Re: [PATCH] Use a temporary index for interactive git-commit

From: Conrad Irwin <conrad.irwin@gmail.com>
Date: Dec 30, 2010, 12:41 UTC
Message-ID: <AANLkTina562i6KzYR4AQbTNExJbCBA3bKHwA_7W45qAG@mail.gmail.com>
In-Reply-To: <20101230043355.GA24555@sigill.intra.peff.net>

On 30 December 2010 04:33, Jeff King <peff@peff.net> wrote:

Show 5 quoted lines

>
> this behavior will not apply to "git add -p". So doesn't that introduce
> a new confusing inconsistency, that ^C from "git commit -p" abandons
> changes entirely, but from "git add -p" will silently stage changes?
>

That is an interesting observation. The intention of the commit was not to make the interactive add atomic, just isolated from the real index (much like git commit /path/to/file.c or git commit -a). This means that if you leave the commit message empty (even after a successful git-add--interactive) you have not staged any new files.

I'd agree that it would also make sense for git-add--interactive to be atomic, so that when you abort it forgets everything you've told it to do (and that this patch highlights that very nicely). At the moment git-add--interactive is atomic at a file level, i.e. it remembers each decision for git add --interactive, and for git add --patch, it saves your decisions only once you've got to the end of a file (not that you'd notice).

While I could fix git add -p in the same manner as git commit -p, by using a temporary isolated index, it would be better to change git-add--interactive directly, and thus git checkout -p and git reset -p too.

Conrad