Orchid XP v8 wrote:
> Obviously, stopping your entire program (all parallel threads!) so you 
> can run the GC isn't a fantastic idea. So various more sophisticated 
> schemes have been invented (e.g., generational GC). In any case, for 
> most designs of GC, the entire chain of objects will be reclaimed in a 
> single pass of the GC. (Although it might be the next "major" rather 
> than "minor" pass or something, depending on the exact design details.)

I could imagine that there is an optimization available when individual 
references go out of scope (although simply deallocating those objects 
should suffice), and in a language that checks these things, there may 
be opportunities whenever a reference is changed.

Not a full-blown check of every dynamic object (which would hurt 
performance), but a limited check.  But I'll leave the details to 
whoever implements it.

In a language that doesn't have built-in GC, properly encapsulating 
references, and properly building destructors, could do most of the GC 
work; but then the point of GC is to allow the programmer to forget 
these things.  But at least I have something to consider if I ever do a 
ground-upwards rewrite of my modeler (which is in C++).

Regards,
John

Post a reply to this message