Parent-less branches

Parent-less branches

[Pedro Melo]

Hi,

inside an existing repository I want to create a new branch without a  
parent commit. Is there a set of commands to do this? Usually only  
the initial commit of a

The idea is to use that branch to track an external source tarball.  
 From time to time, I would checkout the branch and dump a new  
tarball with fresh source code, commit it to the branch and merge  
again with my master.

I could create the setup this setup if I create a separate repo for  
each external source tarball and then fetch them into my main  
repository. But I don't need those repos after that so I'm just  
wondering if there is a way to create the empty branch directly.

Thanks,
-- 
Pedro Melo
Blog: http://www.simplicidade.org/notes/
XMPP ID: melo@simplicidade.org
Use XMPP!

Re: Parent-less branches

[Junio C Hamano]

Pedro Melo <melo@simplicidade.org> writes:

> inside an existing repository I want to create a new branch without a
> parent commit. Is there a set of commands to do this?

Typically these commits come about because you created a completely
separate history elsewhere in a separate repository and fetched and/or
pulled from there.

So "a set of commands to do this" would be:

	go away
	git init
        build history as usual
	come back
        git fetch ..from that other repository..

Re: Parent-less branches

[Wincent Colaiuta]

El 2/5/2008, a las 23:47, Junio C Hamano escribió:
> Pedro Melo <melo@simplicidade.org> writes:
>
>> inside an existing repository I want to create a new branch without a
>> parent commit. Is there a set of commands to do this?
>
> Typically these commits come about because you created a completely
> separate history elsewhere in a separate repository and fetched and/or
> pulled from there.
>
> So "a set of commands to do this" would be:
>
> 	go away
> 	git init
>        build history as usual
> 	come back
>        git fetch ..from that other repository..

I wanted to do this just the other day, but I did it all inside the  
one repo, and made some notes on the process here:

http://rails.wincent.com/wiki/Creating_independent_branches_with_Git

Basically, using "git write-tree" and "git commit-tree" plumbing.

Cheers,
Wincent

Re: Parent-less branches

[Pedro Melo]

Hi,

On May 2, 2008, at 10:47 PM, Junio C Hamano wrote:

> Pedro Melo <melo@simplicidade.org> writes:
>
>> inside an existing repository I want to create a new branch without a
>> parent commit. Is there a set of commands to do this?
>
> Typically these commits come about because you created a completely
> separate history elsewhere in a separate repository and fetched and/or
> pulled from there.
>
> So "a set of commands to do this" would be:
>
> 	go away
> 	git init
>         build history as usual
> 	come back
>         git fetch ..from that other repository..

Yes, that is what I was using until today.

I was asking for a in-repo solution, without the second git init.

The following email by Wincent Colaiuta does exactly what I wanted :).

Thanks!
-- 
Pedro Melo
Blog: http://www.simplicidade.org/notes/
XMPP ID: melo@simplicidade.org
Use XMPP!

Re: Parent-less branches

[Pedro Melo]

Hi,

On May 2, 2008, at 11:32 PM, Wincent Colaiuta wrote:

> El 2/5/2008, a las 23:47, Junio C Hamano escribió:
>> Pedro Melo <melo@simplicidade.org> writes:
>>
>>> inside an existing repository I want to create a new branch  
>>> without a
>>> parent commit. Is there a set of commands to do this?
[...]
> I wanted to do this just the other day, but I did it all inside the  
> one repo, and made some notes on the process here:
>
> http://rails.wincent.com/wiki/Creating_independent_branches_with_Git

This is exactly what I needed, thanks!

Best regards,
-- 
Pedro Melo
Blog: http://www.simplicidade.org/notes/
XMPP ID: melo@simplicidade.org
Use XMPP!

Re: Parent-less branches

[Ulrik Sverdrup]

> I wanted to do this just the other day, but I did it all inside the  
> one repo, and made some notes on the process here:
> 
> http://rails.wincent.com/wiki/Creating_independent_branches_with_Git
> 
> Basically, using "git write-tree" and "git commit-tree" plumbing.
> 
> Cheers,
> Wincent

I think I've seen this done much simpler, in this blog post:

http://madduck.net/blog/2007.07.11:creating-a-git-branch-without-ancestry/

Even though your instructions seem more complicated than they are since
they mix setting up the parentless branch and preparing the first
commit content.

regards,
ulrik

(I am not on the list, please CC)

Re: Parent-less branches

[Randal L. Schwartz]

>>>>> "Pedro" == Pedro Melo <melo@simplicidade.org> writes:

Pedro> I was asking for a in-repo solution, without the second git init.

