Re: git-bug: Distributed bug tracker embedded in git

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Aug 18, 2018, 22:50 UTC
Message-ID: <20180818225052.GE144170@aiede.svl.corp.google.com>
In-Reply-To: <CACSZ0Pwzs2e7E5RUEPDcEUsa=inzCyBAptU7YaCUw+5=MutSsA@mail.gmail.com>

(cc-ing Elijah Newren for the points about merging) Hi again,

To avoid the other thread shadowing more important things:

Michael Muré wrote:

> Someone suggested in the Hacker News thread [0] to post it here as well.

Thanks to Ævar for that.

[...]

> git-bug use as identifier the hash of the first commit in the chain
> of commit of the bug.

Clever!  I like this approach to the naming problem.

[...]

> Git doesn't provide a low-level command to rebase a branch onto
> another without touching the index.

Thanks for pointing this out. There's been some recent work to make Git's merge code (also used for cherry-pick) less reliant on the index and worktree. See https://crbug.com/git/12 for some references. There's also been some heavy refactoring of "git rebase" code to be in C and be able to make use of library functions instead of being a shell script.

That's all to say that we're in a pretty good place to consider introducing commands like

  git cherry-pick --onto=<branch> <revisions>

In absence of that kind of thing, you can run commands that need to touch the index (but not the working tree) by setting the GIT_INDEX environment variable to point to a temporary index file.

> I'd love to have some feedback from you. Contribution are also very
> much welcomed.

Can you say more about the federation model it intends to support? For example, do you imagine

- having multiple copies of a git bugs repo that automatically fetch
  updates from each other

- having explicit "pull request" synchronization moments when the
  owners of one copy of a bug tracker push or request a fetch of
  changes that have been happening on another

- individual contributors using an offline copy of the bug tracker
  and pushing push/pull mostly to synchronize with a single
  centralized copy

- something else?

Thanks, Jonathan