threads / discuss / 26833

Re: How to remove a git subtree and its history?

Subject: Re: How to remove a git subtree and its history?

## tl;dr

3 messages between Mar 22, 2011 and Mar 23, 2011.

replies: 2people: 2as markdown or json

Avery Pennarun· Mar 22, 2011, 20:44 UTC · lore

On Tue, Mar 22, 2011 at 6:19 AM, Robert Pollak <robert.pollak@jku.at> wrote:

Show 12 quoted lines

> in the git repository of my project I have successfully replaced a git
> submodule by a subtree, essentially by using
>
> $git subtree add --prefix=mySubtree mySubmodule/master
>
> with your git-subtree.sh .
>
> Before accepting my change, my project partners want to be sure that
> removing this new subtree and its history at some later point in time
> stays possible. How can I achieve this? I have unsuccessfully tried "git
> filter-branch" as described by pgs and fmarc in
> http://stackoverflow.com/questions/955793#955793 .

Well, you can certainly remove anything you want with git filter-branch. Are you concerned about shrinking the size of your repo, or just getting files out of the way in your tree, or removing stuff from the git history?

If you want to avoid cluttering the git history, you should use git subtree with the --squash option. That makes it quite painless.

If you want to just get the files out of your tree someday in the future (for example, to switch back to submodules), then simply 'git rm -r' the directory. Rewriting history is massive overkill for this.

If you want to be able to shrink your repository size later, well, then you need git filter-branch. I've never tried to do this since the history rewrite you'd be doing in this case would completely invalidate all your old history: by removing the subtree in all past revisions, you make all those past revisions unbuildable, which seems like missing the point of version control. Nevertheless, I'm sure you could make it happen if you wanted; you can do pretty much anything you want by filtering git history.

> (Be free to CC git@vger.kernel.org if you reply, if you think that's
> appropriate for archiving the info.)

I don't understand why you didn't cc: the git list yourself then.  But okay :)

Have fun,

Avery

Robert Pollak· Mar 23, 2011, 09:01 UTC · re: Avery Pennarun · lore

Am 22.03.2011 21:44, schrieb Avery Pennarun:

Show 28 quoted lines

> On Tue, Mar 22, 2011 at 6:19 AM, Robert Pollak <robert.pollak@jku.at> wrote:
>> in the git repository of my project I have successfully replaced a git
>> submodule by a subtree, essentially by using
>>
>> $git subtree add --prefix=mySubtree mySubmodule/master
>>
>> with your git-subtree.sh .
>>
>> Before accepting my change, my project partners want to be sure that
>> removing this new subtree and its history at some later point in time
>> stays possible. How can I achieve this? I have unsuccessfully tried "git
>> filter-branch" as described by pgs and fmarc in
>> http://stackoverflow.com/questions/955793#955793 .
> 
> Well, you can certainly remove anything you want with git
> filter-branch.  Are you concerned about shrinking the size of your
> repo, or just getting files out of the way in your tree, or removing
> stuff from the git history?
> 
> If you want to avoid cluttering the git history, you should use git
> subtree with the --squash option.  That makes it quite painless.
> 
> If you want to just get the files out of your tree someday in the
> future (for example, to switch back to submodules), then simply 'git
> rm -r' the directory.  Rewriting history is massive overkill for this.
> 
> If you want to be able to shrink your repository size later, well,
> then you need git filter-branch.

Yes, I am aware of the other options and that's what I want to do. (I never really want to use this, I only need a test run to convince my partners to switch to git-subtree.)

Show 7 quoted lines

> I've never tried to do this since
> the history rewrite you'd be doing in this case would completely
> invalidate all your old history: by removing the subtree in all past
> revisions, you make all those past revisions unbuildable, which seems
> like missing the point of version control.  Nevertheless, I'm sure you
> could make it happen if you wanted; you can do pretty much anything
> you want by filtering git history.

I see. As quite a git newbie I hoped you had done this already and could help me with the necessary git filter-branch command line.

>> (Be free to CC git@vger.kernel.org if you reply, if you think that's
>> appropriate for archiving the info.)
> 
> I don't understand why you didn't cc: the git list yourself then.  But okay :)

I was not sure whether discussion of git-subtree belongs there, since it is not part of git (yet?).

> Have fun,

I certainly do!

Thank you, Robert

Avery Pennarun· Mar 23, 2011, 17:09 UTC · re: Robert Pollak · lore

On Wed, Mar 23, 2011 at 2:01 AM, Robert Pollak <robert.pollak@jku.at> wrote:

Show 11 quoted lines

> Am 22.03.2011 21:44, schrieb Avery Pennarun:
>> I've never tried to do this since
>> the history rewrite you'd be doing in this case would completely
>> invalidate all your old history: by removing the subtree in all past
>> revisions, you make all those past revisions unbuildable, which seems
>> like missing the point of version control.  Nevertheless, I'm sure you
>> could make it happen if you wanted; you can do pretty much anything
>> you want by filtering git history.
>
> I see. As quite a git newbie I hoped you had done this already and could
> help me with the necessary git filter-branch command line.

I don't really have time to decode it all for you, but there are basically two things you need to do here:

1) filter out the subtree merge commits
2) filter out the actual files.

You didn't really say which part of git-filter-branch *didn't* work for you. I'm assuming you were successful at #2, since that part comes straight out of the filter-branch documentation, and you ended up with some leftover git history (ie. log entries from the parent project) that you don't want to see.

The easiest way I know to remove commits from the middle of history (other than rebase, but that's too painful when there are merges) is to use grafts: https://git.wiki.kernel.org/index.php/GraftPoint

Or maybe 'git replace' is easier than grafts nowadays.

After replacing the offending merge commits, run filter-branch one more time to make the changes permanent.

Show 7 quoted lines

>>> (Be free to CC git@vger.kernel.org if you reply, if you think that's
>>> appropriate for archiving the info.)
>>
>> I don't understand why you didn't cc: the git list yourself then.  But okay :)
>
> I was not sure whether discussion of git-subtree belongs there, since it
> is not part of git (yet?).

The git list seems to be used for discussion of git side projects sometimes. As it happens, I'm not subscribed to the list anymore (nothing personal; I just can't handle the volume right now), so it's important to leave me on the cc: list though if you want me to see it.

Have fun,

Avery