ref(reader): Return date and datetimes instead of strings

[tkaemming]

This code will need to be shared between the HTTP (JSON) reader (#406) and the native reader, so this extracts it from the NativeDriverReader.

Formatting the datetime is object is already handled as part of the JSON serializer in the API:

snuba/snuba/api.py

Line 156 in 4b36a88

default=lambda obj: obj.isoformat() if isinstance(obj, datetime) else obj),

It makes more sense to pass around a datetime object than a string, since this avoids unnecessary type conversions in components such as the splitter.

[manuzope]

How do you feel about having some tests for this function?

[tkaemming]

There should be general coverage via the API tests, but a unit test wouldn't be a bad thing either. At the same time, this is all pretty likely going to go through another round of changes in the near future so I'm hesitant to invest too much time in it.

[manuzope]

Fair 👍

[tkaemming]

There will probably be a change soon to change the type of Result, I promise to add them at that point (and fix the bug with non-scalar values noted here.)

I'm putting this in writing so you can hold me to it. 🙂

[tkaemming]

Out of scope for this change (not a regression), but there is also a bug here:

>>> import pprint
>>> from snuba import settings
>>> from snuba.api import clickhouse_ro
>>> from snuba.reader import NativeDriverReader
>>> 
>>> native = NativeDriverReader(clickhouse_ro)
>>> pprint.pprint(native.execute('SELECT now() FORMAT JSON'))
{'data': [{'now()': datetime.datetime(2019, 8, 14, 18, 22, 54, tzinfo=tzutc())}],
 'meta': [{'name': 'now()', 'type': 'DateTime'}]}
>>> pprint.pprint(native.execute('SELECT [now()] FORMAT JSON'))
{'data': [{'array(now())': (datetime.datetime(2019, 8, 14, 18, 22, 54),)}],
 'meta': [{'name': 'array(now())', 'type': 'Array(DateTime)'}]}