From: Phillip Wood <phillip.wood123@gmail.com>
Date: Tue, 06 Aug 2019 16:32:07 GMT
Subject: Re: amend warnings with no changes staged
Message-ID: <b12f58f6-663a-bd0f-5854-d7ff0b83848b@gmail.com>
In-Reply-To: <xmqqh86vkl0j.fsf@gitster-ct.c.googlers.com>

On 06/08/2019 04:53, Junio C Hamano wrote:
> Junio C Hamano <gitster@pobox.com> writes:
> 
>> Jonathan Nieder <jrnieder@gmail.com> writes:
>>
>>>>> Some non-judgemental descriptive output like
>>>>>
>>>>> 	$ git commit --amend --no-edit
>>>>> 	No changes.
>>>>> 	$
>>>>>
>>>>> would address this case, without bothering people who are doing it
>>>>> intentionally.  So I think there's room for a simple improvement here.
>>>>
>>>> I do that to refresh the committer timestamp.
>>>
>>> I do, too.  The proposal is, paraphrasing,
>>>
>>> 	$ git commit --amend --no-edit
>>> 	Ah, I see that you want me to refresh the committer timestamp.
>>> 	Done, as requested.
>>> 	$
>>
>> Ah, OK then.  I somehow misread "No changes." as an error message.
> 
> Well, on second thought, I think "fatal: no changes" that exits with
> non-zero, with "--force" as an escape hatch for those who want to
> refresh the committer timestamp, would probably be more in line with
> the expectation Lukas had when this thread was started, and I further
> suspect that it might be a bit more end-user friendly.

I agree that this would be the must user friendly way of doing it. I 
think refreshing the timestamp is probably a niche use of '--amend' (out 
of interest what the the motivation for doing that?) although it does 
seem to popular with the other contributors to this thread. We could 
always have a transition period with an opt-in config variable. I find 
the current behavior quite annoying when I've forgotten to stage 
anything as there is no indication the the commit's content is 
unchanged. I've been using a wrapper script that errors out if there are 
no new changes staged with --amend --no-edit and found it very helpful 
(as is a proper --reword option rather than having to give --amend --only)

Best Wishes

Phillip

> 
> It is a backward incompatible behaviour, but I suspect that if I
> were inventing "commit --amend" today, unlike 8588452c ("git-commit
> --amend: allow empty commit.", 2006-03-04), I probably would design
> it that way.  After all, failing and stopping is always a safer
> option than going ahead with or without a report.
> 
> I am not sure which one between "go ahead anyway but report" and
> "fail by default but allow forcing" I would prefer more.  At least
> not yet.  But I won't rule the latter out at this point.
> 
> Thanks.
>