Add projection argument to Reference.fetch()

[whophil]

Adds projection argument to Reference.fetch(), allowing partial reference documents to be returned.

Addresses half of issue #196

[lafrech]

I left a minor comment.

The issue you point to points to #98 which raises concerns about projection and caching mechanism. And the fact that DB names should be expected (unless umongo tries to translates field names into DB names but I think it wouldn't be the only place where DB names are needed).

In any case, you may merge this if you're happy with it. If we can't have a perfect solution, this is better than nothing. It would be nice if known limitations were documented, though.

[whophil]

Thanks @lafrech. I had actually not seen the discussion on issue #98. Let me think over this a bit and potentially rework.

[whophil]

@lafrech I've implemented cook_find_projection, as suggested in #98, so that the projections can use uMongo field names and not DB names.

Other things discussed in #98:

• The cached _document is just the first-fetched document, not a cache which is looked up by the projection. The projection-keyed cache may still be useful in the future. If implemented, it should have a configurable max size.
• _reference_io_validate is untouched (it still uses .fetch()). I implemented Reference.exists for PyMongo and Motor, and modified the corresponding _reference_io_validate to use them. I did not implement for TxMongo, as I ran into some environment issues (see Import error with PyMongo==4.1.1. twisted/txmongo#278).

[whophil]

FYI, the suggestion to not use.fetch() in _reference_io_validate makes a huge difference in performance in some cases.

In my application, I have a parent document which references over a thousand others. The time to .commit() the parent document went from 21.8 seconds to 1.5 seconds (PyMongo with mongo running on localhost).

[lafrech]

LGTM!

You may merge and release!

[whophil]

@lafrech how do I make the CI "go" on this?

[lafrech]

That's a good question. I see no obvious reason. Hooks are enabled. Can't look into this right now.

Frankly, Azure CI is a nightmare to configure and rather than trying to fix this I'd be tempted to move to GitHub actions. You may give it a try in another PR if you want. Setting up tests is relatively straightforward. See what I did on flask-smorest. The tricky part would be launching MongoDB.

---

OK I peeked into Azure logs and CI did go but apparently didn't comment here with the results. Note sure if the pipeline logs there are public.

Here's the last result:

./tests/conftest.py:38:101: E501 line too long (106 > 100 characters)
./umongo/frameworks/motor_asyncio.py:415:101: E501 line too long (105 > 100 characters)

[whophil]

OK I peeked into Azure logs and CI did go but apparently didn't comment here with the results. Note sure if the pipeline logs there are public.

I don't know where to find the link to these results, so I can't tell you if they're public or not :)

I fixed the too-long lines. Merging now!

As far as a release goes - what is the process on this project?