This is an example of what I often do when I want to add some information to an exception:
std::stringstream errMsg;
errMsg << \"Could not load config
A really nicer way would be creating a class (or classes) for the exceptions.
Something like:
class ConfigurationError : public std::exception {
public:
ConfigurationError();
};
class ConfigurationLoadError : public ConfigurationError {
public:
ConfigurationLoadError(std::string & filename);
};
The reason is that exceptions are much more preferable than just transferring a string. Providing different classes for the errors, you give developers a chance to handle a particular error in a corresponded way (not just display an error message). People catching your exception can be as specific as they need if you use a hierarchy.
a) One may need to know the specific reason
} catch (const ConfigurationLoadError & ex) {
// ...
} catch (const ConfigurationError & ex) {
a) another does not want to know details
} catch (const std::exception & ex) {
You can find some inspiration on this topic in https://books.google.ru/books?id=6tjfmnKhT24C Chapter 9
Also, you can provide a custom message too, but be careful - it is not safe to compose a message with either std::string or std::stringstream or any other way which can cause an exception.
Generally, there is no difference whether you allocate memory (work with strings in C++ manner) in the constructor of the exception or just before throwing - std::bad_alloc exception can be thrown before the one which you really want.
So, a buffer allocated on the stack (like in Maxim's answer) is a safer way.
It is explained very well at http://www.boost.org/community/error_handling.html
So, the nicer way would be a specific type of the exception and be avoiding composing the formatted string (at least when throwing).