[Mohit-Ak]

Description of proposed changes

sequence_join decided whether an item was datetime-like by checking for a space in str(item), so any plain string containing a space was fed to np.asarray(item, dtype="datetime64") and raised ValueError: Cannot create a NumPy datetime other than NaT with generic units. Every Alias(..., sep=...) goes through this, so magnetic_rose/directional_rose crashed on free-text labels like labels=["W", "E", "S", "Pole Nord"] before GMT was called.

This replaces the space check with a type check (datetime.date/datetime.datetime/pd.Timestamp, or anything whose array dtype is datetime64, which covers np.datetime64, 0-d arrays and xr.DataArray). Plain strings are passed through unchanged. GMT itself accepts a space inside +l labels when the option is passed as a separate argv entry, so no escaping is needed.

Tested with the new test_sequence_join_strings_with_spaces (fails on main, passes here), the sequence_join/alias doctests, test_helpers, test_alias_system, and the magnetic/directional rose image tests, plus ruff check, ruff format --check and mypy on the changed files.

Fixes #4960

[seisman]

@Mohit-Ak Are you interested in implementing the solution in #4960 (comment)?

[Mohit-Ak]

@seisman thanks, done: switched to your try/except approach, so pandas timestamps and datetime strings with spaces still go through np.asarray(..., dtype="datetime64"), and plain labels like "North Pole" fall back to str.
Added tests for the label, a datetime string with a space and a pd.Timestamp. A pyarrow timestamp scalar falls back to str (it used to raise); happy to add explicit handling if you want it here.

[weiji14]

Actually, can these two branches be collapsed somehow?

[seisman]

Yes, I feel the codes like below is more readable:

        _values = []
        for item in value:
            _value = str(item)
            if isinstance(item, np.timedelta64):  # Convert timedelta64 to numeric value
                _value = str(item.astype(int))
            elif " " in _value:  # Likely a datetime-lik object.
                with contextlib.suppress(ValueError):
                    _value = np.datetime_as_string(np.asarray(item, dtype="datetime64"))  # type: ignore[arg-type]
            _values.append(_value)

[Mohit-Ak]

Done, switched to your version; helper tests and doctests still pass.

[seisman]

/format