Prevent single file storage from performing unnecessary N^2 loop

[eguiraud]

For single file storage we wrap the logic of get_multi with the at_once context manager so that self.list() (which is called by self.get()) actually caches the items rather than re-computing them at every call.

This should largely mitigate the performance issue describe at #818 . The issue discussion also contains more background about this patch.

[WhyNotHugo]

Perfect, thanks!

[WhyNotHugo]

CI is failing because the cache is never initialized during multi-get, even if it's not already.

It always needs to happen if self._items is None.

[WhyNotHugo]

I wonder if the failure scenario is only possible in tests or also happens in some untested real-life scenario.

[eguiraud]

I don't understand how this can happen:

vdirsyncer/storage/singlefile.py:138: in get_multi
    async with self.at_once():
/usr/lib/python3.10/contextlib.py:206: in __aexit__
    await anext(self.gen)
vdirsyncer/storage/singlefile.py:200: in at_once
    self._write()
vdirsyncer/storage/singlefile.py:186: in _write
    text = join_collection(item.raw for item, etag in self._items.values())
E   AttributeError: 'NoneType' object has no attribute 'values'

at_once calls self.list() first thing, and list() calls self._items = collections.OrderedDict(). So I would expect _items to be initialized to an empty dict by the time at_once calls _write?

[WhyNotHugo]

See b1ef680