Re: I'm a total push-over..

From: Junio C Hamano <gitster@pobox.com>
Date: Jan 25, 2008, 19:07 UTC
Message-ID: <7v63xh7mgt.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <87abmtvkd8.fsf@jbms.ath.cx>

Jeremy Maitin-Shepard <jbms@cmu.edu> writes:

Show 12 quoted lines

> In theory, I agree that this is possible, but in practice it may not be
> reasonable at all.  Consider two possible comparison functions:
>
> 1. compare file names as strings case-insensitively assuming a latin 1
> encoding
>
> 2. compare file names as strings case-insensitively assuming a UTF-8
> encoding
>
> Actually writing a hash function such that two strings hash to the same
> value if either of these comparison functions says that the strings are
> equal would appear to be rather difficult.

Once you start adding more "case folding" supported filesystems to the repertoire, such a unified hash function Dscho suggests needs to throw paths that other (N-1) "case folding" filesystems treat as distinct but only 1 filesystem treats "equivalent" into the same hash bucket. I would say not just difficult but the resulting function would have too many collisions to make it ineffective.