Xref: utzoo comp.object:653 comp.lang.c++:5994
Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!apple!xanadu!michael
From: michael@xanadu.com (Michael McClary)
Newsgroups: comp.object,comp.lang.c++
Subject: Re: Dumb question
Message-ID: <1990Jan2.112028.19312@xanadu.com>
Date: 2 Jan 90 11:20:28 GMT
References: <61737@aerospace.AERO.ORG> <2122@tukki.jyu.fi>
Reply-To: michael@xanadu.UUCP (Michael McClary)
Organization: Xanadu Operating Company, Palo Alto, CA
Lines: 25

In article <2122@tukki.jyu.fi> sakkinen@jytko.jyu.fi (Markku Sakkinen)
SAKKINEN@FINJYU.bitnet (alternative) writes:
>
>Note 2: An old favourite idea of mine is that classes should offer,
>as a primitive operation, rather one comparison operator (values:
>equal, less, greater, incomparable, equivalent; possibly others
>in special cases) than several relational operators. The latter
>can be derived from the former, and class interfaces would be simpler.

I recommend against this approach.  Such an operator must fully
characterize the relationship between two objects to answer any
question about that relationship.  Then the bulk of this work is
discarded.

This may seem a trivial efficiency quibble, but consider:  There
are data types where some relations can be determined in microseconds
from information in RAM, while others require hours of disk activity
(or a tape mount in your Sri Lanka branch office, which is closed due
to civil unrest...)

When you finally encounter such a data type, will you
 - accept the ten-orders-of-magnitude performance penalty,
 - redesign all your code, or
 - install a special-purpose kludge, thus dropping all your
   existing tools on the floor?