[Query] Separate hooks for Git worktrees

Hi,

I have a typical use case, where I am using the same repository for both Android and Linux kernel branches.

Android needs us to keep a special hook "commit-msg" which adds a "Change-Id" to every commit we create.

While this works fine with Android, the behavior doesn't change by simply changing to a upstream kernel branch and eventually by mistake I may end up sending patch with Change-Id to upstream kernel as well. And I want to avoid that.

I am looking at ways to make this configuration work for me by applying the hook only for Android branches.

I tried using the "git worktrees" command to create a separate linked tree for my android branch, but it doesn't have a .git directory but just a file linking to the main repository.

Any idea how I can get around this problem without having separate repositories for kernel and android ?

Thanks in advance.

-- viresh

On Thu, Nov 9, 2017 at 2:58 AM, Viresh Kumar <viresh.kumar@linaro.org> wrote:

The standard hook for Gerrit can be {en,dis}-abled via
  git config gerrit.createChangeId {true, false}

We have no worktree specific config yet, though patches for this were floated on the mailing list.

Though recent versions of git learned to conditionally include config files. (look for includeIf in man git-config), which I think could be used to set the option gerrit.createChangeId depending on the worktree you are in.

The proposed approach above might be hacky but sounds as if it should work?

Thanks, Stefan

On 09-11-17, 11:14, Stefan Beller wrote:

Yeah its hacky for sure, but it solved my problem. Thanks for your help Stefan :)

Stefan Beller <sbeller@google.com> writes:

If you meant "conditional include" by "proposed approach above", I do not see which part you found possibly hacky. It is to allow different set of configurations to apply depending on where you are working at, which I think was invented exactly for something like this.

It certainly is not any hackier than using the same repository to house separately manged projects even if they may be related projects.

Where does the aversion of "having separate repositories" primarily come from? Is it bad due to disk consumption? Is it bad because you cannot do "git diff android-branch kernel-branch"? Something else?

If it is purely disk consumption that is an issue, perhaps the real solution is to make it easier to maintain separate repositories while sharing as much disk space as possible. GC may have to be made a lot more robust in the presense of alternate object stores, for example.

On Thu, Nov 9, 2017 at 9:00 PM, Junio C Hamano <gitster@pobox.com> wrote:

Compared to a per-worktree configuration that you can setup via

    git config --for-worktree=X key value

the setup using conditional includes seems hacky to me. (I just realize that these conditional includes can be set using regular git-config, so maybe it is not as hacky as I thought.)

From a UX perspective, I can imagine a way easier workflow, but the data model seems to make sense.

Well it is the same project with different upstream workflows. For example I would imagine that Viresh wants to cherry-pick from one branch to another, or even send the same patch (just with different commit messages, with or without the ChangeId) to the different upstreams?

Yeah, that is an interesting question! (I suspect workflow related things, diff/cherry-pick)

On 10-11-17, 10:00, Stefan Beller wrote:

Right.