Removing some files from history

Removing some files from history

[Gennady Kushnir]

Hello all
I'm not yet subscribed, but I wish I shall get reply anyway

I'm going to make my repository public, but I have found that one of
my files contains some private data that I would not like to share.
Is it possible to remove that file from all commits in my local
repository history before publishing it?
Or it would be easier to start publishing with just my current state
(whith all private data cleaned up)?

Thanks in advance.
Gennady

Re: Removing some files from history

[Jeff King]

On Fri, Oct 17, 2008 at 02:38:54PM +0400, Gennady Kushnir wrote:

> I'm going to make my repository public, but I have found that one of
> my files contains some private data that I would not like to share.
> Is it possible to remove that file from all commits in my local
> repository history before publishing it?

Yes, see "git help filter-branch". The first example given does exactly
what you are asking for.

-Peff

Re: Removing some files from history

[Michael J Gruber]

Gennady Kushnir venit, vidit, dixit 17.10.2008 12:38:
> Hello all
> I'm not yet subscribed, but I wish I shall get reply anyway
> 
> I'm going to make my repository public, but I have found that one of
> my files contains some private data that I would not like to share.
> Is it possible to remove that file from all commits in my local
> repository history before publishing it?
> Or it would be easier to start publishing with just my current state
> (whith all private data cleaned up)?
> 
> Thanks in advance.
> Gennady
Use

git filter-branch --index-filter 'git rm --cached secret' -- --all

or

git filter-branch --tree-filter 'rm -f secret' -- --all

where 'secret' is the name of the file to be removed.

After that, make sure you clean up your repo before publishing:
Clean out the original references (command on 1 line):

git for-each-ref  --format='%(refname)' refs/original |while read ref;
do git update-ref -d $ref;done

Clean out the reflog:

git reflog --expire=0 expire

Remove the old objects and packs:

git prune
git repack -adf

[Makes me feel this should be easier.]

Michael

Re: Removing some files from history

[Gennady Kushnir]

thank you for precise instructions.

however I did not completely understand the part about references and reflog
what are these? git tags?

and another question:
did I understand it right, that I can even make some changes to that
file in history - not just simply delete one?

Gennady

2008/10/17 Michael J Gruber <michaeljgruber+gmane@fastmail.fm>:
> Gennady Kushnir venit, vidit, dixit 17.10.2008 12:38:
>> Hello all
>> I'm not yet subscribed, but I wish I shall get reply anyway
>>
>> I'm going to make my repository public, but I have found that one of
>> my files contains some private data that I would not like to share.
>> Is it possible to remove that file from all commits in my local
>> repository history before publishing it?
>> Or it would be easier to start publishing with just my current state
>> (whith all private data cleaned up)?
>>
>> Thanks in advance.
>> Gennady
> Use
>
> git filter-branch --index-filter 'git rm --cached secret' -- --all
>
> or
>
> git filter-branch --tree-filter 'rm -f secret' -- --all
>
> where 'secret' is the name of the file to be removed.
>
> After that, make sure you clean up your repo before publishing:
> Clean out the original references (command on 1 line):
>
> git for-each-ref  --format='%(refname)' refs/original |while read ref;
> do git update-ref -d $ref;done
>
> Clean out the reflog:
>
> git reflog --expire=0 expire
>
> Remove the old objects and packs:
>
> git prune
> git repack -adf
>
> [Makes me feel this should be easier.]
>
> Michael
>

Re: Removing some files from history

[Jakub Narebski]

Please do not toppost, and quote only fragments you are replying to.

"Gennady Kushnir" <genkush@rujel.net> writes:

> thank you for precise instructions.
> 
> however I did not completely understand the part about references
> and reflog what are these? git tags?

git-filter-branch saves old version under refs/original/* namespace.
You have to remove them after making sure that your rewritten history
is correct.  

Additionally git keeps log where branch (tip/head of branch) was in
your repository; this is kind of generalization of ORIG_HEAD, and
allows to recover from such mistakes as wrong "git reset --hard".
Reflog entries from your original history also would pin commits, so
you have to expire reflog.

And finally the commits you have are there in repository, so you have
to "prune" repository to get rid of them.  Reflog entries and
refs/original/* pin those commits, so you have to remove them prior to
pruning.  Pruning simply removes objects which are unreferenced.

> and another question:
> did I understand it right, that I can even make some changes to that
> file in history - not just simply delete one?

Yes, see man git-filter-branch.

-- 
Jakub Narebski
Poland
ShadeHawk on #git