Re: [PATCH 5/5] submodule: submodule_move_head omits old argument in forced case

From: Stefan Beller <sbeller@google.com>
Date: Dec 20, 2017, 19:36 UTC
Message-ID: <CAGZ79kZpY8iXhCUOB=KuwGPOr7zvV7z+Q-L43p28=7ydaTvutg@mail.gmail.com>
In-Reply-To: <20171220000148.GJ240141@aiede.mtv.corp.google.com>

On Tue, Dec 19, 2017 at 4:01 PM, Jonathan Nieder <jrnieder@gmail.com> wrote:

Show 21 quoted lines

> Stefan Beller wrote:
>> On Tue, Dec 19, 2017 at 2:44 PM, Jonathan Nieder <jrnieder@gmail.com> wrote:
>
>>>   checkout -f
>>>         I think I would expect this not to touch a submodule that
>>>         hasn't changed, since that would be consistent with its
>>>         behavior on files that haven't changed.
> [...]
>> I may have a different understanding of git commands than you do,
>> but a plain "checkout -f" with no further arguments is the same as
>> a hard reset IMHO, and could be used interchangeably?
>
> A kind person pointed out privately that you were talking about
> "git checkout -f" with no further arguments, not "git checkout -f
> <commit>".  In that context, the competing meanings I mentioned in
> https://crbug.com/git/5 don't exist: either a given entry in the
> worktree matches the index or it doesn't.
>
> So plain "git checkout -f" is similar to plain "git reset --hard"
> in that it means "make the worktree (and index, in the reset case)
> look just like this".

with "this" == the argument that was given, if the argument was omitted, HEAD is assumed.

>  checkout -f makes the worktree look like the index;

No, here is what my installation of git (recent master) does:

  $ git --version
git version 2.15.1.389.g52015aaf9d

  $ cat test.sh

  rm -rf tmp
  git init tmp

  cd tmp
  git commit --allow-empty -m initial
  echo commit >a
  echo commit >b
  echo commit >c
  git add .
  git commit -m commit

  echo index >a
  git add a
  echo worktree >a

  echo index >b
  git add b

  echo worktree>c

  $ sh test.sh
Initialized empty Git repository in /u/tmp/.git/
[master (root-commit) c109fb5] initial
[master fcc21ea] commit
 3 files changed, 3 insertions(+)
 create mode 100644 a
 create mode 100644 b
 create mode 100644 c
  $ cd tmp
  $ git status
On branch master
Changes to be committed:
  (use "git reset HEAD <file>..." to unstage)

modified: a modified: b

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git checkout -- <file>..." to discard changes in working directory)

modified: a modified: c

  $ git checkout -f
  $ git status
On branch master
nothing to commit, working tree clean

# Let's test the other commands as well:
  $ cd ..
  $ sh test.sh
   ...
  $ cd tmp
  $ git checkout -f HEAD
  $ git status
On branch master
nothing to commit, working tree clean

  # See, there is no difference between giving HEAD as an argument
  # or not! Try again with reset just for completeness:

  $ cd ..
  $ sh test.sh
   ...
  $ cd tmp
  $ git reset --hard
HEAD is now at b71ae70 commit
  $ git status
On branch master
nothing to commit, working tree clean

  # The only difference between reset and the checkout is that reset
  # tells me where we are.

> reset --hard makes the worktree and index look like HEAD.

I agree.

Show 5 quoted lines

> In that context, more aggressively making the submodule in the
> worktree get into the defined state sounds to me like a good change.
>
> Hopefully my confusion helps illustrate what a commit message focusing
> on the end user experience might want to mention.

I'll try to come up with a better commit message. Thanks for bearing with me here.

Stefan

>
> Thanks again,
> Jonathan