Re: [feature request] Warn about or prevent --amend commits that don't change anything

[Bergi <a.d.bergi@web.de>]

Junio C Hamano schrieb:
> Bergi <a.d.bergi@web.de> writes:
>
>> when nothing is staged in the index then `git commit` warns about this
>> fact with either "nothing to commit, working directory clean" or "no
>> changes added to commit".
>> However, `git commit --amend --no-edit` will happily record a new
>> commit that differs in nothing than its commit date from the original.
>
> What kind of "mistake" are you afraid of?

The particular mistake that happened to me was that I `git push -f 
origin`'ed the newly created commit (which worked as expected, 
overwriting the original commit) and then leaving my PC in the belief 
that I had successfully pushed my changes.

> I can sort of see that "git commit --amend" might want to see two
> summary diffstat output at the end, unlike "git commit" that shows
> what changes were recorded relative to the parent.  In addition to
> that "final result is different in this way from the parent" output,
> you might also want "this is the change you made by amending" and
> knowing the fact that you can notice you didn't add anything by the
> latter being empty _might_ give you an additional peace of mind.

Yes, that would be incredibly helpful as well, but is not what I am 
after here.

However `git commit --amend` already shows me that I still have "changes 
not staged for commit" when editing the commit message, so that would've 
been enough for my use case.
The problem is that I used `git commit --amend --no-edit`, which did 
neither warn me about missing staged changes nor existing unstaged changes.
If nothing is staged and the user does not intend to edit any metadata, 
git should be able to deduce that the user (me) did forget something.

> But is that the kind of mistake you are worried about?  IOW, you
> thought you made and added changes X, Y and Z to the index before
> running your "commit --amend", but you forgot the "add" step and did
> not add anything?

Yes, exactly.

> If so, […] your "you need --allow-empty if you really do not want
> any changes to the tree" would not [help much] either, if you added
> X and Y but forgot to add Z.

A good remark.
Maybe in that case it could warn about unstaged edits in the case of 
--no-edit but still succeed.

> So I am sensing a slight XY problem here.

That might well be, I'm just here to tell my story, it's the designers 
that need to decide at which step I went wrong and which parts could be 
improved.

Kind regards,
  Bergi