From: Junio C Hamano <gitster@pobox.com>
Date: Thu, 03 May 2012 19:03:52 GMT
Subject: Re: Index format v5
Message-ID: <7v8vh9f5nr.fsf@alter.siamese.dyndns.org>
In-Reply-To: <87obq5p1t0.fsf@thomas.inf.ethz.ch>

Thomas Rast <trast@student.ethz.ch> writes:

> So what's wrong in this thinking?

We tolerate giving occasional false positive "Yes" to "Has this changed?",
but do not allow false negative "No".  So the value st_{ino,dev,uid,gid}
gives us is strictly "these are not likely to change, but if even a single
bit in them change, we need to suspect that it may have changed."  The
primary field that protect us is mtime, and all the rest are more or less
belt-and-suspender safety.