Error Handling Alternatives
With the traditional programming languages of yore, a developer's alternatives for handling errors consisted mainly of two options:
-
1. Ignore them.
2. Check return codes (fastidiously).
-
The first alternative should find its way only into toy programs such as school assignments or magazine articles that aren't discussing exceptions. Such a non-strategy just won't do for anything you plan to actually execute.
Consider the program in Listing 1, for example, which deletes an entire directory tree using standard POSIX directory-handling functions. (For an explanation of these functions, see my column, "Code Capsules," CUJ, June 1993). What could possibly go wrong? Well, I see the following potentialities:
- The chosen directory doesn't exist.
- The directory is read-protected.
- Files in the directory are delete-protected.
- The directory itself is delete protected.
For these reasons, most C library functions return a status code which you can test to detect if an error occurred. As Listing 2 illustrates, making use of these codes requires a lot of checking throughout a deeply nested call chain. The deeper it gets, the more tired you get. Since the goal in such cases is to return to some safe state, it would be nice if there were some way to yank the thread of execution "out of the deep" and place it higher up in the call chain.
Well, there is such a mechanism, which brings us to a third error handling alternative:
3. Use non-local jumps to re-route the thread of execution.
This is what C's setjmp/longjmp mechanism is all about.
As you can see in Listing 3, setjmp uses a jump buffer, of type jmp_buf, which is a compiler-defined structure that records information sufficient to restore control to a previous point in the program. Such information might include the stack pointer, instruction pointer, etc. A jump buffer must be global, of course. (Think about it — references to the jump buffer transcend a single function.)
When setjmp first executes, it initializes the buffer and returns zero. If a longjmp call further down in the call chain refers to the same jmp_buf, then control transfers immediately back to the setjmp call, this time as if it returned the second argument from the longjmp call. It would really confuse things if you could return a zero via a longjmp call (which you would never do, I realize), so if some other foolish programmer makes the attempt, setjmp will return a 1 instead.
You may be wondering why the volatile keyword appears in this version of the program. Compilers do funny things sometimes, and it turns out that any local variables declared in the same block as the setjmp call are not guaranteed to retain their original values after a longjmp, unless you adorn their declaration with the volatile qualifier.
It would seem that all problems are solved, and I can bid you adieu. It turns out, however, that jumping out of a function into another one higher up in the call chain can be risky in C++. The main problem, as Listing 4 illustrates, is that automatic variables created in the execution thread between the setjmp and longjmp may not get properly destroyed. Fortunately, C++ provides for such alternate returns that know about destructors, via the exception handling mechanism. So, if you program in C++, you might consider this final error handling alternative:
4. Use Exceptions.
That's what the remainder of this article is about.
Stack Unwinding
The program in Listing 5 is a rewrite of Listing 4 using exceptions. You turn on exception handling in a region of code by surrounding it with a try block. You place exception handlers immediately after the try block, in a series of catch clauses. An exception handler is much like a definition for a function, named by the keyword catch.
You raise an exception with a throw expression. The statement
throw 1;
causes the compiler to search back up the chain of nested function calls for a handler that can catch an integer, so control passes to that handler in main. The run-time environment invokes destructors for all automatic objects constructed after execution entered the try block. This process of destroying automatic variables on the way to an exception handler is called stack unwinding.
As you can see, executing a throw expression is semantically similar to calling longjmp, and a handler is like a setjmp. Just as you cannot longjmp into a function that has already returned, you can only jump to a handler associated with an active try block — one for which execution control has not yet exited. After a handler executes, control passes to the first statement after all the handlers associated with the try block (the return statement, in this case).
The notable differences between exceptions and the setjmp/longjmp facility are:
1. Exception handling is a language mechanism, not a library feature. The overhead of transferring control to a handler is invisible to the programmer.
2. The compiler generates code to keep track of all automatic variables that have destructors and executes those destructors when necessary (it unwinds the stack).
3. Local variables in the function that contains the try block are safe — you don't need to declare them volatile to keep them from being corrupted as you do with setjmp.
4. Handlers are found by matching the type of the exception object that you throw. This allows you to handle categories of exceptions with a single handler, and to classify them via inheritance.
5. Exception handling is a run-time mechanism. You can't always tell which handler will catch a specific exception by examining the source code.
The last point is significant. C++ exception handling allows a clean separation between error detection and error handling. A library developer may be able to detect when an error occurs, such as an argument out of range, but the developer doesn't know what to do about it. You, the user, can't always detect an exceptional condition, but you know how your application needs to handle it. Hence, exceptions constitute a protocol for runtime error communication between components of an application.
It is also significant to realize that C++ exception handling is designed around the termination model. That is, when an exception is thrown, there is no direct way to get back to the throw point and resume execution where you left off, as you can in languages like Ada that follow the resumption model. C++ exceptions are intended for reporting rare, synchronous events.
