Posting-number: Volume 5, Issue 59
Submitted-by: "Brandon S. Allbery" <allbery@ncoast.UUCP>
Archive-name: pwgen

[Dr. Jekyll and Mr. Hyde, anyone?  ;-)  ++bsa]

I've described this program, in its OSI Superboard-II incarnation, in
comp.unix.wizards and news.sysadmin.  Now, here's the code for a UN*X
version.  It seems to work OK on ncoast, except for some oddness in the
"random" selections -- which are probably the result of our not having a
good RNG.  (Could someone mail me one or more of the alternative
generators?  Or maybe even the source to BSD random(), assuming that it
falls outside the subset of BSD code that can be traced to AT&T?  To put it
mildly, our random number generator isn't.)

This is not as complete as the original program, which actually had a list
of characters (the current one uses phonemes) which were most likely to
follow other phonemes.  On the other hand, this version does use phonemes
rather than characters, so its creations are more pronounceable as a rule
than those of the original.  The resulting passwords aren't quite as
"natural" as the original, however.  (That may be for the best; the original
was intended as a simple experiment to see if a rule for "proper English
words" could be defined in terms of rules which related letters.  This was
before I got to college and learned that such things were rather unlikely,
but the original program still did a pretty good job of spewing out some
common English words.  That can't be said for *this* program.)

To compile:	cc -O -o pwgen pwgen.c -lm
(The sqrt() function is used in an attempt to produce a random number which
is "weighted" toward one end of the range, so as to prefer one end of a list
of "spellings" for a phoneme over the other end.  It seems to work, but with
this rotted RNG, I can't be absolutely certain.  A trial distribution seems
to be correct, however.)

What's the intent?  I find that I can remember a "word" better if I can
pronounce it.  This may or may not be true for other people, but *anything*
that encourages the use of random passwords is an improvement on what I
typically see at client sites every day.  So this program spits out
passwords which are virtually guaranteed not to be found in the dictionary,
but are (usually) pronounceable to speakers of English.  (The tables can be
modified for other languages, but they're rather hairy.  Perhaps I'll write
a version which loads "compiled" language descriptions, and let the compiler
deal with the hairy stuff.  Hey, they were even hairier in the BASIC
version!)

Oh, well, shar and enjoy.