Re: How to checkout a revision that contains a deleted submodule?

From: Philippe Blain <levraiphilippeblain@gmail.com>
Date: Sep 21, 2020, 23:14 UTC
Message-ID: <FFBB71FD-8D1F-4E86-9E37-813018AFC690@gmail.com>
In-Reply-To: <4eb688f2-0c17-9b85-e60e-f07485895622@gmail.com>

Hi Luke and Kaartic,

Show 54 quoted lines

> Le 20 sept. 2020 à 14:02, Kaartic Sivaraam <kaartic.sivaraam@gmail.com> a écrit :
> 
> On 20/09/20 3:14 pm, Luke Diamand wrote:
>> On Sat, 19 Sep 2020 at 10:03, Luke Diamand <luke@diamand.org> wrote:
>>> 
>>> Maybe this is a FAQ, but I couldn't figure it out!
>>> 
>>> I have a repo which has a couple of submodules.
>>> 
>>> At some point in the past I deleted one of those submodules:
>>> 
>>>     git rm sub2
>>>     git add -u
>>>     git commit -m 'Deleting sub2'
>>>     git push origin
>>>     ...
>>>     ... more commits and pushes...
>>> 
>>> Now I go and clone the head revision. This gives me a clone which has
>>> nothing present in .git/modules/sub2.
>>>     login on some other machine
>>>     git clone git@my.repo:thing
>>>     cd thing
>>>     ls .git/modules
>>>     <sub2 not present>
>>> 
>>> So when I go and checkout an old revision where sub2 is still around I get:
>>>     git checkout oldrevision
>>>     fatal: not a git repository: sub2/../.git/modules/sub2
>>> 
>>> What am I doing wrong?
>>> What set of commands do I need to use to ensure that this will always
>>> do the right thing?
>>> 
>>> Thanks
>>> Luke
>> 
>> Replying to myself, adding Jens who added the section below.
>> 
>> This is a known bug:
>> 
>> https://git-scm.com/docs/git-rm
>> 
>>> BUGS
>>> ----
>>> Each time a superproject update removes a populated submodule
>>> (e.g. when switching between commits before and after the removal) a
>>> stale submodule checkout will remain in the old location. Removing the
>>> old directory is only safe when it uses a gitfile, as otherwise the
>>> history of the submodule will be deleted too. This step will be
>>> obsolete when recursive submodule update has been implemented.
>> 
> 
> I don't think that part of the documentation applies to your case.

I also don't think this part of the doc applies here.

Show 19 quoted lines

> So,
> I also don't think this is a known bug. As a matter of fact, I couldn't
> reproduce this with the following:
> 
> 
> git init checkout-removed-submodule &&
> cd checkout-removed-submodule/ &&
> echo "Hello, world" >foo &&
> git add foo && git commit -m "Initial commit" &&
> git init ../submodule &&
> cd ../submodule/ &&
> echo "Foo bar" >foobar.txt &&
> git add foobar.txt && git commit -m "Foo bar baz" &&
> cd ../checkout-removed-submodule/ &&
> git submodule add ../submodule/ foobar &&
> git commit -m "Add foobar submodule" &&
> git rm foobar/ &&
> git commit -m "Remove foobar submodule" &&
> git checkout HEAD~ # Checking out the "Add foobar submodule" commit

Yes. At this point "foobar" would be empty because '--recurse-submodules' was not used on 'checkout'. Using `git checkout --recurse-submodules HEAD~` instead would populate it, and it would work correctly because the Git repository of foobar does exist at .git/modules/foobar.

> I also tried with a cloned version of that repository as follows:

here let's make sure we re-checkout 'master' before cloning: git checkout -

Show 7 quoted lines

> git clone /me/checkout-removed-submodule/ cloned-repo &&
> cd cloned-repo &&
> git co HEAD~
> 
> I get:
> 
> HEAD is now at 25270d8 Add foobar submodule

I get the same thing, with or without '--recurse-submodules'.

However, if I you have the 'submodule.active' configuration set to '.', which is the case if you *cloned* with '--recurse-submodules', and you then checkout with '--recurse-submodules', then it fails as Luke describes:

git clone --recurse-submodules  checkout-removed-submodule cloned-repo 
cd cloned-repo &&
git co --recurse-submodules HEAD~
  fatal: not a git repository: ../.git/modules/foobar
  fatal: could not reset submodule index

This bug was reported earlier in May [1], and I suggested a couple ways the experience could be improved.

I might add here that maybe a good idea would be that 'checkout' be taught to try to clone the missing submodules if it does not find their repository at .git/modules.

Cheers,

Philippe.

[1] https://lore.kernel.org/git/20200501005432.h62dnpkx7feb7rto@glandium.org/T/#u