The analyzer has detected an exception constructor that may throw another exception. Using such a class may cause the program to behave unexpectedly when handling exceptions.
Consider the following synthetic example:
#include <stdexcept>
class divide_by_zero_error : public std::invalid_argument
{
public:
divide_by_zero_error() : std::invalid_argument("divide_by_zero")
{
....
if (....)
{
throw std::runtime_error("oops!"); // <=
}
}
};
void example(int a, int b)
{
try
{
if (b == 0)
throw divide_by_zero_error ();
....
}
catch (const divide_by_zero_error &e)
{
....
}
// my_exception thrown and unhandled
}
In the code of the 'example' function, the programmer intends the raised 'divide_by_zero_error' exception to be handled, but instead of it, an 'std::runtime_error' exception will be thrown, which will not be caught by the subsequent 'catch' block. As a result, the exception will leave the 'example' function, thus causing one of the following situations:
Write and use custom exception classes with particular care because their constructors may throw exceptions at unexpected points – for example, when calling other functions. In the following example, when creating a logging exception, a second exception may be thrown by the 'Log' function:
#include <ios>
static void Log(const std::string& message)
{
....
// std::ios_base::failure may be thrown by stream operations
throw std::ios_base::failure("log file failure");
}
class my_logging_exception : public std::exception
{
public:
explicit my_logging_exception(const std::string& message)
{
Log(message); // <=
}
};
This diagnostic is classified as: