make Timestamp.strptime() return None

[Dr-Irv]

mypy_nightly started failing because typeshed got updated to have datetime.strptime() return Self, and since Timestamp is a subclass of datetime, the nightly mypy build was saying that the datetime result of strptime() was invalid since it should return Timestamp if the superclass returns Self.

But Timestamp.strptime() is not implemented, and we always raise, so changed the arguments to Never meaning that type checkers will always say it is invalid to call this method.

[twoertwein]

Has typeshed documented how to handle such cases (using Never, NotImplementd, ...)? I did a quick search but I didn't find a comment on that.

[Dr-Irv]

Has typeshed documented how to handle such cases (using Never, NotImplementd, ...)? I did a quick search but I didn't find a comment on that.

I didn't look. But this all depends on whether you want to strongly enforce the Liskov substitution principle. mypy does, while pyright does not.

In this particular case, we don't want people to call Timestamp.strptime() . Right now, if you do, an exception is raised. If we remove the declaration of strptime() from timestamps.pyi, then since Timestamp is a subclass of datetime.datetime, both mypy and pyright will say that a call to Timestamp.strptime() is valid, because it will find the stub for datetime.datetime.strptime(). By using the Never trick, we are telling the type checker that this is an invalid function to call, and I think that is the best user experience.

[twoertwein]

I didn't look. But this all depends on whether you want to strongly enforce the Liskov substitution principle. mypy does, while pyright does not.

(I prefer pyright's approach, while mypy's approach is theoretically better - it is just not compatible with python: everything inherits from object - but some classes violate its __hash__). I definitely like informing users of the error (instead of adding a type ignore in pandas-stubs), but I wasn't sure whether there is a "typeshed-recommended" way to do so.