Re: performance

[Rob Browning <rlb@cs.utexas.edu>]

A few other things I should have mentioned in the first message:

* I wasn't trying to perform a completely fair test to guile.  To be
  more fair, I would have needed to test and show the results using
  hobbit (which would, I presume, be guile's rsc equivalent), but I
  don't have hobbit handy, and I'm not familiar with it, so I didn't
  include those results.  I was mostly just trying to contribute
  whatever information I could collect quickly.

* Stalin, for those who haven't heard, is now GPLed.  I had originally
  wanted to package stalin for Debian, but about half way through I
  checked the license more carefully and realized it wasn't free.
  After talking to Jeff Siskind (the author) about it, he said he was
  interested in GPLing it.  Some possibly tricky corporate lawyer
  wrangling and about 9 months later, he sent me mail announcing the
  latest version and announcing that it was under the GPL.  Woohoo.
  (There's now a Debian package (with one minor bug --- you have to
  use -copt -I -copt /usr/include/gc ATM, but I'm going to fix that
  shortly)).

* Stalin being GPLed might (or might not) have interesting
  implications for guile.  I've just asked Jeff how interested he
  might be in making it easy to call stalin object files from
  RScheme/guile, but I haven't heard back.  I don't even know how hard
  this is likely to be, but even if you had to do some ugly print/read
  interface hack, the performance benefit cases where you'd likely
  care about this kind of hardcore optimization would far outweigh the
  interface boundary overheard.

FWIW.

-- 
Rob Browning <rlb@cs.utexas.edu> PGP=E80E0D04F521A094 532B97F5D64E3930