NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/lang/c-part1/2808

Internet Message Format | 1996-08-05 | 5.1 KB

Path: newshost.lanl.gov!tanmoy
From: [e-mail] (Tanmoy Bhattacharya)
Newsgroups: comp.lang.c
Subject: Re: Passing multidimensional arrays to functions in C
Date: 24 Jan 1996 04:00:37 GMT
Organization: Los Alamos National Laboratory
Message-ID: <[e-mail]>
References: <[e-mail]>
NNTP-Posting-Host: qcd.lanl.gov
Mime-Version: 1.0
Content-Type: text
In-reply-to: [e-mail]'s message of Tue, 23 Jan 1996 20:13:31 GMT

--text follows this line--
In article <[e-mail]>
[e-mail] (Jim Vijay) writes:
<snip>
   The C FAQ (Q's 6.16 thru 6.20) explains how to pass 2D arrays to functions
   in C.  None of the methods shown is as simple or general as languages such
   as (gasp!) FORTRAN.

Different languages are suited for different things, and details like
this often result from subtle differences in viewpoints. I use both
Fortran and C, and am sometimes irritated by small inconveniences like
this.

   While realizing that I'm inviting flames, and realizing this topic
   may have been discussed to death in the past (but not in the FAQ),

I remember having participated in a similar discussion once in the
past: and remember having seen it more than that one occassion. It is
a fit topic for discussion, probably even in comp.std.c.

   let me ask:

      What would break (or even be damaged) in C if the language allowed
      one to pass the width of a 2D array (in general, n-1 dimensions of
      an n dim array) to a function?  It should be easy for a compiler to
      emit code that dynamically depends on these n-1 dimensions.
      FORTRAN (IV?) at the time of C's birth could do it!

   Please note that I'm not asking how the C language defines and uses multi
   dimensional array parameters to functions.  I would like to know _why_ the
   current mechanism was chosen - initially (by K&R?) and later by the ANSI
   C committees.

What follows is completely my personal opinion. My knowledge of
history is poor, so take everything with a grain of salt. (I know I
should not violate my own rule of not posting if one does not know:
especially since I know I am getting careless as shown by a lot of my
recent postings; I hope this disclaimer warns you enough.)

Originally C was designed in the spirit that it would be a portable
syntax for writing code close to machine language: sort of an high
level assembly language. No assembly language at that time had the
concept of an array: an array therefore was not a basic type in
C. Assembly languages provided means of allocating blocks of memory
... an array declaration in C originally did precisely that. As a
result array types are somewhat special: there are no array values (or
rvalues), any attempt to use them as an rvalue resulted in a pointer
to the allocated block of memory being used instead. As you could
never pass an array in the first place, how could you hope to specify
the bounds of the array? It may be that this bias towards arrays as
allocation units as opposed to true types contributed to
multidimensional arrays not being treated specially.

Today, the rules are somewhat better. Array rvalues are still
absolutely useless: in most uses arrays still become pointers. But,
today, because of the better type checking rules and non-equivalence
of all pointer types (which were probably never specified equivalent,
but often treated that way), arrays are first class types (or at least
one and half class :-). But allowing array sizes to be variables is
still a problem.

1) Currently, according to the ANSI standard, a object allocated `on
   the stack' has little runtime penalty. The compiler
   always knows the exact amount of space it needs for all its local
   variables etc. (excluding of course the heap allocation through
   malloc etc.) Allowing variable sized arrays as a type would mean
   a possible run time allocation. Allowing variable sized arrays
   _only_ in parameter definition would alleviate this
   problem. (FORTRAN chose this rule upto FORTRAN 77)

2) The existence of that type (even if only in parameter
   declarations), would imply changes in the semantics of
   sizeof. Currently sizeof returns a compile time constant and depeds
   on the type of expression, _no_ part of which is evaluated.

3) Allowing the generality of FORTRAN would also imply that the bounds
   declarations be allowed to be forward declarations. This creates
   slight parsing difficulties, and needs resolution of a few
   ambiguities.

Having said all this, do I believe it would be useful? Certainly! This
is onw extension which I use a lot, even though ultimately it is
nothing but syntactic sugar. But then, what isn't?

Cheers
Tanmoy
--
[e-mail](128.165.23.46) DECNET: BETA::"[e-mail]"(1.218=1242)
Tanmoy Bhattacharya O:T-8(MS B285)LANL,NM87545 H:#9,3000,Trinity Drive,NM87544
Others see <gopher://yaleinfo.yale.edu:7700/00/Internet-People/internet-mail>,
<http://alpha.acast.nova.edu/cgi-bin/inmgq.pl>or<ftp://csd4.csd.uwm.edu/pub/
internetwork-mail-guide>. -- <http://nqcd.lanl.gov/people/tanmoy/tanmoy.html>
fax: [phone]   voice: 1 [phone]    [ Home: 1 [phone] ]