Re: [PATCH 3/4] git-commit: Refactor creation of log message.

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 5, 2008, 01:39 UTC
Message-ID: <7vzluggp0q.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <47A747EB.9020909@gnu.org>

Paolo Bonzini <bonzini@gnu.org> writes:

Show 17 quoted lines

> Johannes Schindelin wrote:
>>
>> On Mon, 21 Jan 2008, Paolo Bonzini wrote:
>>
>>> This means that: 1) the commit may be aborted after editing the message
>>> if there is a problem writing out the tree object (slight disadvantage);
>>
>> I consider this more than a slight disadvantage.  I regularly take
>> ages coming up with a good commit message, because I think that the
>> overall time balance is better with me spending more time on the
>> message, but every reader spending less time to guess what I meant.
>>
>> So I would be quite annoyed to edit a message, only to find out that
>> for whatever reason the commit was not successful.
>
> Just to make it clearer, the piece of code that would have to fail,
> for the behavior to change, is this:

I suspect Dscho was worried about the case where he says "git commit", types message and then write-tree finds out that the index is still unmerged and the tree cannot be written out.

And I'd be majorly annoyed if the "slight disadvange" was about that.

Show 9 quoted lines

>         discard_cache();
>         read_cache_from(index_file);
>         if (!active_cache_tree)
>                 active_cache_tree = cache_tree();
>         if (cache_tree_update(active_cache_tree,
>                               active_cache, active_nr, 0, 0) < 0) {
>                 rollback_index_files();
>                 die("Error building trees");
>         }

I think this _could_ error out if your index is unmerged.

However, if you have other code to error out early upon unmeregd index before you collect the message from the editor, I think you are Ok.