support timedelta in Q_.init

[andrewgsavage]

• Closes Convert timedelta to quantity #608 Timedeltas Quantities not acting as expected #690
• Executed pre-commit run --all-files with no errors
• The change is fully covered by automated unit tests
• Documented in docs/ as appropriate
• Added an entry to the CHANGES file

[keewis]

I've got a couple of comments, mostly regarding timedelta64 and where to put the code handling it.

I would also like to have a way to convert a quantity with dimension [time] to a timedelta64, but that can be its own PR (the API is not quite clear to me: should to special-case units of timedelta64[us], for example?).

[keewis]

would it be possible to special-case numpy.timedelta64, probably in the numpy facet? That way we could roundtrip exactly

[keewis]

this is not going to work in general: timedelta64 has its own units, which can range from days (or years?) to nanoseconds, and astype(float) is the equivalent of magnitude. I don't have any advice on how to extract that, though, other than parsing the string repr of the dtype.

[jakirkham]

It looks like this is ready for another review

[codspeed-hq]

CodSpeed Performance Report

Merging #1978 will not alter performance

_{Comparing andrewgsavage:infer_time (92e70a1) with master (6e0d038)}

Summary

✅ 448 untouched benchmarks

[andrewgsavage]

poke @keewis

[andrewgsavage]

I would also like to have a way to convert a quantity with dimension [time] to a timedelta64, but that can be its own PR (the API is not quite clear to me: should to special-case units of timedelta64[us], for example?).

I think to do we should make time into a nonmultiplactive unit like temperature, so there is then a delta_time unit
the new c++ library does this for all units with two different classes, a point and a vector
#2120

[keewis]

sorry for disappearing for quite a while.

Looks good to me in general, but I have two comments.

[keewis]

I wonder if this should rely on is_duck_array_type? That would already guarantee the existence of dtype. Also, this will return None if the condition evaluates to False:

Suggested change

	if isinstance(obj, ndarray) and obj.dtype.type == np_timedelta64:
	return True
	return is_duck_array_type(obj) and obj.dtype.type == np_timedelta64

[keewis]

it feels somewhat weird to have the plain facet deal with timedelta arrays. Should that part be in the numpy facet?

Then again, I admit I don't fully grasp the concept of facets.