Re: [PATCH] checkout: teach --worktree

[Denton Liu <liu.denton@gmail.com>]

On Sun, Jun 14, 2020 at 03:44:39AM -0400, Denton Liu wrote:
> Hi Eric,
> 
> On Sat, Jun 13, 2020 at 10:51:47PM -0400, Eric Sunshine wrote:
> > On Sat, Jun 13, 2020 at 10:25 AM Denton Liu <liu.denton@gmail.com> wrote:
> > > [...]
> > > Teach `git checkout --worktree`, allowing users to checkout files
> > > directly into the worktree without affecting the index.
> > >
> > > Signed-off-by: Denton Liu <liu.denton@gmail.com>
> > > ---
> > > diff --git a/Documentation/git-checkout.txt b/Documentation/git-checkout.txt
> > > @@ -264,6 +266,12 @@ When switching branches with `--merge`, staged changes may be lost.
> > > +-W::
> > > +--worktree::
> > > +       When writing contents, only modify files in the worktree. Do not
> > > +       modify the index. This option is essentially a no-op when used
> > > +       without a `<tree-ish>`.
> > 
> > Why a no-op rather than actually diagnosing that --worktree makes no
> > sense in that case and erroring out?
> 
> I decided on this behaviour because I assumed that an empty
> `git checkout` has `git restore` behaviour but I guess I was mistaken.
> I'll change it to error out.

...Disregard the above. I misread your comments.

I thought about it some more and I think that the real bug is in how I
phrased it in the documentation. I meant that --worktree itself was
essentially no-op, not the whole checkout operation.

I think that it makes sense to allow this behaviour. The documentation
states that we only modify files in the worktree. So if we do
`git checkout --worktree <path>`, we should overwrite the worktree with the
index. This should be exactly the same as running `git checkout <path>`.

However, one additonal behaviour I should implement is running
`git checkout --worktree` should behave like running `git checkout .`.
So this would make --worktree not a no-op without a tree-ish.