🚀 go-pugleaf

Some questions and remarks on LEDA3.3b:
In my application (computation of colimits of diagrams of algebraic 
specifications and attributed graphs) I tried to find out which is
the best representation for morphisms. I tried all map- and dictionary-types
and noticed the following facts:

- the "print_statistics"-results doesn't coincide with the real memory 
  consumption, because for hash-maps the size of the table isn't taken into
  account (I use a huge amount of small maps).
- the initial size of (hash-)map tables is at least 512 (I would like to know
  why the constructor argument is overwritten).
- I patched the minimum table size for map from 512 to 8 and got the best 
  performance both in speed and memory consumption compared to the other
  dictionary types on my linux machine at home.
- unfortunately I got a segvio on my SUNOS-Sparc at work, which disappears when
  I change the minimum table size for map back to 512 (even 256 doesn't work).

So I am very interested in a working implementation of small hash-maps.

Thank you in advance,
  Dietmar Wolz                               E-Mail: dietmar@cs.tu-berlin.de
  Technische Universitaet Berlin                    Tel:  (030) 314-73137
  FB 13 / Sekr. FR 6-1 / Raum FR 6026              Fax: +49-30-314-23516
  Franklinstr. 28/29   10587 Berlin  Germany
-- 
 Dietmar Wolz                               E-Mail: dietmar@cs.tu-berlin.de
 Technische Universitaet Berlin                    Tel:  (030) 314-73137
 FB 13 / Sekr. FR 6-1 / Raum FR 6026              Fax: +49-30-314-23516
 Franklinstr. 28/29   10587 Berlin  Germany