killing a branch

killing a branch

[Jens Axboe]

Hi,

When I need to delete a branch from my git tree, I'm currently using
this (slooow) approach:

$ rm .git/refs/heads/branch-name
$ git prune

to rid myself of the branch and associated objects. Is there a faster
approach?

(please cc me, not on the list)

-- 
Jens Axboe

Re: killing a branch

[Junio C Hamano]

Jens Axboe <axboe@suse.de> writes:

> $ rm .git/refs/heads/branch-name
> $ git prune
>
> to rid myself of the branch and associated objects. Is there a faster
> approach?

Perhaps the "official" way is

	$ git branch -d branch-name ;# you may need -D
        $ git prune

but that essentially is the same as what you are doing.  Note
that having dangling objects in your repository is not a crime,
and you do not have to religiously do "git prune" every time.

Re: killing a branch

[Jens Axboe]

On Tue, Jan 10 2006, Junio C Hamano wrote:
> Jens Axboe <axboe@suse.de> writes:
> 
> > $ rm .git/refs/heads/branch-name
> > $ git prune
> >
> > to rid myself of the branch and associated objects. Is there a faster
> > approach?
> 
> Perhaps the "official" way is
> 
> 	$ git branch -d branch-name ;# you may need -D
>         $ git prune
> 
> but that essentially is the same as what you are doing.  Note

So no time saved there :-)

> that having dangling objects in your repository is not a crime,
> and you do not have to religiously do "git prune" every time.

I know, it just doesn't feel nice!

-- 
Jens Axboe

Re: killing a branch

[Junio C Hamano]

Jens Axboe <axboe@suse.de> writes:

>> that having dangling objects in your repository is not a crime,
>> and you do not have to religiously do "git prune" every time.
>
> I know, it just doesn't feel nice!

Sorry, but I can think of only three reasons (and a half) why
somebody cannot live with "one git prune at the end of the day
before leaving" (or "week" for that matter) workflow:

 * the filesystem quota is too tight and you cannot afford to
   leave unused loose objects around.  May still be true on
   student accounts, perhaps, but I doubt this is much of an
   issue in the modern world anymore.

 * rsync is used to sync from a repository that dropped a branch
   just now, and you do not want to push the garbage out.  Well,
   if you are still using rsync, I'll tell about it to Linus ;-)
   Pushing via git native protocol over ssh would not send
   unreferenced objects out and will not contaminate the other
   end with the garbage.

 * you do not want to leave after starting prune before it
   finishes.  If it is your hobby to watch the paint dry, I
   cannot help you, but you could run prune under nohup (or
   always work inside "screen", which is what I do).

 + having unused things on the disk just does not _feel_ right.
   Well, maybe.  I can argue with a reason but not with a
   feeling.  On a bright side, leaving recently abandoned
   objects around for a while lets you run git-lost-found to
   recover if you accidentally deleted a still-useful branch.

Re: killing a branch

[Jens Axboe]

On Tue, Jan 10 2006, Junio C Hamano wrote:
> Jens Axboe <axboe@suse.de> writes:
> 
> >> that having dangling objects in your repository is not a crime,
> >> and you do not have to religiously do "git prune" every time.
> >
> > I know, it just doesn't feel nice!
> 
> Sorry, but I can think of only three reasons (and a half) why
> somebody cannot live with "one git prune at the end of the day
> before leaving" (or "week" for that matter) workflow:
> 
>  * the filesystem quota is too tight and you cannot afford to
>    leave unused loose objects around.  May still be true on
>    student accounts, perhaps, but I doubt this is much of an
>    issue in the modern world anymore.

Yeah obviously not true.

>  * rsync is used to sync from a repository that dropped a branch
>    just now, and you do not want to push the garbage out.  Well,
>    if you are still using rsync, I'll tell about it to Linus ;-)
>    Pushing via git native protocol over ssh would not send
>    unreferenced objects out and will not contaminate the other
>    end with the garbage.

Haven't used rsync in a long time, I use git:// and ssh:// exclusively.

>  * you do not want to leave after starting prune before it
>    finishes.  If it is your hobby to watch the paint dry, I
>    cannot help you, but you could run prune under nohup (or
>    always work inside "screen", which is what I do).
> 
>  + having unused things on the disk just does not _feel_ right.
>    Well, maybe.  I can argue with a reason but not with a
>    feeling.  On a bright side, leaving recently abandoned
>    objects around for a while lets you run git-lost-found to
>    recover if you accidentally deleted a still-useful branch.

:-)

Alright, I'll just have to shake the habit of running git prune to rid
myself of that dirty dirty feeling.

-- 
Jens Axboe

Re: killing a branch

[Linus Torvalds]

On Tue, 10 Jan 2006, Jens Axboe wrote:
> 
> Alright, I'll just have to shake the habit of running git prune to rid
> myself of that dirty dirty feeling.

Yeah, I'm slowly shaking it off too. I used to run git-fsck-objects 
religiously just because I worried about bugs. I still do it, but 
especially with the recursive-strategy merging, I get "dangling blob" 
messages every once in a while that are _not_ due to bugs, but just due to 
the temporary merge object.

So I'm learning to ignore them, and prune the tree only occasionally, 
instead of compulsively every time.

		Linus

Re: killing a branch

[Jens Axboe]

On Tue, Jan 10 2006, Linus Torvalds wrote:
> 
> 
> On Tue, 10 Jan 2006, Jens Axboe wrote:
> > 
> > Alright, I'll just have to shake the habit of running git prune to rid
> > myself of that dirty dirty feeling.
> 
> Yeah, I'm slowly shaking it off too. I used to run git-fsck-objects 
> religiously just because I worried about bugs. I still do it, but 
> especially with the recursive-strategy merging, I get "dangling blob" 
> messages every once in a while that are _not_ due to bugs, but just due to 
> the temporary merge object.
> 
> So I'm learning to ignore them, and prune the tree only occasionally, 
> instead of compulsively every time.

Compulsive is the right word, it just itches to run the prune after
killing a branch... I'll try and learn as well.

-- 
Jens Axboe

Re: killing a branch

[linux]

Note that "git prune" could be sped up ENORMOUSLY if git-fsck-cache
could be taught to (optionally) not open, uncompress, hash, and
verify any blob objects.  Just assume that they're okay.

I had a look at the code briefly, but it was a little bit hairier (a
more invasive change) than I felt like dealing with.