From: matz@... (Yukihiro Matsumoto)
Date: 2001-12-29T01:15:05+09:00
Subject: [ruby-talk:29539] Re: GC and values in extensions

Hi,

In message "[ruby-talk:29532] Re: GC and values in extensions"
    on 01/12/28, Paul Brannan <paul@atdesk.com> writes:

|  VALUE foo(VALUE self) {
|    VALUE x = rb_str_new2("foo");
|    long y = (long)x; /* Assume x and y now have the same bit pattern */
|    x = Qnil;
|    ...
|    return Qnil;
|  }
|
|then the newly allocated string will not be freed until foo returns? 

It won't.

|Or
|if I happen to store an integer that happens to look like a pointer to
|the heap, then it is possible that some object that has no more
|references may not actually get garbage-collected?

They may not.

|So there is no way
|to depend on objects being garbage-collected, even if there are no more
|references?

No.  That is what "conservative" means.

|This also sounds a little inefficient to me (though it makes things
|convenient for extension writers).  Is it?

It's pretty efficient.  I've read the survey that misdetermination was
less than 5% on a Lisp implementation using conservative GC.  And from
my brief experimental result, it happens even less in Ruby (virtually
zero).  I think you can ignore accidental misdetermination (except
aritificial ones).

							matz.