Re: git submodule update - reset instead of checkout?

From: Stefan Beller <sbeller@google.com>
Date: Mar 26, 2018, 23:33 UTC
Message-ID: <CAGZ79kZ16+ZtcXx45nvcOeT0wSM7TEi8iDoxB6An91d-FanZhg@mail.gmail.com>
In-Reply-To: <20180309150719.GA30596@inner.h.apk.li>

On Fri, Mar 9, 2018 at 7:07 AM Andreas Krey <a.krey@gmx.de> wrote:

> Hi everyone,

> I've been reading up on the current state of git submodules
> (our customer's customers want to use them, causing a slight
> groan from our customer).

> The usability thing I wonder about - with git submodule update,
> is it necessary to detach the head to the current (or upstream)
> head, instead of resetting the current branch to that state?

Try "git checkout --recurse-submodules" or "git reset --recurse-submodules" (there is also the submodule.recurse option in case you don't want to type the option all the time)

> Primary interest in the question: Seeing 'detached head' scares
> almost everybody. To brainstorm:

I agree on that. That is what we are trying to work out eventually, too.

One idea is to "reattach the submodule branch if it fits" another idea would be a submodule ref store that is (partially) tied to the superproject, such that the HEAD of the submodule is non-existent for most of the time. https://public-inbox.org/git/cover.1512168087.git.jonathantanmy@google.com/

> - as we can already use 'submodule update --remote' to update
>    to the remote head of a branch, it would be logical to have
>    that branch checked out locally (and unfortunately, potentially
>    have that branch's name conflict with the remote branch setup).

> - when developers more or less accidentally commit on the detached
>    head, all is not lost yet (I remember this being differently),
>    but if the next 'submodule update' just resets again, the commit
>    just made is still dropped, just as in the detached head state.

> - So, we'd need to have 'submodule update' act closer to the pull or
>    rebase counterparts and refuse to just lose commits (or uncommitted
>    changes).

> Having a checked-out branch in the submodule would have the advantage
> that I can 'just' push local commits. At the moment, doing that requires
> a relatively intricate dance, not at all similar to working in the
> base (parent) module.

Show 5 quoted lines

> I'm working on hooks that automatically update the submodules after
> a commit change (merge, pull, checkout) in the parent module, but
> with the additional feature of (a) keeping a branch checked out
> and (b) avoid discarding local changes. Probably means I shouldn't
> invoke 'submodule update' at all, and deal with everyting myself.

> Any thoughs/comments/helpful hints?

Our plan is to deprecate "git submodule" just like "git remote" is not a well known tool any more. (e.g. Instead of git remote update, use git fetch, that learned about updating the remote tracking branches on its own)

> (Addional complexity: egit/jgit is in use as well, and the work model
> we will arrive at probabaly needs to work with the current egit.)

That sounds familiar, we also have JGit/Gerrit in the setup.

Stefan