Catching Exceptions
Since exceptions are a run-time and not a compile-time feature, standard C++ specifies the rules for matching exceptions to catch-parameters a little differently than those for finding an overloaded function to match a function call. You can define a handler for an object of type T several different ways. In the examples that follow, the variable t is optional, just as it is for ordinary functions in C++:
catch(T t)
catch(const T t)
catch(T& t)
catch(const T& t)
Such handlers can catch exception objects of type E if:
1. T and E are the same type, or
2. T is an accessible base class of E at the throw point, or
3. T and E are pointer types and there exists a standard pointer conversion from E to T at the throw point. T is an accessible base class of E if there is an inheritance path from E to T with all derivations public.
To understand the third rule, let E be a type pointing to type F, and T be a type that points to type U. Then there exists a standard pointer conversion from E to T if:
1. T is the same type as E, except it may have added any or both of the qualifiers const and volatile, or
2. T is void*, or
3. U is an unambiguous, accessible base class of F. U is an unambiguous base class of F if F's members can refer to members of U without ambiguity (this is usually only a concern with multiple inheritance).
The bottom line of all these rules is that exceptions and catch parameters must either match exactly, or the exception caught by pointer or reference must be derived from the type of the catch parameter. For example, the following exception is not caught:
#include <iostream>
void f();
main()
{ try
{
f();
}
catch(long)
{
cerr << "caught a long" << endl;
}
}
void f()
{
throw 1; // not a long!
}
When the system can't find a handler for an exception, it calls the standard library function terminate, which by default aborts the program. You can substitute your own termination function by passing a pointer to it as a parameter to the set_terminate library function. (See Listing 6)
The following exception is caught, since there is a handler for an accessible base class:
#include <iostream>
using namespace std;
class B {};
class D : public B {};
void f();
main()
{
try
{
f();
}
catch(const B&)
{
cerr << "caught a B" << endl;
}
}
void f()
{
throw D();
}
An exception caught by a pointer can also be caught by a void* handler.
Question: Since the context of a throw statement is lost when control transfers to a handler, how can the exception object still be available in the handler?
Answer: Good question! Here is another area where exception handling differs from function calls. The runtime mechanism creates a temporary copy of the thrown object for use by the handler. This suggests that it is never really a good idea to define catch parameters with value semantics, since a second copy will be made. Also, it can be dangerous to define a catch parameter as a pointer, because you may not know if it came from the heap or not (in which case you must delete it). The safest strategy is always to catch by reference. And even though the exception is a temporary, you can catch it as a non-const reference if you want (yet another departure from the normal function-processing rules to accommodate exception handling).
Standard Exceptions
The Standard C++ library throws exception objects of types found in the following class hierarchy:
exception
logic_error
domain_error
invalid_argument
length_error
out_of_range
runtime_error
range_error
overflow_error
underflow_error
bad_alloc
bad_cast
bad_exception
bad_typeid
A logic error indicates an inconsistency in the internal logic of a program, or a violation of pre-conditions on the part of client software. For example, the substr member function of the standard string class throws an out_of_range exception if you ask for a substring beginning past the end of the string. Run-time errors are those that you cannot easily predict in advance, and are usually due to forces external to a program. A range error, for instance, violates a post-condition of a function, such as arithmetic overflow from processing legal arguments.
A bad_alloc exception occurs when heap memory is exhausted. (See "Memory Management" in Part 2 next month.) C++ will generate a bad_cast exception when a dynamic_cast to a reference type fails. If you rethrow an exception from within an unexpected handler, it gets converted into a bad_exception. (See "Exception Specifications" in Part 2.) If you attempt to apply the typeid operator to a null expression, you get a bad_typeid exception.
The program in Listing 7 derives the dir_error exception class from runtime_error, since the associated errors are detected by the return status of system services. The exception base class has a member function called what that returns the string argument that created the exception. I use this member function to pass information about the error to the exception handler.
The handler with the ellipsis specification (catch (...)) will catch any exception, so the order of handlers in program text is significant. You should always order catch clauses according to their respective types, from the most specific to the most general.
We're Not Done Yet
I know what you're thinking. Things can't be this easy. There has to be a "catch" somewhere (ha ha). Well there is. The error handling plot really thickens when allocating resources fails. Next month I'll explore that issue, and also share Chuck's Pretty Good Error Handling Strategy for C++ development. o
Acknowledgment
This article is based on material from the author's forthcoming book, C and C++ Code Capsules: A Guide for Practitioners, Prentice-Hall, 1998.
Chuck Allison is Consulting Editor and a former columnist with CUJ. He is the owner of Fresh Sources, a company specializing in object-oriented software development, training, and mentoring. He has been a contributing member of J16, the C++ Standards Committee, since 1991, and is the author of C and C++ Code Capsules: A Guide for Practitioners, Prentice-Hall, 1998. You can email Chuck at cda@freshsouces.com.
