`Repository.open_with_config`

[TomNicholas]

Problem

Sometimes you want to know about the repo config before you open the repo. For this there is already the classmethod Repository.fetch_config(). But once you have this config, Repository.open() will re-download the config that you already have. This means an unnecessary extra roundtrip to remote storage.

Proposed Solution

A new piece of API that allows opening a repo without re-downloading the config. e.g. Repository.open_with_config.

Additional context

I'm not totally sure what to name this new method - Repository.open() already has a config kwarg, but its meaning is subtly different:

• config in Repo.open means "fetch the stored repo config, merge it with the config passed by the user, then use that merged config to open the repo",
• config in Repo.open_with_config means "don't fetch the stored repo config at all (so ignoring it entirely), and use the config passed by the user to open the repo".

So maybe we call the new method .open_using_config or .open_with_config_override?

In fact maybe a new method isn't required at all, we could just add a new kwarg override_config: bool kwarg to Repository.open...

I also have to say that I don't understand how Repository.reopen() is different.

[paraseba]

I think reopen may be all you need actually. It still merges the passed config with the Repo's one, but if you are passing a full config that's the same as overriding the whole thing.

What you'll need to decide is if first calling open and then reopen is performant enough. open does much more than just reading the config.

One minor issue: what happens if somebody changes the config between the times you read it and call open_with_config? This flow is somewhat unsafe in that way...

[TomNicholas]

What you'll need to decide is if first calling open and then reopen is performant enough.

Surely it's not? Not if I'm trying to minimize the number of sequential fetches from S3. The minimum possible number is the same as the number that open currently does.

reopen won't work because if I first do fetch_config (which is all I need initially), I now don't have a way to pass storage in at reopen-time. Which takes us back to my suggestions above.