Chapter 15

Use Smart Pointers

Smart pointers allow you to write code that automatically prevents memory leaks with exception handling. As you read in Chapter 13, smart pointer objects are allocated on the stack, so whenever the smart pointer object is destroyed, it calls delete on the underlying dumb pointer. Here is an example of funcTwo() using the auto_ptr smart pointer template class in the Standard Library:

#include <memory>

using namespace std;

void funcOne() throw(exception)

{

string str1;

auto_ptr<string> str2(new string(“hello”)); funcTwo();

}

With smart pointers, you never have to remember to delete the underlying dumb pointer: the smart pointer destructor does it for you, whether you leave the function via an exception or leave the function normally.

Common Error-Handling Issues

Whether or not you use exceptions in your programs is up to you and your colleagues. However, we strongly encourage you to formalize an error-handling plan for your programs, regardless of your use of exceptions. If you use exceptions, it is generally easier to come up with a unified error-handling scheme, but it is not impossible without exceptions. The most important aspect of a good plan is uniformity of error handling throughout all the modules of the program. Make sure that every programmer on the project understands and follows the error-handling rules.

This section discusses the most common error-handling issues in the context of exceptions, but the issues are also relevant to programs that do not use exceptions.

Memory Allocation Errors

Despite the fact that all of our examples so far in this book have ignored the possibility, memory allocation can, and will, fail. However, production code must account for memory allocation failures. C++ provides several different ways to handle memory errors.

The default behaviors of new and new[] are to throw an exception of type bad_alloc, defined in the <new> header file, if they cannot allocate memory. Your code should catch these exceptions and handle them appropriately. The definition of “appropriate” depends on your particular application. In some cases, the memory might be crucial for your program to run correctly, in which case printing an error message and exiting is the best course of action. Other times, the memory might be necessary only for a particular operation or task, in which case you can print an error message and fail the particular operation, but keep the program running.

Thus, all your new statements should look something like this:

424

Handling Errors

try {

ptr = new int[numInts]; } catch (bad_alloc& e) {

cerr << “Unable to allocate memory!\n”;

// Handle memory allocation failure. return;

}

// Proceed with function that assumes memory has been allocated.

You could, of course, bulk handle many possible new failures with a single try/catch block at a higher point on the program, if it will work for your program.

Another consideration is that logging an error might try to allocate memory. If new fails, there might not be enough memory left even to log the error message.

Nothrow new

As mentioned in Chapter 13, if you don’t like exceptions, you can revert to the old C model in which memory allocation routines return the NULL pointer if they cannot allocate memory. C++ provides nothrow versions of new and new[], which return NULL instead of throwing an exception if they fail to allocate memory.

ptr = new(nothrow) int[numInts];

if (ptr == NULL) {

cerr << “Unable to allocate memory!\n”; // Handle memory allocation failure. return;

}

// Proceed with function that assumes memory has been allocated.

The syntax is a little strange: you really do write “nothrow” as if it’s an argument to new (which it is).

Customizing Memory Allocation Failure Behavior

C++ allows you to specify a new handler callback function. By default, there is no new handler, so new and new[] just throw bad_alloc exceptions. However, if there is a new handler, the memory allocation routine calls the new handler upon memory allocation failure instead of throwing an exception. If the new handler returns, the memory allocation routines attempt to allocate memory again, calling the new handler again if they fail. This cycle could become an infinite loop unless your new handler changes the situation with one of four alternatives. Practically speaking, some of the four options are better than others. Here is the list with commentary:

Make more memory available. One trick to expose space is to allocate a large chunk of memory at program start-up, and then to free it with delete in the new handler. If the current request for memory is for a size smaller than the one you free in the new handler, the memory allocation routine will now be able to allocate it. However, this technique doesn’t buy you much. If you hadn’t preallocated that chunk of memory, the request for memory would have succeeded in the first place and wouldn’t have needed to call the new handler. One of the only benefits is that you can log a warning message about low memory in the new handler.

