Re: Submodules & worktree

From: Philippe Blain <levraiphilippeblain@gmail.com>
Date: Jan 28, 2020, 04:01 UTC
Message-ID: <EE184A05-60BD-4939-93E6-C156A0DB3DF8@gmail.com>
In-Reply-To: <CAKQS2dqx9Nux11ot7a-Wxd2qpcdAAWw-e7Li+X3bw7TxJS=Kyw@mail.gmail.com>

Hi Ole,

Show 9 quoted lines

> Le 1 févr. 2019 à 09:00, Ole Pinto <olepinto@gmail.com> a écrit :
> 
> Adding a worktree from a working copy with submodules doesn't work.
> 
> In the config file I have
>   [submodule]
>       recurse = true
> 
> It that's not present, I don't find the problem.

I also stumbled upon this bug and fixed it in 4782cf2ab6 (worktree: teach "add" to ignore submodule.recurse config, 2019-10-27), which was merged in 05fc6471e3 (Merge branch 'pb/no-recursive-reset-hard-in-worktree-add', 2019-12-01) and made it to Git 2.25. `git worktree add` now works correctly in a project with submodules, and cd-ing to the new worktree directory and issuing `git submodule update` inits and clones the submodule(s) to .git/worktrees/<worktree-name>/modules/<submodule-name>, as it should.

I must warn you however that I discovered another bug that can happen with the submodule.recurse config when using worktrees with projects that have submodules. In short, issuing `git checkout` (or `git reset` or `git read-tree`) in one of the worktrees will leave the .git file in the submodule incorrectly pointing to the repository of the submodule in the main worktree, i.e. .git/modules/<submodule-name> instead of .git/worktrees/<worktree-name>/modules/<submodule-name>. I also fixed this bug [1], my patch series for this is currently in next.

[1] https://github.com/gitgitgadget/git/pull/523

Show 16 quoted lines

> # Problem 2 (submodule status seems not to detect some cases)
> 
> $ pwd
> /tmp/main_repo
> 
> $ git submodule status
> ebc9325f7b19164c9bdfd05263481ded66d7bd7d submod (heads/master)
> 
> $ rm submod/file.txt submod/.git
> 
> $ git submodule status
> ebc9325f7b19164c9bdfd05263481ded66d7bd7d submod
> 
> So, I have a plain, empty directory where the submodule should be. Not
> having any 'first char', I'd expect it to be checked out and up to
> date. Is this expected?

This is indeed not expected, and used to work correctly (the hash should be prefixed with a ‘-‘) in Git 2.12. This bevaviour regressed when `git submodule status` was ported to C in a9f8a37584 (submodule: port submodule subcommand 'status' from shell to C, 2017-10-06). Another user reported this regression to the list recently and a patch series is in the works to fix it [2].

[2] https://lore.kernel.org/git/xmqqv9owbw5o.fsf@gitster-ct.c.googlers.com/

Cheers, Philippe.