🚀 go-pugleaf

In <1992Jul28.072501.6238@sics.se> Mats Carlsson writes:

> I don't know how other WAM based Prologs do it, but in SICStus Prolog,
> the compiler ensures that all permanent variables are initialized at
> the first 'call' instruction of every clause.  The variables are initialized
> by 'put_variable' if they are not otherwise initialized already.

> This is done to prevent there from being random bit patterns in the
> permanent variables in the event of a garbage collection.

Garbage collection has to deal with 3 types of permanent variables

	initialized and active (1)
	initialized and no longer active (2)
	not yet initialized (3)

(1) poses no problem

There are at least three ways to take care of (3):

1. at the first call, initialize the ones that are not yet active;
   it is the SICStus solution Mats described
2. make info available to the gc process so that it knows which permanent
   vars are active --> no need to initialize the non active ones
   it is the ProLog by BIM solution
3. let the gc process use the random bit patterns in a carefull way, i.e.
   only if they denote reasonable objects - e.g. if a pointer points in a
   non-active part of the heap, do not follow it - ; in this way, gc marks more
   than non-garbage, in fact it should be possible to write a program that
   attempts gc and recovers nothing while other schemas would collect
   something
   it is, as far as I know, the Quintus solution

(2) is dealt with in systems with trimming by making the length of the
    environment somehow available to the gc process, often by storing
    this length in *(CP-1)
SICStus does not trim environments, so perhaps it doesn't collect all garbage ?

> I don't have any figures, but I suspect that variables created by this
> mechanism would significantly contribute to the heap growth if one
> gets rid of unsafe variables.

Getting rid of unitialized variables by a globalizing put_variable,
seems indeed a bad idea.

Bart Demoen