LispM Market Share

["William D. Gooch" <ai.gooch@MCC.COM>]

    Date: Sat, 13 Jan 90 17:11 EST
    From: JCMA@REAGAN.AI.MIT.EDU (John C. Mallery)

    If the lisp machine is the best program development environment available
    today, why isn't it more popular?

    Here are some hypotheses that come to mind:

	    3- LispMs are too slow for numerical computations -- ...

My experience says that this is a false assumption.  The raw, untuned
performance of some lisp programs doing math may be poor due to (often
unbeknownst to the programmer) consing of bignums, rationals,... and/or
due to the higher (on conventional hardware) overhead of generic versus
strongly typed arithmetic.  These problems are generally solvable - in
many cases I have found that simple metering quickly points to ways of
tuning away such lossages.  Sometimes, a bit more effort is needed, as
when using system-internal math routines to do double precision without
consing so that one can really benefit from having those FPA chips.  I
have been told that "the new compiler" for Genera, when it is released,
will provide optimizer support for non-consing rational, complex and
double precision floating point arithmetic.

Of course it all depends on how one defines "too slow."