Stream States
Let’s take a closer look at what happens for inappropriate input. A cin or cout object contains a data member (inherited from the ios_base class) that describes the stream state. A stream state (defined as type iostate, which, in turn, is a bitmask type, such as described earlier) consists of the three ios_base elements: eofbit, badbit, and failbit. Each element is a single bit that can be 1 (set) or 0 (cleared). When a cin operation reaches the end of a file, it sets eofbit. When a cin operation fails to read the expected characters, as in the earlier example, it sets failbit. I/O failures, such as trying to read a non-accessible file or trying to write to a write-protected disk, also can set failbit to 1. The badbit element is set when some undiagnosed failure may have corrupted the stream. (Implementations don’t necessarily agree about which events set failbit and which set badbit.) When all three of these state bits are set to 0, everything is fine. Programs can check the stream state and use that information to decide what to do next. Table 17.4 lists these bits, along with some ios_base methods that report or alter the stream state.
Setting States
Two of the methods in Table 17.4, clear() and setstate(), are similar. Both reset the state, but they do so in a different fashion. The clear() method sets the state to its argument. Thus, the following call uses the default argument of 0, which clears all three state bits (eofbit, badbit, and failbit):
clear();
Similarly, the following call makes the state equal to eofbit; that is, eofbit is set, and the other two state bits are cleared:
clear(eofbit);
The setstate() method, however, affects only those bits that are set in its argument. Thus, the following call sets eofbit without affecting the other bits:
setstate(eofbit);
So if failbit was already set, it stays set.
Why would you reset the stream state? For a program writer, the most common reason is to use clear() with no argument to reopen input after encountering mismatched input or end-of-file; whether doing so makes sense depends on what the program is trying to accomplish. You’ll see some examples shortly. The main purpose for setstate() is to provide a means for input and output functions to change the state. For example, if num is an int, the following call can result in operator>>(int &) using setstate() to set failbit or eofbit:
cin >> num; // read an int
I/O and Exceptions
Suppose that an input function sets eofbit. Does this cause an exception to be thrown? By default, the answer is no. However, you can use the exceptions() method to control how exceptions are handled.
First, here’s some background. The exceptions() method returns a bitfield with three bits corresponding to eofbit, failbit, and badbit. Changing the stream state involves either clear() or setstate(), which uses clear(). After changing the stream state, the clear() method compares the current stream state to the value returned by exceptions(). If a bit is set in the return value and the corresponding bit is set in the current state, clear() throws an ios_base::failure exception. This would happen, for example, if both values had badbit set. It follows that if exceptions() returns goodbit, no exceptions are thrown. The ios_base::failure exception class derives from the std::exception class and thus has a what() method.
The default setting for exceptions() is goodbit—that is, no exceptions thrown. However, the overloaded exceptions(iostate) function gives you control over the behavior:
cin.exceptions(badbit); // setting badbit causes exception to be thrown
The bitwise OR operator (|), as discussed in Appendix E, allows you to specify more than one bit. For example, the following statement results in an exception being thrown if either badbit or eofbit is subsequently set:
cin.exceptions(badbit | eofbit);
Listing 17.12 modifies Listing 17.11 so that the program throws and catches an exception if failbit is set.
// cinexcp.cpp -- having cin throw an exception
#include <iostream>
#include <exception>
int main()
{
using namespace std;
// have failbit cause an exception to be thrown
cin.exceptions(ios_base::failbit);
cout << "Enter numbers: ";
int sum = 0;
int input;
try {
while (cin >> input)
{
sum += input;
}
} catch(ios_base::failure & bf)
{
cout << bf.what() << endl;
cout << "O! the horror!
";
}
cout << "Last value entered = " << input << endl;
cout << "Sum = " << sum << endl;
return 0;
}
Here is a sample run of the program in Listing 17.12; the what() message depends on the implementation:
Enter numbers: 20 30 40 pi 6
ios_base failure in clear
O! the horror!
Last value entered = 40.00
Sum = 90.00
So that’s how you can use exceptions for input. But should you use them? It depends on the context. For this example, the answer is no. An exception should catch an unusual, unexpected occurrence, but this particular program uses a type mismatch as the intended way to exit the loop. It might make sense, however, for this program to throw an exception for badbit because that circumstance would be unexpected. Or if the program were designed to read numbers from a data file up to end-of-file, it might make sense to throw an exception for failbit because that would represent a problem with the data file.
Stream State Effects
An if or while test such as the following tests as true only if the stream state is good (all bits cleared):
while (cin >> input)
If a test fails, you can use the member functions in Table 17.4 to discriminate among possible causes. For example, you could modify the central part of Listing 17.11 to look like this:
while (cin >> input)
{
sum += input;
}
if (cin.eof())
cout << "Loop terminated because EOF encountered
";
Setting a stream state bit has a very important consequence: The stream is closed for further input or output until the bit is cleared. For example, the following code won’t work:
while (cin >> input)
{
sum += input;
}
cout << "Last value entered = " << input << endl;
cout << "Sum = " << sum << endl;
cout << "Now enter a new number: ";
cin >> input; // won't work
If you want a program to read further input after a stream state bit has been set, you have to reset the stream state to good. This can be done by calling the clear() method:
while (cin >> input)
{
sum += input;
}
cout << "Last value entered = " << input << endl;
cout << "Sum = " << sum << endl;
cout << "Now enter a new number: ";
cin.clear(); // reset stream state
while (!isspace(cin.get()))
continue; // get rid of bad input
cin >> input; // will work now
Note that it is not enough to reset the stream state. The mismatched input that terminated the input loop is still in the input queue, and the program has to get past it. One way is to keep reading characters until reaching white space. The isspace() function (see Chapter 6, “Branching Statements and Logical Operators”) is a cctype function that returns true if its argument is a white-space character. Or you can discard the rest of the line instead of just the next word:
while (cin.get() != '
')
continue; // get rid rest of line
This example assumes that the loop terminated because of inappropriate input. Suppose, instead, that the loop terminated because of end-of-file or because of a hardware failure. Then the new code disposing of bad input makes no sense. You can fix matters by using the fail() method to test whether the assumption was correct. Because for historical reasons, fail() returns true if either failbit or eofbit is set, the code has to exclude the latter case. The following code shows an example of such exclusion:
while (cin >> input)
{
sum += input;
}
cout << "Last value entered = " << input << endl;
cout << "Sum = " << sum << endl;
if (cin.fail() && !cin.eof() ) // failed because of mismatched input
{
cin.clear(); // reset stream state
while (!isspace(cin.get()))
continue; // get rid of bad input
}
else // else bail out
{
cout << "I cannot go on!
";
exit(1);
}
cout << "Now enter a new number: ";
cin >> input; // will work now