Re: [PATCH 7/7] push: document --lockref

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Jul 15, 2013, 04:44 UTC
Message-ID: <20130715044454.GA2962@elie.Belkin>
In-Reply-To: <7v4nbwfooj.fsf@alter.siamese.dyndns.org>

Junio C Hamano wrote:

> Jonathan Nieder <jrnieder@gmail.com> writes:

Show 11 quoted lines

>> Then I will use
>>
>> 	(5c) git push --force
>>
>> which means not to use this new lockref trick that looks at my
>> remote-tracking branch and instead to just force the ref update.
>
> I am not sure I follow.  Do other contributors update this remote
> repository?  They are "only using fast-forward updates", so their
> updates may not lose anything we pushed, but with "--force", aren't
> you losing their work on top of yours?

Yep, I meant that when you really *do* want to force a push regardless of what's on the remote end, the current --force behavior is more useful than --lockref.

The example I used to introduce (5c) is too vague to be useful. A more compelling example (to me, at least) is the one from later in that message involving a relay, which does not involve other contributors at all.

That is, suppose I maintain a mirror of the branches from git://repo.or.cz/git.git by pushing regularly to a hosting service where I do not have shell access. Since I can't fetch from the target repository or push from the source, I instead fetch and then push from a relay, like this:I might push like this:

	git fetch upstream
	git push --force origin refs/remotes/upstream/*:refs/heads/*

Or, in the same spirit, with a detached HEAD:

	git fetch upstream refs/heads/*:refs/heads/*
	git push --force origin :

The --force is to account for "pu" and "next" rewinding.

In this scenario, assuming I have exclusive access to the repository and the push updates the remote-tracking branches, --lockref and --force work equally well. The commands might run once every 6 hours using a cronjob.

Now suppose my relay has some downtime. That's fine --- I can still maintain the mirror by running the same commands on another machine. But when the old relay comes back up, "push --lockref" will fail and "pu" and "next" in my mirror are not updated any more.

That is why I said that --force is more appropriate than --lockref for this application.

Thanks, Jonathan