Thanks for tackling some more in-memory work. Exciting!

I think that this is fallout from the fact that we never really decided how to support building a wholly in-memory repository and add backends to it. I don't remember all the details offhand, so I'll have to debug into this - but this is something that libgit2 itself supports (and tests) so I think the problem is that we're trying to use something before we've fully configured it.

So i looked into this some today.

Appears for a in-memory repository that _odb is null (https://github.com/libgit2/libgit2/blob/master/src/repository.c#L1073) and thus when the lazy ObjectDatabase is constructed in the Repository constructor (for the ObjectDatabase property) it explodes in the above test.
Seems like a bit of a chicken/egg scenario. So there may have to be some changes around Repository and ObjectDatabase that is needed.

The easiest way for me to get it working is a breaking change, property not desired. (I just hacked this together while debugging the issue to figure out what it was)

I altered the RepositoryOptions to accept a defaultOdb so inside the constructor it can set the defaultOdb that will then be picked up by the ObjectDatabase constructor Proxy.git_repository_odb call.

Scratch those code changes/thoughts above. I think I have it figured out. ObjectDatabase needs to call git_odb_new then git_repository_set_odb when its bare inmemory. (Should be no breaking changes) I'll have a PR for that in a few minutes.

@Zoxive FYI I've rewritten your PR to target the current libgit2 API #1739 . @ethomson I tried to hew closely to how refdb backends in libgit2 are plugged in, but I'm new to this repo so I could use guidance.