From: matz@... (Yukihiro Matsumoto)
Date: 2003-07-22T04:15:35+09:00
Subject: Re: matz thoughts on Rite ?

Hi,

In message "Re: matz thoughts on Rite ?"
    on 03/07/22, Lothar Scholz <mailinglists@scriptolutions.com> writes:

|It has one problem that a lot of GC algorithms have:
|fragmentation. But with a language like ruby it should be possible to
|write a separate memory compaction algorithm that uses some heuristics
|on the "Memory Allocated / Memory Used" quotient and call the
|compaction if absolute necessary, but this would make memory mangement
|of C extensions really hard to write.
|But this is independent of the used GC algorithm/library.

I don't care much about fragmentation.

|Have you thought about a user selectable (compile time) GC ?

I'm not sure what you mean by "user selectable (compile time)".

|The only problematic way with this is when using finalizeres.
|Eiffel is going the Java way here, which is "never trust that a finalizer is
|called". GC is only used for pure memory resources and nothing else
|(file handles etc.)

Current Ruby depends heavily upon finalizers.  That is a problem.

							matz.