Re: how to delete the entire history before a certain commit

From: Jakub Narebski <jnareb@gmail.com>
Date: May 3, 2010, 23:42 UTC
Message-ID: <m3k4rkft7k.fsf@localhost.localdomain>
In-Reply-To: <hrnl7o$nnf$1@dough.gmane.org>

Gelonida <gelonida@gmail.com> writes:

Show 18 quoted lines

> Jakub Narebski wrote:
>> Gelonida <gelonida@gmail.com> writes:
>> 
>>> We have a git repository, whose size we want to reduce drastically due
>>> to frequent clone operations and a slow network connection.
>>  
>> Why frequent *clone* operations, instead of using "git fetch" or
>> equivalent ("git pull" which is fetch+merge, or "git remote update")?
> 
> The clone is part of the deployment process and would IIRC be equivalent
> to a 'svn export'
> Almost certainly one can also improve this, but this should probably
> discussed in another thread.
> 
> The sequence on some remote hosts is.
> - git clone tag dirname
> - rm -rf dirname/.git
> - tar cvfz dirname.tgz dirname

 
Why not simply (after enabling 'upload-archive' service in git-daemon
if you serve via git:// URL, and probably similar in the case of
SSH access management by gitosis or gitolite)

  $ git archive --remote=<repo> <tag>

(where <repo> is <dirname> in your example)?

Show 21 quoted lines

>>> The idea is following:
>>>
>>> * archive the git repository just in case we really have to go back in
>>> history.
>>>
>>>
>>> create a new git repository, which shall only contain last month's activity.
>>>
>>> all changes before should be squashed together.
>>> It would be no problem if the very first commit remains unmodified.
>> 
>> If you want to simply _remove_ history before specified commit,
>> instead of squashing it, the best solution would be to use grafts to
>> cauterize (cut) history, check using [graphical] history viewer that
>> you cut it correctly, and then then use git-filter-branch to make this
>> cut permanent.
> 
> This sounds exactly as what I'd like to do.
> I used "git gui" => "Visualize All Branch History" y to choose a nice
> single cutoff point.
> I just didn't know how to apply the cut.

You can read about grafts in git-filter-branch(1) manpage, in gitrepository-layout(5) git repository layout description, and in gitglossary(7) a git glossary.

In short, each line in .git/info/grafts consist of sha1 id of object, followed by space-separated list of its effective (grafted) parents. So to cut history e.g. after commit a3eb250f996bf5e, you need to put line containing only this SHA-1 in .git/info/grafts file, e.g.:

  $ git rev-parse --verify a3eb250f996bf5e >> .git/info/grafts

> So the command to look for is git-filter-branch, right ?
> I'll read the doc.

As you would see in git-filter-branch(1) documentation, simple

  $ git filter-branch --all

(no filter) would make history described by grafts permanent.

Note that this will be rewriting history, and you would make it (much)
harder on any contributor who based his/her work on commits from
before "rebase".

Show 6 quoted lines

>> 
>> You can later use grafts or refs/replaces/ mechanism to join "current"
>> history with historical repository.
> 
> Probably we wont need this, but this sounds rather interesting and is
> good to know.

Grafts were for example used to fuse (join) current and historical Linux kernel repositories, after Linux kernel moved from BitKeeper to Git.

The 'git-replace' mechanism is meant as modern, transferable and safe replacements for grafts file.

-- 
Jakub Narebski
Poland
ShadeHawk on #git