We could e.g. use the first 16 chars off sha256sum of the name :)

For consistency, I would for now use the build in UID generator which also ensures uniqueness. In the long term however basing UIDs from name SHAs could be a valid option.

For consistency, I would for now use the build in UID generator which also ensures uniqueness

There is the problem that if we generate a new UID every time we see the object, it's gonna be recreated/etc. for every new commit. We need "stable" generation of UIDs, hence.
We should also make the UIDs themselves a bit longer IMO

@twelho pointed out that this isn't easily possible with the current architecture. We will move this to v0.5.0 along with the other, larger changes in the storage backend.

I'm gonna move this out; for now it's a reasonable requirement to require the UID being in the files you post to Git.

Just noting here that we can have good default logic for this when the user provides a name but no UID.

However, if a user provides a UID, we should use it.