The following program load tests GCPtr by repeatedly allocating and discarding objects until free memory is exhausted. When this occurs, a bad_alloc exception is thrown by new. Inside the exception handler, collect( ) is explicitly called to reclaim the unused memory, and the process continues. You can use this same technique in your own programs.
// Load test GCPtr by creating and discarding
// thousands of objects.
#include <iostream>
#include <new>
#include <limits>
#include "gc.h"
using namespace std;
// A simple class for load testing GCPtr.
class LoadTest {
int a, b;
public:
double n[100000]; // just to take up memory
double val;
LoadTest() { a = b = 0; }
LoadTest(int x, int y) {
a = x;
b = y;
val = 0.0;
}
friend ostream &operator<<(ostream &strm, LoadTest &obj);
};
// Create an inserter for LoadTest.
ostream &operator<<(ostream &strm, LoadTest &obj) {
strm << "(" << obj.a << " " << obj.b << ")";
return strm;
}
int main() {
GCPtr<LoadTest> mp;
int i;
for(i = 1; i < 20000; i++) {
try {
mp = new LoadTest(i, i);
} catch(bad_alloc xa) {
// When an allocation error occurs, recycle
// garbage by calling collect().
cout << "Last object: " << *mp << endl;
cout << "Length of gclist before calling collect(): "
<< mp.gclistSize() << endl;
GCPtr<LoadTest>::collect();
cout << "Length after calling collect(): "
<< mp.gclistSize() << endl;
}
}
return 0;
}
A portion of the output from the program (with the display option off) is shown here. Of course, the precise output you see may differ because of the amount of memory available in your system and the compiler that you are using.
Last object: (518 518)
Length of gclist before calling collect(): 518
Length after calling collect(): 1
Last object: (1036 1036)
Length of gclist before calling collect(): 518
Length after calling collect(): 1
Last object: (1554 1554)
Length of gclist before calling collect(): 518
Length after calling collect(): 1
Last object: (2072 2072)
Length of gclist before calling collect(): 518
Length after calling collect(): 1
Last object: (2590 2590)
Length of gclist before calling collect(): 518
Length after calling collect(): 1
Last object: (3108 3108)
Length of gclist before calling collect(): 518
Length after calling collect(): 1
Last object: (3626 3626)
Length of gclist before calling collect(): 518
Length after calling collect(): 1
There are a few restrictions to using GCPtr:
- You cannot create global GCPtrs. Recall that a global object goes out of scope after the rest of the program ends. When a global GCPtr goes out of scope, the GCPtr destructor calls collect( ) to try to release the unused memory. The trouble is that, depending on how your C++ compiler is implemented, gclist may have already been destroyed. In this case, executing collect( ) will cause a runtime error. Therefore, GCPtr should be used only when creating local objects.
- When using dynamically allocated arrays, you must specify the size of the array when you declare a GCPtr that will point to it. There is no mechanism that enforces this, however, so be careful.
- You must not attempt to release the memory pointed to by a GCPtr by explicitly using delete. If you need to immediately release an object, call collect( ).
- A GCPtr object must point only to memory that is dynamically allocated via new. Assigning to a GCPtr object a pointer to any other memory will cause an error when the GCPtr object goes out of scope because an attempt will be made to free memory that was never allocated.
- It is best to avoid circular pointer references for reasons described earlier in this chapter. Although all allocated memory is eventually released, objects containing circular references remain allocated until the program ends, rather than being released when they are no longer used by another program element.
Some Things to Try It is easy to tailor GCPtr to the needs of your applications. As explained earlier, one of the changes that you might want to try is collecting garbage only after some metric has been reached, such as gclist reaching a certain size, or after a certain number of GCPtrs have gone out of scope.
An interesting enhancement to GCPtr is to overload new so that it automatically collects garbage when an allocation failure occurs. It is also possible to bypass the use of new when allocating memory for a GCPtr and use factory functions defined by GCPtr instead. Doing this lets you carefully control the dynamic allocation of memory, but it makes the allocation process fundamentally different than it is for C++’s built-in approach.
You might want to experiment with other solutions to the circular reference problem. One way is to implement the concept of a weak reference, which does not prevent garbage collection from occurring. You would then use a weak reference whenever a circular reference was needed.
Perhaps the most interesting variation on GCPtr is found in Chapter 3. There, a multithreaded version is created in which garbage collection takes place automatically, when free CPU time exists.