Re: Issue when changing staged files in a pre-commit hook

[Junio C Hamano <gitster@pobox.com>]

Duy Nguyen <pclouds@gmail.com> writes:

> On Wed, Jan 20, 2016 at 1:44 AM, Junio C Hamano <gitster@pobox.com> wrote:
>> Duy Nguyen <pclouds@gmail.com> writes:
>>
>>> I think it's the intended behavior.
>>
>> Yeah, pre-commit was designed for inspecting and rejecting, not for
>> tweaking and munging.  Perhaps "git commit" can be tightened to make
>> sure that pre-commit that returns successfully did not muck with the
>> working tree and the index?
>
> That was my impression from the docs, but then I saw this comment,
>
> /*
> * Re-read the index as pre-commit hook could have updated it,
> * and write it out as a tree.  We must do this before we invoke
> * the editor and after we invoke run_status above.
> */
>
> which comes from 2888605 (builtin-commit: fix partial-commit support -
> 2007-11-18) that admits "the hook can modify it (the index)". And I
> was about to update the docs, but the other way around, about updating
> index and side effects.

I do not think the comment is wrong per-se, but the code we have,
either back then or even today, is insufficient to allow pre-commit
hook that mucks with the fake index that is shown to it.

Re-reading the in-core index at that point may help creating a
commit whose tree matches what the hook did, but the extra change
made by the hook is not ported forward in the real index that the
user will use after a partial commit (and there is no easy way to do
so cleanly--the change the hook makes may even overlap the change in
the real index that are added but left uncommitted, and you would
end up needing to run a threeway merge).

The only sensible thing we can do at that point in the code after
re-reading the index is to make sure that hasn't been changed by the
pre-commit hook and complain loudly to die if we find it modified, I
think.