Re: marking mystery code

James Kanze US/ESC 60/3/141 #40763 <kanze@lts.sel.alcatel.de>
21 Feb 1996 00:06:16 -0500

          From comp.compilers

Related articles
[3 earlier articles]
Re: marking mystery code jgj@ssd.hcsc.com (1996-02-16)
Re: marking mystery code hagerman@ece.cmu.edu (1996-02-16)
Re: marking mystery code mnp@compass-da.com (Mitchell Perilstein) (1996-02-16)
Re: marking mystery code toon@moene.indiv.nluug.nl (Toon Moene) (1996-02-16)
Re: marking mystery code preston@tera.com (1996-02-17)
Re: marking mystery code rfg@monkeys.com (1996-02-19)
Re: marking mystery code kanze@lts.sel.alcatel.de (James Kanze US/ESC 60/3/141 #40763) (1996-02-21)
Re: marking mystery code xanthian@qualcomm.com (1996-02-24)
| List of all articles for this month |

From: James Kanze US/ESC 60/3/141 #40763 <kanze@lts.sel.alcatel.de>
Newsgroups: comp.compilers
Date: 21 Feb 1996 00:06:16 -0500
Organization: Compilers Central
References: 96-02-168 96-02-198
Keywords: design, debug

Mitchell Perilstein <mnp@compass-da.com> writes:


> I will try to defend the FSF here, since no one else has yet.


        [...]
> But let's talk about the effectiveness of the systems. As proof of
> the GNU concept, check out "Fuzz Revisited: a Re-examination of the
> Reliability of UNIX Utilities and Services," by Barton Miller
> <bart@cs.wisc.edu> et al at U of WI CS. (Anyone know where it was
> published? I just have a prelim paper dated 4/95.)


The initial paper was published quite some time ago, in the
Communications of the ACM. I've not seen a publication of the update
you are talking about.


> Anyway, they tested several UNIX implementations for reliability of
> their programs under unexpected conditions. They concluded that GNU
> and Linux utilities (Linux utilities are almost all GNU) had
> noticably better reliability than commercial 1995 UNIXes. The
> percentages of tools which crashed or hung were: SunOS 23, HP-UX 18,
> AIX 20, Solaris 23, Irix 15, Ultrix 21, NEXT 43, GNU 7, Linux 9.
        [...]


I'm familiar with the above survay. While I think it rather bad of
the commercial vendors *not* to have fixed their products, I think I
can see why. The particular problems tested simply don't affect the
average user, so reparing the problem is not cost efficient.


On the other hand, while the GNU utilities do not crash on input they
were not designed for, they do give wrong results; often enough, IMHO,
to make them unusable. I find this particularly frustrating, since
the GNU utilities generally use more recent algorithms, and are at
times several orders of magnitude faster than the standard Unix
versions. (My own experience is with diff. I tried the GNU diff
because of performance. I went back to the standard one when I found
that the GNU diff pretty regularly reported more than the minimum set
of differences, in some cases, more than double.)


--
James Kanze Tel.: (+33) 88 14 49 00 email: kanze@gabi-soft.fr
GABI Software, Sarl., 8 rue des Francs-Bourgeois, F-67000 Strasbourg, France
--


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.