Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!uwm.edu!zaphod.mps.ohio-state.edu!usc!ucsd!ucbvax!UNHH.BITNET!D_FELDMA
From: D_FELDMA@UNHH.BITNET
Newsgroups: comp.lang.forth
Subject: A word on my Forth wishlist
Message-ID: <9001221802.AA23796@jade.berkeley.edu>
Date: 22 Jan 90 07:01:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Reply-To: Forth Interest Group International List <FIGI-L%SCFVM.bitnet@jade.berkeley.edu>
Organization: The Internet
Lines: 49

For me, the most tedious aspect of programming in Forth is getting items
on the data stack in the desired order.  It would be nice to have a word, say,
REORGANIZE that would lift this burden.

I can think of two ways that this might work:

RUNTIME version:

a_1 a_2 ... a_n str1 str2 REORGANIZE ====> a_j_1 ... a_j_m

where the a_i are arbitrary data elements, and str1 and str2 are (addresses of)
strings, so that for example:

a_1 a_2 a_3 (abc) (cbbac) REORGANIZE ===> a_3 a_2 a_2 a_1 a_3

This would be easy enough to write, but slow.

COMPILETIME version:  str1 str2 REORGANIZE as a string of Forth words to
accomplish the desired rearrangement.  For example:

(a) (aa)     REORGANIZE   would compile as    dup
(ab) (ba)    REORGANIZE   would compile as    swap
(ab) (aba)   REORGANIZE   would compile as    over
(ba) (baaba) REORGANIZE   might compile as    over over swap over
                                          or  dup dup 4 pick swap
                                          or  ???
             (I hope I have the numbering convention for  pick  correct;
              I've been writing a lot of PostScript lately.)
The point is that to write such a word, one must solve a (possibly
implementation dependent) optimization problem.  The reason for the dependency
on implementation is that on any given system, e.g. pick might be more or
less expensive than  swap.  So problem: given the relative cost of Forth's
stack primitives in a given implementation, generate the cheapest sequence
for rearranging the stack in a given way.  Or easier, just find one way to
implement REORGANIZE that's presumably close to optimal.  If the word generated
different code on different systems, but all with the same runtime behavior,
would it still be considered portable?

Having such a word would make Forth programs more readable (because a series
of stack manipulations is seldom transparent at first glance), and possibly
more efficient (if REORGANIZE could be optimized.)  In particular, it would
allow one to do on the stack what one might do less efficiently with reads
and writes to memory while not compromising the readability of one's code.

Any thoughts on how this might be done?  On on why one would or wouldn't
want to, if it could?

                                David Feldman
                                d_feldman@UNHH