Re: gc and repack ignore .git/*HEAD when checking reachability

[Josh Triplett <josh@joshtriplett.org>]

On Sat, Jul 09, 2016 at 09:35:24AM +0200, Johannes Schindelin wrote:
> On Fri, 8 Jul 2016, Junio C Hamano wrote:
> > Josh Triplett <josh@joshtriplett.org> writes:
> > 
> > > That sounds reasonable.  And if they *do* end up taking any time to
> > > traverse, it's because they weren't reachable from other anchoring
> > > points, so taking the extra time to traverse them seems fine.
> > 
> > The only thing that is hard is to clearly define _what_ are the new
> > anchoring points.
> > 
> > It cannot be "anything directly under .git that has all-caps name
> > that ends with _HEAD".  The ones we write we know are going to be
> > removed at some point in time (e.g. "git reset", "git bisect reset",
> > "git merge --abort", etc.).  We do not have any control on random
> > ones that the users and third-party tools leave behind, holding onto
> > irrelevant objects forever.
> 
> Please note that bisect already uses the (transient) refs/bisect/
> namespace. So I do not think we need to take specific care of the
> BISECT_* files.
> 
> If we had thought of it back then, we could have used such a transient
> namespace also for FETCH_HEAD, CHERRY_PICK_HEAD and also for detached
> HEADs (which we should have called "unnamed branches").
> 
> Now, how about special-casing *just* these legacy files in gc: HEAD,
> FETCH_HEAD, MERGE_HEAD and CHERRY_PICK_HEAD? Any new transient refs should
> live in the refs/ namespace, which is already handled.

That seems workable as well; in that case, we should also document this
(in the git-gc manpage at a minimum), and explicitly suggest creating
refs in refs/ but outside of refs/heads/ and refs/tags/, rather than
directly in .git/.

> BTW this issue is getting much more problematic when you have a lot of
> worktrees, some of which operate on detached HEADs. Which I do.
> 
> Ciao,
> Dscho