signals missing values. Will check what's done in other cases, not really sure

[jbrockmendel]

I would've thought so - the docs say [...] and pd.NA is what signals missing values. Will check what's done in other cases, not really sure

The meaning of "missing" has changed since that doc was written. In particular NaT has nan-like semantics.

[lukemanley]

Is it obvious we want to allow this? If we consider pd.NA as meaningfully distinct from pd.NaT, then silently converting would be wrong

I dont have a strong opinion but I'll note:

1. Other na-likes already get converted to NaT (e.g. None -> NaT, NaN -> NaT)

2. The current behavior does seem inconsistent:

pd.to_timedelta(pd.NA)  # -> NaT
pd.to_datetime(pd.NA)   # -> NaT
pd.Timedelta(pd.NA)     # -> NaT
pd.Timestamp(pd.NA)     # -> TypeError

[jbrockmendel]

The current behavior does seem inconsistent:

fair enough. no objection here.

[MarcoGorelli]

cool, going to go ahead and approve then, thanks @lukemanley !

[mroeschke]

This might cause issue if we every introduced a nullable numpy datetime/timedelta types with pd.NA as the missing value, but that's not planned right?

[jbrockmendel]

This might cause issue if we every introduced a nullable numpy datetime/timedelta types with pd.NA as the missing value, but that's not planned right?

Does the numpy-nullable part matter here? timestamp[pyarrow] already behaves that way

[mroeschke]

Does the numpy-nullable part matter here? timestamp[pyarrow] already behaves that way

Sure, I was just confirming that there wasn't appetite to have BaseMaskedArray support datetime/timedelta types for example

[mroeschke]

Thanks @lukemanley