[Ecls-list] Hopefully some innovation..
ehuels at gmail.com
Mon Jan 7 13:57:42 UTC 2008
On 1/6/08, Juan Jose Garcia-Ripoll <jjgarcia at users.sourceforge.net> wrote:
> I am working on a major fix in ECL which, however, does not involve a
> lot of coding. I am currently conducting the usual tests and the code
> HAS NOT YET been committed to CVS. If after reading this email
> somebody sees a major objection on this change or has an idea for
> other kind of testing, please speak up :-)
> The idea is very simple. When generic functions are called, the
> appropiate method has to be selected based on the argument types. This
> computation is costly and it is cached on a hash table. Traditionally,
> this table is a per-function one. This is how PCL was coded and it is
> also how ECL, CMUCL, SBCL and probably others evolved.
SBCL has recently grown hashes which allow thread-safe reading and
writing without requiring locks. This is said to have improved
hash-access in a threaded context considerably. Do you use
atomic/non-locked hash read/update algorithms? If not, that would have
been a possibility for optimization in the old situation. Maybe it
still is for other hash-uses?
More information about the ecl-devel