Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!gem.mps.ohio-state.edu!ginosko!uunet!algor2.algorists.com!jeffrey
From: jeffrey@algor2.algorists.com (Jeffrey Kegler)
Newsgroups: comp.std.c
Subject: Do non-trivial strictly conforming programs exist?
Summary: I don't think so.
Keywords: ANSI C strictly conforming
Message-ID: <1989Sep8.230612.6629@algor2.algorists.com>
Date: 8 Sep 89 23:06:12 GMT
Reply-To: jeffrey@algor2.UUCP (Jeffrey Kegler)
Organization: Algorists, Inc.
Lines: 52

One of the hopes I had for the standard was the ability to write
certain programs where portability is guaranteed.  I was hoping that
making them strictly conforming would do it.  Reading the standard has
dashed that hope.  I honestly hope the following is wrong and that
someone can point out how to write non-trivial strictly conforming
programs.

Section 2.2.4.1 says that "the implementation shall be able to
translate and execute at least one program that contains at least one
instance of every one of the following limits."  This is loosely
enough worded to allow a conforming implementation to reject almost
any program.

This apparently did not escape the Committee, whose Rationale points
out "a deficient implementation could probably contrive a program that
meets this requirement, yet still succeed in being useless."  More of
a threat than completely useless conforming implementations are a
number of seriously deficient ones, each deficient in a different
area.  Since a strictly conforming program must be portable to them
all, it must embody the lowest common denominator.  This can be low
indeed.

The Rationale expresses the hope that "ingenuity [ in contriving a
useless implementation ] would probably require more work than making
something useful."  In the real world of compiler vendors, examples of
implementations going out of their way to be defective are not hard to
find.

But let's assume the only danger is laziness and stupidity.  Take the
limit that requires one to allow "31 parameters in one macro
definition."  Space is at a premium, I want to allow a lot of macros,
and I am too lazy or stupid to do anything but reserve the maximum
space required for 31 parameters for each macro.  My implementation
can have a "oversize flag" in the macro table, putting the parameters
in a special overflow area if more than one parameter is required.
That overflow table need only have room for one macro's parameters for
the implementation to be conforming.

I will leave it as an exercise to the reader to think of even sicker
implementations which conform.

There is one group of programs I can find that are strictly
conforming.  The standard requires a strictly conforming program "not
produce output depending on any unspecified, undefined, or
implementation-defined behavior, ... ."  Any program which produces no
output may therefore take advantage of implementation-defined
behavior.
-- 

Jeffrey Kegler, Independent UNIX Consultant, Algorists, Inc.
jeffrey@algor2.ALGORISTS.COM or uunet!algor2!jeffrey
1762 Wainwright DR, Reston VA 22090