Throw an exception. new and new[] have throw lists that say they will throw exceptions only of type bad_alloc. So, unless you want to create a call to unexpected(), if you throw an exception from the new handler, throw bad_alloc or a subclass. However, you don’t need a new handler to throw an exception: the default behavior does so for you. Thus, if that’s all your new handler does, there’s no reason to write a new handler in the first place.

425

Chapter 15

Set a different new handler. Theoretically, you could have a series of new handlers, each of which tries to create memory and sets a different new handler if it fails. However, such a scenario is usually more complicated than useful.

Terminate the program. This option is the most practical and useful of the four. Your new handler can simply log an error message and terminate the program. The advantage of using a new handler over catching the bad_alloc exception and terminating in the exception handler is that you centralize the failure handling to one function, and you don’t need to litter your code with try/catch blocks. If there are some memory allocations that can fail but still allow your program to succeed, you can simply set the new handler back to its default of NULL temporarily before calling new in those cases.

If you don’t do one of these four things in your new handler, any memory allocation failure will cause an infinite loop.

You set the new handler with a call to set_new_handler(), declared in the <new> header file. set_new_handler() completes the trio of C++ functions to set callback functions. The other two are setterminate() and set_unexpected(), which are discussed earlier in the chapter. Here is an example of a new handler that logs an error message and aborts the program:

void myNewHandler()

{

cerr << “Unable to allocate memory. Terminating program!\n”; abort();

}

The new handler must take no arguments and return no value. This new handler calls the abort() function declared in <cstdlib> to terminate the program.

You can set the new handler like this:

#include <new> #include <cstdlib> #include <iostream>

using namespace std;

int main(int argc, char** argv)

{

//Code omitted

//Set the new new_handler and save the old.

new_handler oldHandler = set_new_handler(myNewHandler);

//Code that calls new

//Reset the old new_handler. set_new_handler(oldHandler);

//Code omitted

return (0);

}

Note that new_handler is a typedef for the type of function pointer that set_new_handler() takes.

426

Handling Errors

Errors in Constructors

Before C++ programmers discover exceptions, they are often stymied by error handling and constructors. What if a constructor fails to construct the object properly? Constructors don’t have a return value, so the standard preexception error-handling mechanism doesn’t work. Without exceptions, the best you can do is to set a flag in the object specifying that it is not constructed properly. You can provide a method, with a name like checkConstructionStatus(), which returns the value of that flag, and hope that clients remember to call the function on the object after constructing it.

Exceptions provide a much better solution. You can throw an exception from a constructor, even though you can’t return a value. With exceptions you can easily tell clients whether or not construction of the object succeeded. However, there is one major problem: if an exception leaves a constructor, the destructor for that object will never be called. Thus, you must be careful to clean up any resources and free any allocated memory in constructors before allowing exceptions to leave the constructor. This problem is the same as in any other function, but it is subtler in constructors because you’re accustomed to letting the destructors take care of the memory deallocation and resource freeing.

Here is an example of the constructor from the GameBoard class from Chapter 11 retrofitted with exception handling:

GameBoard::GameBoard(int inWidth, int inHeight) throw(bad_alloc) :

mWidth(inWidth), mHeight(inHeight)

{

int i, j;

mCells = new GamePiece* [mWidth];

try {

for (i = 0; i < mWidth; i++) { mCells[i] = new GamePiece[mHeight];

}

} catch (...) {

//

//Clean up any memory we already allocated, because the destructor

//will never be called. The upper bound of the for loop is the index

//of the last element in the mCells array that we tried to allocate

//(the one that failed). All indices before that one store pointers to

//allocated memory that must be freed.

//

for (j = 0; j < i; j++) { delete [] mCells[j];

}

delete [] mCells;

// Translate any exception to bad_alloc. throw bad_alloc();

}

}

It doesn’t matter if the first new throws an exception because the constructor hasn’t allocated anything else yet that needs freeing. If any of the subsequent new calls throw exceptions, though, the constructor must clean up all of the memory already allocated. It catches any exception via ... because it doesn’t know what exceptions the GamePiece constructors themselves might throw.

427