Re: [BUG] git-new-workdir doesn't understand packed refs

[Peter Baumann <waste.manager@gmx.de>]

On Wed, Apr 18, 2007 at 11:42:24AM -0700, Junio C Hamano wrote:
> Peter Baumann <waste.manager@gmx.de> writes:
> 
> <ot>
> 
> Getting more and more annoyed by your stupid Mail-Followup-To...
> I do *not* want to bother Julian with a message that points out
> a flaw (in my opinion) in YOUR reasoning but you are forcing me
> to send my message that way, which I have to waste time
> correcting every time.  Grumble.
> 
> </ot>

Hm. Sorry. I don't understand. I'm just pressing 'g' for group reply in
mutt which should do the right thing; even your mail has a CC to Julian
set so I _really_ don't understand the problem. I addressed him in the
begining because he was the author of git-new-workdir. But please
forgive me if I'm breaking some netiquette rules but I just started to
hang out activly on mailinglists ...

> 
> > On Wed, Apr 18, 2007 at 11:17:43AM -0700, Junio C Hamano wrote:
> >> Peter Baumann <waste.manager@gmx.de> writes:
> >> ...
> >> > I thought about the case where packed-refs is a symlink to another symlink
> >> > and then decided that it's not worth to implement this because a workdir
> >> > should be linked to a _repo_ and not another workdir.
> >> 
> >> That's incredibly weak, as the initial motivation of this patch
> >> is that you did not want to say "you should run gc only in the
> >> _repo_ not in workdir".
> >
> > Yes. That's my motivation and it works right now
> >
> > 	git init a
> > 	<hack, hack, hack,>
> > 	git commit -a
> >
> > 	git-new-workdir a b 	# allowed
> > 	git-new-workdir a c	# allowed
> >
> > 	git-new-workdir b d	# NOT ALLOWED
> 
> But I do not think you are disallowing it; instead you are
> making the same problem appear without telling the user.
> 
> Also, how is the above different from this?
> 
> 	git init a
>         cd a ; git gc ; cd ..	# allowed
> 	git new-workdir a b
> 	cd b ; git gc ; cd ..	# NOT ALLOWED
> 

Sorry, you lost me here. Your above sequence _is_ allowed and that was
just the point of the patch. I lightly tested it that it does the right
thing, so perhaps I'm missing something?

What isn't allowed is the following:

	mkdir a; cd a; git-init; cd ..
	git new-workdir a b
	cd b; git gc ; cd .. # IS ALLOWED
	git new-workdir b c
	cd b; git gc ; cd .. # NOT ALLOWED

Because now you created a new workdir c which doesn't point to a repo,
but only to another _workdir_ b. And only in this case you get a symlink
chain like this:

c/.git/packed-refs -> b/.git/packed-refs -> a/.git/packed-refs

This is even dissallowed by the code in git-new-workdir (Sorry, I just
saw it now; otherwise I wouldn't spend so much time in arguing this)):

# don't link to a workdir
if test -L "$orig_git/.git/config"
then
        die "\"$orig_git\" is a working directory only, please specify" \
                "a complete repository."
fi

> You are saying "you should run workdir only in the _repo_ not in
> workdir".
> 

This sentence doesn't make any sense to me. Did you mean "you should run
gc only ..." ?

> As I already said, certain things work differently between a
> proper repository and a worktree that borrows .git/refs from a
> proper repository, and you always have to know what you are
> doing when you use such a setup.  If your goal is to minimize
> the difference, I do not think it makes much sense to allow gc
> and not allow new-workdir.
> 

I think you missunderstud me. Hopefully the above explanation clears this
missunderstanding. The case I feared (symlink chain of workdirs) is not
allowed in git-new-workdir from the very begining of this script, so
there shouldn't be any problem with the symlink handling in my patch.

> On the other hand, if we admit that things work differently, I
> think erroring out gc or pack-refs when we see .git/packed-refs
> is a symbolic link is much simpler, less error prone and easier
> to explain.
> 

But with my patch it just works! I really tested it again. The link
in b/.git/packed-refs -> a/.git/packed-refs (using the example from above)
isn't broken up and in the new generated packed-refs are stored inside
the repo a (as they should).

-Peter