Interfacing the Garbage Collector

Warning: Warning: Article under development

This article tries to explain, how the EiffelStudio GC works together with the runtime. It should help the programmer to better understand the C and byte code generated by EiffelStudio. The exact mechanisms behind the GC are not of concern here.

Introduction

Most important thing first, EiffelStudio has a mark and sweep moving Garbage Collector (GC) (instead of moving GC one may say compacting GC).

A mark and sweep GC traverses during the mark phase all the objects, that are reachable from a given root set (locals, arguments uws...). After that, the not reached objects are detected as garbage and may be freed. To avoid memory fragmentation the GC may move around the reached objects.

Thats about all there is to know for interfacing the GC, for modifying or even extending him a little more in-deft might be handy but is not provided here.

To recap, the GC needs to know the current root set and enough information to traverse the objects currently on the heap. The runtime and the generated code on the other side need to be able to cope with objects that change their memory location after a GC run.

The root set

The root set of EiffelStudio is surprisingly complex. Only the more interesting part is shown here.

The loc_set

The following two snippets shows the feature foo:

foo (s: STRING; i: INTEGER): LIST [ANY]
   local
      b: BOOLEAN
      a: ANY
   do
      ...
   end

This feature has two arguments, two locals and a return value. The ones that are of a reference type, namely s, a and the return value, need to be protected during the execution of feature foo. They are pushed on the loc_set stack and thus become part of the root set as long as the frame is active.

EIF_REFERENCE Fadkw7f (EIF_REFERENCE Current,
                       EIF_REFERENCE arg1, 
                       EIF_INTEGER_32 arg2) 
{
   EIF_BOOLEAN loc1 = (EIF_BOOLEAN) 0;
   EIF_REFERENCE loc2 = (EIF_REFERENCE) 0;
   EIF_REFERENCE Result = (EIF_REFERENCE) 0;
 
   RTLI(4);                //Reserve space for 4 references on the loc_stack
   RTLR(0,arg1);           //Push the address of s onto loc_stack
   RTLR(1,Current);        //Push the address of Current onto loc_stack
   RTLR(2,loc2);           //Push the address of a onto loc_stack
   RTLR(3,Result);         //Push the address of the Result onto loc_stack
   ...
 
   RTLE;                   //Pop the 4 entries from the loc_stack.
   return Result;
}

Feature foo has