Re: [PATCH v3 10/21] checkout: split part of it to new command 'switch'

[Phillip Wood <phillip.wood123@gmail.com>]

Hi Elijah

On 11/03/2019 22:04, Elijah Newren wrote:
> On Mon, Mar 11, 2019 at 1:51 PM Phillip Wood <phillip.wood123@gmail.com> wrote:
>> On 11/03/2019 17:24, Elijah Newren wrote:
>>> On Mon, Mar 11, 2019 at 4:47 AM Duy Nguyen <pclouds@gmail.com> wrote:
>>>> On Mon, Mar 11, 2019 at 6:16 PM Phillip Wood <phillip.wood123@gmail.com> wrote:
>>>>> On 08/03/2019 09:57, Nguyễn Thái Ngọc Duy wrote:
>>>>>> "git checkout" doing too many things is a source of confusion for many
>>>>>> users (and it even bites old timers sometimes). To remedy that, the
>>>>>> command will be split into two new ones: switch and
>>>>>> something-to-checkout-paths.
>>>>>
>>>>> I think this is a good idea, thanks for working on it. I wonder if it
>>>>> would be a good idea to have the new command refuse to checkout a new
>>>>> branch if there is a cherry-pick/revert/merge/rebase in progress (with
>>>>> an option to override the check) as switching branches in the middle of
>>>>> one of those is likely to be confusing to users (if I do it it is
>>>>> normally because I've forgotten that I've not run 'git whatever
>>>>> --continue').
>>>>
>>>> Interesting. I think this would be a good default if we have an escape
>>>> hatch (which could even come later). I often wander off to some other
>>>> branch and go back. But then half the time I end up forgetting I'm in
>>>> a middle of something and just "git rebase --quit" :P
>>>>
>>>> Of course with git-stash (*) and git-worktree, I guess there's little
>>>> reason to just switch away from a something-in-progress worktree. I'll
>>>> try to implement this in the next round, unless someone objects.
>>>
>>> No objection here; I like this idea.
>>>
> 
> Keeping this hunk since it's now relevant to the comment below...
> 
>>>>>> +-f::
>>>>>> +--force::
>>>>>> +     Proceed even if the index or the working tree differs from
>>>>>> +     HEAD. Both the index and working tree are restored to match
>>>>>> +     the switching target. This is used to throw away local
>>>>>> +     changes.
>>>>>
>>>>> I'd always thought that --force meant "throw away my local changes if
>>>>> they conflict with the new branch" not "throw them away regardless"
>>>>> (which is better as it is deterministic). Maybe we can come up with a
>>>>> clearer name here --discard-changes? At the moment --force does not
>>>>> throw away conflicts properly (see the script below in my comments about
>>>>> --merge).
>>>>
>>>> Yeah if you want to redefine --force, now is a very good time.
>>>> Personally I'd rather have separate options than the "one catch all"
>>>> --force (or worse, multiple of --force). I'll leave this for the
>>>> community to decide.
>>>>
>>>> Adding Elijah too. He also had some concern about "git restore
>>>> --force". Maybe he's interested in this as well.
>>>
>>> I like Phillip's suggestion of --discard-changes.  I also like how he
>>> came up with a simple testcase showing one bug each with checkout's
>>> current -m and -f handling; we should fix those.
>>
>> With regard to discarding conflicts, do we want it to clear up any state
>> associated with the conflicts (like reset)? They rarely happen in
>> isolation, there's a MERGE_HEAD or CHERRY_PICK_HEAD etc. I'm not sure
>> what it should do in the middle of a rebase or when cherry-picking a
>> range of commits. I think it would be surprising if it was the
>> equivalent of rebase/cherry-pick --quit but just clearing the conflicts
>> in those contexts may not be very useful in practice.
> 
> You already suggested above (outside the context of --discard-changes)
> that we should just error out if there is some special mid-operation
> state (be it from a merge, cherry-pick, rebase, or bisect).  The user
> can then manually resolve the operation first, or, perhaps use a
> special override to force the switch command to proceed despite the
> presence of mid-operation state.
> 
> Personally, I'm leaning towards --discard-changes operating within
> that same context; I think that mid-operation special state should
> require a more explicit and operation-specific step to remove (e.g.
> rebase --quit).

I think that makes sense, I was wondering if --discard-changes should 
remove the state of a conflicted merge/single cherry-pick/revert but it 
is probably simpler and easier to understand just to error out. I think 
it should still clean up conflicts from other checkouts and applying 
stashes though as there is no other state in those cases.

Best Wishes

Phillip