Specification of the gnu Smalltalk virtual machine

This document exposes the virtual machine’s architecture with respect to bytecode execution, and documents how the methods are stored as objects in memory. The first part of the specification details a subset of the reflection classes that the virtual machine accesses in order to execute the code. The second part details the bytecode set and the invariants that should hold for a method to be valid. This is a low-level specification. A formalization of the Smalltalk language semantics, on which the virtual machine is based, can be found in [Wol87]. 1 Structure of instances of CompiledCode Instances of CompiledCode can have byte-sized indexed instance variables which store the bytecodes; in addition, they have three named instance variables: • literals is an Array that holds the constants and references to globals that are referenced in the code (the method’s literal pool). • flags is a SmallInteger that stores miscellaneous information about the method (number of locals and arguments, stack size, and so on). CompiledCode has two concrete subclasses, CompiledMethod and CompiledBlock. These two classes match the two fundamental constructs for organizing program flow in the Smalltalk language—methods and blocks. Logically, methods represent actions made on an object, or in general requests to an object: by defining a method, an object exposes a contract that hides implementation details; blocks, instead, are objects that access a series of statements, and they can be passed around or stored in an instance variable. For example, the following code myList collect: [ :x | x + 1 ]. uses blocks like this Lisp lambda function: (mapcar (lambda (x) (+ 1 x)) *my-list*) From this example, one can see that methods like collect: live inside a class (the class of myList) and can be invoked through their name; instead a block is anonymous.