Why?  Are they charging you extra to type git init now? :)

-- 
Randal L. Schwartz - Stonehenge Consulting Services, Inc. - +1 503 777 0095
<merlyn@stonehenge.com> <URL:http://www.stonehenge.com/merlyn/>
Perl/Unix/security consulting, Technical writing, Comedy, etc. etc.
See PerlTraining.Stonehenge.com for onsite and open-enrollment Perl training!

Re: Parent-less branches

[Jakub Narebski]

Ulrik Sverdrup <ulrik.sverdrup@gmail.com> writes:
> Wincent Colaiuta <win@wincent.com> writes:
>>
>> I wanted to do this just the other day, but I did it all inside the  
>> one repo, and made some notes on the process here:
>> 
>> http://rails.wincent.com/wiki/Creating_independent_branches_with_Git
>> 
>> Basically, using "git write-tree" and "git commit-tree" plumbing.
> 
> I think I've seen this done much simpler, in this blog post:
> 
> http://madduck.net/blog/2007.07.11:creating-a-git-branch-without-ancestry/

And it is mentioned in http://git.or.cz/gitwiki/BlogPosts, athough not
yet in either GitTips or GitFaq.


In short the solution is to create content-less branch using:

  $ git symbolic-ref HEAD refs/heads/newbranch
  $ rm .git/index
 
then commit.

> Even though your instructions seem more complicated than they are since
> they mix setting up the parentless branch and preparing the first
> commit content.

-- 
Jakub Narebski
Poland
ShadeHawk on #git

Re: Parent-less branches

[Sverre Rabbelier]

On Sat, May 3, 2008 at 2:17 AM, Jakub Narebski <jnareb@gmail.com> wrote:
>  In short the solution is to create content-less branch using:
>
>   $ git symbolic-ref HEAD refs/heads/newbranch
>   $ rm .git/index
>
>  then commit.

I guess what Pedro is asking is if there isn't a builtin way to do
this. Which, if I understand correctly, there isn't. Perhaps though
something like "git branch -N newbranchname" which then creates a new
parentless branch. It could just be a shell script that does the
above, but I wonder if there are no side-effects of the above? It
sounds somewhat 'messy' to remove the index? Too 'low level' to do for
a user I mean, what if sometime in the future we decide to store
something in the index that -is- important? I reckon it would be
cleaner to add an option for this.

-- 
Cheers,

Sverre Rabbelier

Re: Parent-less branches

[Johannes Schindelin]

Hi,

On Fri, 2 May 2008, Junio C Hamano wrote:

> Pedro Melo <melo@simplicidade.org> writes:
> 
> > inside an existing repository I want to create a new branch without a 
> > parent commit. Is there a set of commands to do this?
> 
> Typically these commits come about because you created a completely 
> separate history elsewhere in a separate repository and fetched and/or 
> pulled from there.

It is funny.  I really like that approach, but too many people seem to be 
married to the concept that you have to stay in the same working 
directory, and rather play dirty tricks with the plumbing.

Ciao,
Dscho

Re: Parent-less branches

[Daniel Barkalow]

On Sat, 3 May 2008, Sverre Rabbelier wrote:

> On Sat, May 3, 2008 at 2:17 AM, Jakub Narebski <jnareb@gmail.com> wrote:
> >  In short the solution is to create content-less branch using:
> >
> >   $ git symbolic-ref HEAD refs/heads/newbranch
> >   $ rm .git/index
> >
> >  then commit.
> 
> I guess what Pedro is asking is if there isn't a builtin way to do
> this. Which, if I understand correctly, there isn't. Perhaps though
> something like "git branch -N newbranchname" which then creates a new
> parentless branch. It could just be a shell script that does the
> above, but I wonder if there are no side-effects of the above? It
> sounds somewhat 'messy' to remove the index? Too 'low level' to do for
> a user I mean, what if sometime in the future we decide to store
> something in the index that -is- important? I reckon it would be
> cleaner to add an option for this.

If sometime in the future we decide to store something in the index that 
is important, we'll definitely want to remove the index, because whatever 
it is will be both important and -wrong- for a new parentless commit, 
since a new root should have all the default values, not values acquired 
during the other history in the repo.

On the other hand, a builtin command of that sort probably ought to make 
sure that there aren't any uncommitted changes before doing this. And it 
should probably be called "git checkout -b name -N", since it makes it 
your current branch, rather than creating it like "git branch ..." would 
suggest.

	-Daniel
*This .sig left intentionally blank*