Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!elroy.jpl.nasa.gov!sdd.hp.com!samsung!crackers!jjmhome!smds!sw
From: sw@smds.UUCP (Stephen E. Witham)
Newsgroups: comp.lang.misc
Subject: Re: On whether C has first-class composable functions
Summary: Not an interpreter, but not peaches and cream either.
Message-ID: <311@smds.UUCP>
Date: 30 Jan 91 16:29:05 GMT
References: <3576:Jan2622:55:2991@kramden.acf.nyu.edu> <6800@exodus.Eng.Sun.COM>
Organization: SMDS Inc., Concord, MA
Lines: 45

brnstnd@kramden.acf.nyu.edu (Dan Bernstein) writes:
>In article <21245@yunexus.YorkU.CA> oz@yunexus.yorku.ca (Ozan Yigit) writes:
>> Yes, you were able to write an interpreter that contained said
>> functionality, minus the scoping details.
>> Amazing.
>'Twas hardly amazing. What amazes me is that there are *still* people
>posting to this newsgroup asking questions like ``Ya know, I haven't
>seen any portable first-class composable function implementations that
>don't have arbitrary limits.  Is this because they are impossible, or is
>there some sneaky way?'' ...

At least we have the extremes of the opinion spectrum covered.

The things I've seen (and posted) here are neither "interpreters" nor
"first-class" (nor "obvious," as someone said, and Dan agreed).

"Interpreter" implies a big mechanism that you have to write, that uses a
different internal representation, requires a different kind of thinking to
use, runs slowly, and is hard to interface to routines in the "host" 
language.  The functionoids we've been talking about, on the other hand, 
are not that different from things people regularly do in C.  The syntax of 
using them is, well, close to the syntax of using regular C functions.  
They deal in regular C data types.  They run with compiled code, and not a 
lot of overhead, either.  And there's no big mechanism that has to be 
implemented or added to your program.

"First-class," in my favorite interpretation (i.e., relative to the rest of
the language), also doesn't apply to these functionoids.  In C, all variable-
sized data are second-class compared to the simple, fixed-sized data types.
They can't be put in "automatic" variables or be intermediate values in
expressions, so you have to take care of allocating and freeing their space.  
Also, these functionoids can't be used interchangeably with regular C 
function pointers.  I don't think that's a big practical limitation, but it 
is something function pointers can do that functionoid pointers can't.

Calling this stuff obvious is an insult.  Great, so you guys already knew the
answer.  I remember the time when I didn't, and the time when I first started
using this stuff in a real C program (and it ran faster than the assembly-
language program it replaced).  Objects and lexical closures are standard
stuff in computer science now, but for some of us lowly coders, the fact 
that you can get the same effects relatively painlessly in C is a nifty, 
non-obvious idea.

--Steve Witham
Not-the-fault-of: SMDS, Inc., Concord, MA