From: Junio C Hamano <gitster@pobox.com>
Date: Wed, 08 Oct 2014 17:03:21 GMT
Subject: Re: [PATCH 0/16] make prune mtime-checking more careful
Message-ID: <xmqq38aycx46.fsf@gitster.dls.corp.google.com>
In-Reply-To: <20141008083109.GA3767@peff.net>

Jeff King <peff@peff.net> writes:

> So that explains that bug (as a side note, you might think that if we
> are flipping return values, lots of things would fail when they ask "do
> we have this packed object" and it erroneously says "yes". But that does
> not happen. The wrong return value comes from freshening the file, so we
> only flip "yes" to "no", and never the opposite).
> ...
> When dt/cache-tree-repair is merged, we have a valid cache tree when we
> run "git commit", and we realize that we do not need to write out the
> tree object at all. Thus we never hit the buggy code, the object isn't
> created, and the subsequent prune reports that there is nothing to
> prune.

Wow, that is a tricky "bug" (rather the series with the bug failed
to fail when applied to 'master', so it is an "unbug" that hides a
bug) to hunt down.  Thanks for digging.