Ruby Hacking Guide/Garbage Collection

The Runtime Structure of a Program
Before diving into the content of this chapter, let us review the organization of memory at program execution time. This chapter will involve some low-level components of computer architecture, so familiarizing oneself with some basic concepts ahead of time will be required. Moreover, these concepts will be required for later chapters as well. Let's get them out of the way here.

Segments
Most C programs have the following components in their memory space:

 text area, a store of static and global variables, the machine stack, the heap. 

The text area is where the code is stored. The second component should be clear. The machine stack is where functions' arguments and local variables are stacked. The heap is what is allocated by.

Let's discuss the machine stack in detail. Being called a machine "stack", it obviously has a stack-like construction. In other words, we can keep adding new elements to the top. In practice values are added to the stack in individual  units but conceptually there is a larger unit called the stack frame.

Each stack frame corresponds to one function call. In other words, each function call will add one stack frame and a stack frame is taken off on. If we extremely simplify this, the machine stack could look like Figure 1.

the top +-+ +-+ +-+ +-+ +-+ +-+ the bottom
 * stack frame | <-- the frame for the currently running function
 * stack frame |
 * stack frame |
 * stack frame |
 * stack frame |

Figure 1: The machine stack

In this figure we marked the extreme end of the stack "the top," but the machine stacks does not necessarily address frames from low to high. For example on x86 machines the stack grows from higher addresses to lower ones.

With  memory of arbitrary size can be allocated. is the machine stack version of this. However, memory allocated by  does not need to be deallocated. Or, rather, it may be better to say that that the memory "gets" deallocated with the function's. Thus -allocated values cannot be used as a function's   value. This is the same as saying "pointers to local variables cannot be returned."

All this is fine. It basically means that we can locally allocate arrays which change in length dynamically.

However there are environments where there is no native. Many would prefer to use an  in those environments as well, so a function with the same behavior could be written in C. In this case, however, it may only be implemented to "not require deallocation," but may not necessarily be allocating memory on the machine stack. In fact, it normally would not do this. If it could do that, there may as well be a native implementation of.

How can we implement  in C? The most straight-forward implementation first allocates memory with. It then stores the caller function and the allocated address on a global list. Then the next time  is called, if there is any memory that was allocated for a function which has already concluded, it can be  'ed (see Figure 2).

+---+     ++ +---+      ++ +---+      ++ +---+      | alloca(32) | free the memory allocated for D +---+ +---+
 * main     |      | main       |
 * A        | ===> | A          |
 * B        |      | B          | mark that B -> alloca(32)
 * C        |      ++
 * D        |
 * alloca(8) | mark that D -> alloca(8)

Figure 2: The behavior of a C implementation of 

Ruby's  is an implementation of just such an emulated.

Outline
Now let's begin the main subject of this chapter, garbage collection.

Problems with the allocation framework
Comments, suggestions, and criticisms may be sent to Aoki MINERŌ . Please direct translation comments, suggestions, and criticisms to the translator of this chapter, mitcho (Michael Yoshitaka Erlewine) .