Re: Removing part of the history

[Jakub Narebski <jnareb@gmail.com>]

Samuel Lucas Vaz de Mello <samuellucas@datacom.ind.br> writes:

> I have a repository with code imported from CVS.
> 
> I first imported from CVS using git 1.5.6 and we started to develop
> in git branches.  I run cvsimport periodically and merged the
> changes from CVS (legacy developers) to our git branches.  After
> some time, I realized that the import was broken, upgraded to git
> 1.6 and reimported the data.  Now the imported data seems to be
> correct and I finished with a history tree like this:
> 
> (our branch1)             X--Y--Z
> merges>                  /    \
> (our branch2)       J--K--L--M--N--O--P
> merges>             /     /     /
> (cvs old) ...A--B--C--D--E     /
> merges>                       /
> (cvs new) ...A'-B'-C'-D'-E'-F'
> 
> 
> There is any way to get rid of the (broken) cvs-old tree?
> 
> I've tried to rebase, but the commits from A' and A are different,
> it tries to reapply all history in CVS.
> 
> I've tried to checkout a point before our changes start (B') and
> then, using a script, cherry-pick all commits up to HEAD that are
> not in cvs-old branch. This approach didn't handled well merges
> between our branches (K-L-M and X-Y).
> 
> The history from the beginning up to the point where our changes
> started (...A--B) is quite large, so it would be nice for us to get
> rid of it.
> 
> Any suggestion on how to handle this?

You can use grafts (see "Repository Layout" documentation) to prune
'cvs old' branches, by adding for each merge commit <c_i> with one
side leading to 'cvs old' branch line to .git/info/grafts
  <c_i> <c_i^1>
where <c_i> is full hexadecimal object name.

Then check in gitk or some other history viewer if the history now
looks like you wanted (looks sane).

Then use "git filter-branch --all" to turn grafts into reality, check
out correctness, then remove refs/original/* and do git-gc.

HTH
-- 
Jakub Narebski
Poland
ShadeHawk on #git