[Ecls-list] Hopefully some innovation..

Juan Jose Garcia-Ripoll jjgarcia at users.sourceforge.net
Sun Jan 6 15:46:23 UTC 2008

Ok, it has been tested and committed. It seems to work both in the
single thread and multithreaded versions and adds no significant time
overhead to a full run of Paul Dietz's test suite. The tag is (CVS
2008-01-06 16:18)

Right now the hash table limits are hard code: cache begins with 1024
entries and grows up to 4092 (see src/c/main.d and src/c/gfun.d) This
can be probably changed, but it would be nice if somebody spent some
time profiling different choices and what are the typical values for a
CLOS-intensive application.

There is also another issue, which is that the cache does not use
timestamps. When the cache gets full, it is simply emptied and
everything starts from zero. It would be nice if we could come up with
a very cheap algorithm to overwrite entries in the cache when this has
reached the predetermine size. Note however that the cache only grows
when methods are called with different types of objects. Rather
complicated programs would be required to test the performance of such
rewriting algorithm.


Facultad de Fisicas, Universidad Complutense,
Ciudad Universitaria s/n Madrid 28040 (Spain)

More information about the ecl-devel mailing list