From: strick -- henry strickland <strick@versant.com>
Date: Wed, 22 Sep 93 12:06:56 PDT
To: pmetzger@lehman.com
Subject: Re: Public-Key Crypto Toolkit
In-Reply-To: <9309221511.AA29362@snark.lehman.com>
Message-ID: <9309221906.AA03782@versant.com>
MIME-Version: 1.0
Content-Type: text/plain


THUS SPAKE "Perry E. Metzger" <pmetzger@lehman.com>:
# 
# Paul Robichaux says:
# > I'd really like to see the toolkit in plain ol' C. I realize that
# > other languages and systems have great advantages over C, but C is
# > portable beyond belief.

It should be at two layers.  TCL (pronounced "tickle") is merely a
layer on top of C.  

(By the way, GENERIC TCL is Plain Old C, and Portable Beyond Belief.  A
lot more portable than mediocre C.  GENERIC TCL is an #ifdef that shuts
off all commands that use any operating system services (besides
malloc() and free()) -- it's merely a string-based language
interpreter, with designed with friendly conventions for adding new
commands that are wrappers of C API.  Supplement GENERIC TCL with
some stdio.h routine wrappers (fopen, fclose, fgets, fputs), and you'll
be happy.  Compiled instantly on my macintosh, even.)

# I agree. By having a good toolkit in C, we'd have an easy time not
# just prototyping applications but building real and portable ones.

Yes, you want a complete C api.    I don't argue against that.

First you assemble your C API at a complete layer.  It'll look a lot
like the RSAREF API.  In fact, the RSAREF portion of it *will be* the
RSAREF API.

You see, RSAREF already is a crypto toolkit.  Not as full as you may
like, but enough to do most basic public-key-cryptosystem stuff.  The
reference implementations of MD2 and MD5 are also part of the C toolkit
API.   And the bignumber packages.  We've already got *lots* of these.

But have you tried using RSAREF API to do anything?  C API are a
notorious pain in the butt to use -- allocating & deallocating memory,
twiddling bits, writing functions to copy one data format to another,
etc.  They're the reason we're not writing code!  When you elevate to a
"scrypting" language, it becomes really easy (even fun) to experiment
and hack stuff together.

# I'd suggest that a streams mechanism, so you could attach various
# processes to a bunch of data in sequence the way unix filters work,
# would also be nice. That way sequences like
# 
# compress -- des -- tran -- des -- idea -- radix-64ify
# 
# could be really easily built.

Yes, but now you just suggested Shell and Streams rather than C as the
easy-to-use interface into this kit.

You've rejected then rediscovered the reason for a "scrypting"
language.  I think you don't yet understand how TCL is used.  You'll be
much more productive in TCL than hacking weird data flows and futzing
with temporary files and klutzy syntax and counting your nested escape
and quote characters in Shell.  (Most shell things are not one simple
pipe from front to end.)

There's still a time and a place for shell stuff -- fortunately there
(finally) is a standard "tclsh" unix (or POSIX) main.c that you can use to
mix your TCL scripts with Unix Shell scripts.  (You could do it before,
but it was more of a "test" program or homegrown driver than part of a
standard main.c)



ANYWAY -- I should quit ranting and Just Start Writing Code.

Unfortunately I'll be out of town till early October....
Perhaps I can suggest some specs before then.

                                                             _
                                               menya zavoot cmpuk
                                               strick@versant.com



TCL is available via ftp from sprite.berkeley.edu and is freely distributable.
UNIX is a registered trademark of whomever bought it last.