BREAKING CHANGE: Change the behaviour of git_repository_is_empty to better align with expectations in a world where master isn't always the "initial branch"

[timrogers]

This pull request proposes a change to the definition of an "empty" repository used in git_repository_is_empty.

At present, an empty repository is one which:

1. has just been initialized
2. has no references apart from HEAD
3. has a HEAD pointing to the unborn master branch or, if specified, the unborn branch specified in the init.defaultBranch configuration value

This pull request updates the definition of "empty", removing requirement 3.

Why? Since git_repository_is_empty was first implemented, there has been significant cultural change in the Git community towards using an "initial branch" other than master.

Because of this change, the current implementation of git_repository_is_empty considers repos to be not empty which an ordinary observer would consider to be empty.

Specifically, this happens where a repo is initialized with git init --initial-branch <initial_branch_which_is_not_master>.

An ordinary observer would consider such a repo to be empty, but currently, libgit2 considers it to be non-empty if the initial-branch is not master.

This is because, whilst git_repository_is_empty checks the init.defaultBranch to try to figure out the initial branch, git init --initial-branch doesn't set this value, as you can prove like this:

git config init.defaultBranch
# nothing is printed to STDOUT
mkdir test-repo
cd test-repo
git init --initial-branch main
git config init.defaultBranch
# nothing is printed to STDOUT, despite the `initial-branch` option

I'm not necessarily expecting this PR to be merged as is or ever - but I want to start a conversation about this, given that the current implementation is surprising and likely confusing.

See #6049.

[timrogers]

The tests aren't working yet, but happy to get them working if this is a proposal worth moving forward with.

[ethomson]

Thanks @timrogers. I haven't read the code yet but agree that is_empty... doesn't make much sense as-is.

[ethomson]

I'd be curious for @carlosmn's take since I suspect he's using it production (and I am not).

[timrogers]

Thanks @timrogers. I haven't read the code yet but agree that is_empty... doesn't make much sense as-is.

Thanks! I'm conscious that breaking changes are a big deal, but this behaviour has caused some confusion for our team, and I'm sure we're not alone. I think we should do something to make this better - but I'd love the maintainers' thoughts on the right path.

[timrogers]

@carlosmn 👋🏻 Just a reminder that it would be great if you could take a look at this!