Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!elroy.jpl.nasa.gov!jarthur!ucivax!ucla-cs!twinsun!eggert
From: eggert@ata.twinsun.com (Paul Eggert)
Newsgroups: comp.std.c
Subject: qsort(), bsearch(), and rude comparison functions
Message-ID: <1990Oct6.013443.5364@twinsun.com>
Date: 6 Oct 90 01:34:43 GMT
Sender: news@twinsun.com
Organization: Twin Sun, Inc
Lines: 17
Originator: eggert@ata
Nntp-Posting-Host: ata

Suppose I pass to qsort() a comparison function that isn't a total order.
May qsort() dump core?

For example, NN 6.4.10 passes the function order_subj_date() to qsort().  This
function is not a total order: there are values A,B,C such that A<B, B<C, and
C<A according to this function.  (This bug is fixed in NN 6.4.11.)  SunOS 4.1's
qsort() dumps core sometimes in this situation.  It seems to me that a
conforming qsort() should be allowed to dump core here, but I don't see where
the ANSI C standard permits this.

While we're on the subject, what happens if the comparison function has side
effects, benign or otherwise?  How are the sequence points defined here?  For
example, suppose I qsort() an N-item array using a comparison function that
increments a global counter starting at zero; must the counter be at least N-1
when qsort() returns?

Similar questions apply to bsearch().