[plt-scheme] Performance Targets for MzScheme

From: Neil W. Van Dyke (neil at neilvandyke.org)
Date: Thu May 13 13:18:29 EDT 2004

Brent Fulgham <bfulg at pacbell.net> writes at 09:15 13-May-2004 -0700:
> However, my goal was to identify places in MzScheme where performance
> could be improved.  Currently, it looks like SRFI performance is quite
> bad in some cases.

Some SRFI reference implementation code contains some terribly
inefficient pieces.

(For example, a function that effectively feeds a string through a
string port, converts that back to a string, feeds it through a string
port again, back to a string, string port, string, string port, string.
A compiler that could unravel all of that in reasonable time would be
impressive.  Human intelligence can refactor the code or just rewrite it
from scratch.)

You might run some of your real-world code under DrScheme's profiler,
see which SRFI implementations are especially itchy, and scratch them.

As you're improving on the SRFI reference implementations, keep the
merge problem in mind.  I brutalized the SRFI-19 reference
implementation by "tidying it up a bit" so I could understand it and
work with it.  Others had also made substantial enough changes to the
Schematics version of SRFI-19 that the three-way merge fell to the
bottom of my priority list. :) http://www.neilvandyke.org/srfi19-plt/
Actually, I decided what I really needed was not that hefty SRFI, but
perhaps 100 bytes of MzScheme primitive. :)

Posted on the users mailing list.