Testing: Add type annotations to common/logging

[rdimaio]

Part of #6588

On the ECS_FIELDS and LOG_RECORDS change, here's an execution of that code to show that the final BUILTIN_FIELDS is the same as before:

>>> from typing import TYPE_CHECKING, Any, Literal, Optional, Union, get_args
>>> ECS_FIELDS = Literal[
...     '@timestamp',
...     'message',
...     'log.level',
...     'log.origin.function',
...     'log.origin.file.line',
...     'log.origin.file.name',
...     'log.logger',
...     'process.pid',
...     'process.name',
...     'process.thread.id',
...     'process.thread.name'
... ]
>>>
>>> LOG_RECORDS = Literal[
...     'asctime',
...     'message',
...     'levelname',
...     'funcName',
...     'lineno',
...     'filename',
...     'name',
...     'process',
...     'processName',
...     'thread',
...     'threadName'
... ]
>>>
>>> BUILTIN_FIELDS = tuple((x, y) for x,y in zip(get_args(ECS_FIELDS), get_args(LOG_RECORDS)))
>>> BUILTIN_FIELDS
(('@timestamp', 'asctime'), ('message', 'message'), ('log.level', 'levelname'), ('log.origin.function', 'funcName'), ('log.origin.file.line', 'lineno'), ('log.origin.file.name', 'filename'), ('log.logger', 'name'), ('process.pid', 'process'), ('process.name', 'processName'), ('process.thread.id', 'thread'), ('process.thread.name', 'threadName'))

[rdimaio]

The type ignore here is to ignore the error that occurs when a str that's not in ECS_FIELDS is passed to _ecs_field_to_record_attribute, in which case the function returns field_name.replace('-', '_').replace('.', '_').

I personally don't like this logic being in this function, as the function name doesn't say anything about replacing those symbols within a string, and from the function name _ecs_field_to_record_attribute you would assume that this only takes in ECS_FIELDS and returns LOG_RECORDS.

I think it would be a good idea to change the way this works in future.

[rdimaio]

The type: ignore here is due to the fact that LogDataSource.format returns None if there is no self._formatter, and the static type checker is not able to guarantee that f has a _formatter specified in this case.

This logic should likely be changed so that it only runs LogDataSource.format if _formatter exists, and otherwise returns None, but I think it's best to justtype: ignore for now and address this later.