Re: hash tables not in class hierarchy. Why?

From: Erick Gallesio <eg_at_saxo.essi.fr>
Date: Mon, 16 Dec 1996 21:01:13 +0100

> "Harvey J. Stein" writes
>
> A long time ago (circa v2.1.5, I guess) I asked about about why hash
> tables weren't integrated into the type tree in STk.

The problem with hash table is that they are dynamically loaded.
In old version of STk, the problem was that hash table could be
loaded before STklos and it was difficult to take them into account.
Today the situation a better (a little bit only) since STklos is
statically linked in the interpreter. However, you can always
load hash.so before initializing STklos.
What we need here is a mechanism (compatible with previous
version of dynamic loading to not break existing code).
I think that it could be in the initialization part, something which says:
"If STklos is initialized, I want to define a new class for this
new type of object and it's name shoud be <xxx>". What is not satisfaisant
here is that
        1. the loading order must be taken into account in C code
        2. it seems difficult to explicit an inheritance scheme for the
           new type here without introducing interdependence problems
           if we want something general. But perhaps forbidding inheritance
           and imposing that all the new types defined in STk extensions
           be direct descendant of <object> is acceptable.
If these seem acceptable (after all it's better than nothing ;-), I can
implement it for next release.

                -- Erick
Received on Mon Dec 16 1996 - 21:01:50 CET