RFC: dealing with dead, anonymous feature branches

[Augie Fackler]

On Mon, May 17, 2010 at 8:57 AM, Martin Geisler <mg at aragost.com> wrote:
> Augie Fackler <durin42 at gmail.com> writes:
>
>> On May 17, 2010, at 3:10 AM, Martin Geisler wrote:
>>
>>> The core idea is simple: do not push/pull/clone dead branches.
>>>
>>> I don't know if people will see this as just a halfhearted way of
>>> supporting remove deletions, as if we couldn't figure out how to do a
>>> "proper" server-side deletion.
>>>
>>> I just tried to figure out how Git supports this. There you need to
>>> push your branch name with a colon (?!) and others need to run a
>>> prune command from what I can tell:
>>>
>>>  http://github.com/guides/remove-a-remote-branch
>>>
>>> I think the proposed solution would be more user friendly. We will
>>> have to take care to warn users when they pull in a changesets that
>>> causes the working copy parent changeset to be dead too.
>>
>> The git solution is rather different, as removing the branch from the
>> repo means that all the changes that were referenced only by that
>> branch get garbage collected, but that won't cause git to lose any tag
>> information.
>
> So Git wont garbage collect a changeset if it is referenced by a tag?
> Seems quite sensible :-)
>
>> We run the risk of losing tags if we stop transferring some branches.
>
> I see... You must be talking about the situation where we have
>
>  [A] --- [B] --- [T] --- [C]
>             \
>              [X] --- [Y] --- <Z>
>
> and T is the changeset that adds a tag for Y and Z is dead. If I clone
> this, I will only get A, B, T, and C and the .hgtags file will still
> reference the unknown Y.

Oh, hadn't even thought of that situation. I was more worried about T
adding a tag for Y but C is dead.

>
> Perhaps this suggests that it is a very bad idea to make a tag on one
> branch that refers to a changeset on another branch (again, "branch"
> means topological branch here). The normal case where you do
>
>  hg tag 1.0
>
> to tag the current working copy parent revision works fine -- if you
> have the tag, then you also have the changeset it refers to.
>
> So can we forbid tags that "cross branches" like this?

Not sure. hgsubversion uses the fact that closed branches hang around
in order to emulate Subversion's mutable tags, so there are converted
repositories that probably look something like this...

>
> --
> Martin Geisler
>
> aragost Trifork
> Professional Mercurial support
> http://aragost.com/mercurial/
>