threads / discuss / 24940

getting git to ignore modifications to specific files

Subject: getting git to ignore modifications to specific files

## tl;dr

9 messages between Sep 1, 2010 and Sep 2, 2010.

replies: 8people: 7as markdown or json

Chris Packham· Sep 1, 2010, 18:57 UTC · lore

Hi,

We have a git repository that as some GNU build system (a.k.a. autotools) files, my current problem is with the INSTALL file but I suspect there may be some others. These can get modified if you are running a different version of autotools from when the files were created.

I've had various arguments about which autotools files should or shouldn't be included in our repositories. My general rule of thumb is that if it is automatically generated then it shouldn't go into the repository.

There are a couple of repositories that are local clones of 3rd party repositories which have included the pesky auto-generated files so whle I can remove the offending files from repositories we control I need another solution for the 3rd part ones.

I did a bit of googling and found
  git update-index --assume-unchanged

Which works locally to stop git status from complaining. Is there anyway for me to make a change to our clone (a .gitattribues entry?) so that everyone can get this by default?

Thanks, Chris

Joshua Jensen· Sep 1, 2010, 19:06 UTC · re: Chris Packham · lore

Re: getting git to ignore modifications to specific files

  ----- Original Message -----
From: Chris Packham
Date: 9/1/2010 12:57 PM

Show 6 quoted lines

> I did a bit of googling and found
>    git update-index --assume-unchanged
>
> Which works locally to stop git status from complaining. Is there anyway
> for me to make a change to our clone (a .gitattribues entry?) so that
> everyone can get this by default?

I don't think so (but I hope I'm wrong!).

Good idea, though.  +1 from me for being necessary.

Josh

Bruce Korb· Sep 1, 2010, 19:12 UTC · re: Chris Packham · lore

Re: getting git to ignore modifications to specific files

On Wed, Sep 1, 2010 at 11:57 AM, Chris Packham <judge.packham@gmail.com> wrote:

Show 6 quoted lines

> I did a bit of googling and found
>  git update-index --assume-unchanged
>
> Which works locally to stop git status from complaining. Is there anyway
> for me to make a change to our clone (a .gitattribues entry?) so that
> everyone can get this by default?

Isn't that what .git/info/exclude is about?

Chris Packham· Sep 1, 2010, 20:45 UTC · re: Bruce Korb · lore

Re: getting git to ignore modifications to specific files

On 01/09/10 12:12, Bruce Korb wrote:

Show 9 quoted lines

> On Wed, Sep 1, 2010 at 11:57 AM, Chris Packham <judge.packham@gmail.com> wrote:
>> I did a bit of googling and found
>>  git update-index --assume-unchanged
>>
>> Which works locally to stop git status from complaining. Is there anyway
>> for me to make a change to our clone (a .gitattribues entry?) so that
>> everyone can get this by default?
> 
> Isn't that what .git/info/exclude is about?

Not as far as I've been able to tell. .gitignore and .git/info/exclude are ways to tell git to ignore these untracked files so that they don't show up in the output of git status (and others).

In my case the file is tracked, I just don't care if it changes and I don't want git rebase/checkout to complain if I have local modifications to this file. Ideally I'd like git add INSTALL to treat it as if nothing changed but have git add -f INSTALL stage the changes.

Thinking about what I can do with the existing .gitattributes I could probably have a custom diff driver that always returns 0 (/bin/true perhaps) and a custom merge strategy of theirs (or something sensible) if needed. A clean/smudge filter of git update-index --assume-(un)changed may also be possible but I haven't tried it yet.

Joshua Juran· Sep 1, 2010, 19:24 UTC · re: Chris Packham · lore

Re: getting git to ignore modifications to specific files

On Sep 1, 2010, at 11:57 AM, Chris Packham wrote:

Show 16 quoted lines

> We have a git repository that as some GNU build system (a.k.a.
> autotools) files, my current problem is with the INSTALL file but I
> suspect there may be some others. These can get modified if you are
> running a different version of autotools from when the files were  
> created.
>
> I've had various arguments about which autotools files should or
> shouldn't be included in our repositories. My general rule of thumb is
> that if it is automatically generated then it shouldn't go into the
> repository.
>
> There are a couple of repositories that are local clones of 3rd party
> repositories which have included the pesky auto-generated files so  
> whle
> I can remove the offending files from repositories we control I need
> another solution for the 3rd part ones.

I ran into this at a previous job, except using Perforce, so the files were readonly and make failed. My workaround was to comment out the Makefile rules for rebuilding the autotools files.

The basic issues here are self-modifying code and mixing code and data. The first is why I don't use autotools; the second is why I don't use make.

Josh

Jonathan Nieder· Sep 2, 2010, 04:09 UTC · re: Chris Packham · lore

Re: getting git to ignore modifications to specific files

Chris Packham wrote:

> There are a couple of repositories that are local clones of 3rd party
> repositories which have included the pesky auto-generated files so whle
> I can remove the offending files from repositories we control I need
> another solution for the 3rd part ones.

FWIW one half-solution that has worked well for me is to make a local clone without the autogenerated files and complain to upstream so the local clone becomes superfluous after a while.

Naturally another solutions might work better, though.

Jan Krüger· Sep 2, 2010, 09:25 UTC · re: Chris Packham · lore

Re: getting git to ignore modifications to specific files

Chris Packham <judge.packham@gmail.com> wrote:

> I did a bit of googling and found
>   git update-index --assume-unchanged
> 
> Which works locally to stop git status from complaining.

Be careful with that, as it's technically meant to be used for another purpose (e.g. slow filesystems). Notably, git reset --hard will still kill your local modifications. In your case it seems to be about automatically generated files so it's no big problem, but it might still cause a bit of confusion, so I thought I'd warn you anyway.

-Jan

Michael J Gruber· Sep 2, 2010, 10:20 UTC · re: Chris Packham · lore

Re: getting git to ignore modifications to specific files

Chris Packham venit, vidit, dixit 01.09.2010 20:57:

Show 23 quoted lines

> Hi,
> 
> We have a git repository that as some GNU build system (a.k.a.
> autotools) files, my current problem is with the INSTALL file but I
> suspect there may be some others. These can get modified if you are
> running a different version of autotools from when the files were created.
> 
> I've had various arguments about which autotools files should or
> shouldn't be included in our repositories. My general rule of thumb is
> that if it is automatically generated then it shouldn't go into the
> repository.
> 
> There are a couple of repositories that are local clones of 3rd party
> repositories which have included the pesky auto-generated files so whle
> I can remove the offending files from repositories we control I need
> another solution for the 3rd part ones.
> 
> I did a bit of googling and found
>   git update-index --assume-unchanged
> 
> Which works locally to stop git status from complaining. Is there anyway
> for me to make a change to our clone (a .gitattribues entry?) so that
> everyone can get this by default?

For tracking my git configuration including .gitk, I used

[filter "dotgitk"]
        clean = fgrep -v geometry

together with an appropriate attribute for gitk.

I don't know how your autogenerated files differ, but maybe you can set up a filter (or custom diff driver) which makes git think the files are equal though they are not, just like in my case (ignoring window geometry changes for gitk).

Michael

Chris Packham· Sep 2, 2010, 19:40 UTC · re: Michael J Gruber · lore

Re: getting git to ignore modifications to specific files

On 02/09/10 03:20, Michael J Gruber wrote:

Show 38 quoted lines

> Chris Packham venit, vidit, dixit 01.09.2010 20:57:
>> Hi,
>>
>> We have a git repository that as some GNU build system (a.k.a.
>> autotools) files, my current problem is with the INSTALL file but I
>> suspect there may be some others. These can get modified if you are
>> running a different version of autotools from when the files were created.
>>
>> I've had various arguments about which autotools files should or
>> shouldn't be included in our repositories. My general rule of thumb is
>> that if it is automatically generated then it shouldn't go into the
>> repository.
>>
>> There are a couple of repositories that are local clones of 3rd party
>> repositories which have included the pesky auto-generated files so whle
>> I can remove the offending files from repositories we control I need
>> another solution for the 3rd part ones.
>>
>> I did a bit of googling and found
>>   git update-index --assume-unchanged
>>
>> Which works locally to stop git status from complaining. Is there anyway
>> for me to make a change to our clone (a .gitattribues entry?) so that
>> everyone can get this by default?
> 
> For tracking my git configuration including .gitk, I used
> 
> [filter "dotgitk"]
>         clean = fgrep -v geometry
> 
> together with an appropriate attribute for gitk.
> 
> I don't know how your autogenerated files differ, but maybe you can set
> up a filter (or custom diff driver) which makes git think the files are
> equal though they are not, just like in my case (ignoring window
> geometry changes for gitk).
> 
> Michael

In my case the file differs quite a lot (GNU copyright statement has been updated) so a simple clean filter like that won't work.

One question is if I could come up with a filter, would it even work to stop git status from saying its modified?

I tried the custom diff

  [diff "ignoretracked"]
          command = true

but that didn't stop git status from saying it was modified. It did however make "git diff" show no differences as expected.

The best advice seems to be get the file removed. I've sent a patch upstream so hopefully that will fix the problem going forward.