When should I really use noexcept?

I think it is too early to give a "best practices" answer for this as there hasn't been enough time to use it in practice. If this was asked about throw specifiers right after they came out then the answers would be very different to now.

Having to think about whether or not I need to append noexcept after every function declaration would greatly reduce programmer productivity (and frankly, would be a pain).

Well, then use it when it's obvious that the function will never throw.

When can I realistically expect to observe a performance improvement after using noexcept? [...] Personally, I care about noexcept because of the increased freedom provided to the compiler to safely apply certain kinds of optimizations.

It seems like the biggest optimization gains are from user optimizations, not compiler ones due to the possibility of checking noexcept and overloading on it. Most compilers follow a no-penalty-if-you-don't-throw exception handling method, so I doubt it would change much (or anything) on the machine code level of your code, although perhaps reduce the binary size by removing the handling code.

Using noexcept in the big four (constructors, assignment, not destructors as they're already noexcept) will likely cause the best improvements as noexcept checks are 'common' in template code such as in std containers. For instance, std::vector won't use your class's move unless it's marked noexcept (or the compiler can deduce it otherwise).

---

As I keep repeating these days: semantics first.

Adding noexcept, noexcept(true) and noexcept(false) is first and foremost about semantics. It only incidentally condition a number of possible optimizations.

As a programmer reading code, the presence of noexcept is akin to that of const: it helps me better grok what may or may not happen. Therefore, it is worthwhile spending some time thinking about whether or not you know if the function will throw. For a reminder, any kind of dynamic memory allocation may throw.

Okay, now on to the possible optimizations.

The most obvious optimizations are actually performed in the libraries. C++11 provides a number of traits that allows knowing whether a function is noexcept or not, and the Standard Library implementation themselves will use those traits to favor noexcept operations on the user-defined objects they manipulate, if possible. Such as move semantics.

The compiler may only shave a bit of fat (perhaps) from the exception handling data, because it has to take into account the fact that you may have lied. If a function marked noexcept does throw, then std::terminate is called.

These semantics were chosen for two reasons:

immediately benefiting from noexcept even when dependencies do not use it already (backward compatibility)

allowing the specification of noexcept when calling functions that may theoretically throw, but are not expected to for the given arguments

---

This actually does make a (potentially) huge difference to the optimizer in the compiler. Compilers have actually had this feature for years via the empty throw() statement after a function definition, as well as propriety extensions. I can assure you that modern compilers do take advantage of this knowledge to generate better code.

Almost every optimization in the compiler uses something called a "flow graph" of a function to reason about what is legal. A flow graph consists of what are generally called "blocks" of the function (areas of code that have a single entrance and a single exit) and edges between the blocks to indicate where flow can jump to. Noexcept alters the flow graph.

You asked for a specific example. Consider this code:

void foo(int x) {
    try {
        bar();
        x = 5;
        // Other stuff which doesn't modify x, but might throw
    } catch(...) {
        // Don't modify x
    }

    baz(x); // Or other statement using x
}

The flow graph for this function is different if bar is labeled noexcept (there is no way for execution to jump between the end of bar and the catch statement). When labeled as noexcept, the compiler is certain the value of x is 5 during the baz function - the x=5 block is said to "dominate" the baz(x) block without the edge from bar() to the catch statement.

It can then do something called "constant propagation" to generate more efficient code. Here if baz is inlined, the statements using x might also contain constants and then what used to be a runtime evaluation can be turned into a compile-time evaluation, etc.

Anyway, the short answer: noexcept lets the compiler generate a tighter flow graph, and the flow graph is used to reason about all sorts of common compiler optimizations. To a compiler, user annotations of this nature are awesome. The compiler will try to figure this stuff out, but it usually can't (the function in question might be in another object file not visible to the compiler or transitively use some function which is not visible), or when it does, there is some trivial exception which might be thrown that you're not even aware of, so it can't implicitly label it as noexcept (allocating memory might throw bad_alloc, for example).

---

noexcept can dramatically improve performance of some operations. This does not happen at the level of generating machine code by the compiler, but by selecting the most effective algorithm: as others mentioned, you do this selection using function std::move_if_noexcept. For instance, the growth of std::vector (e.g., when we call reserve) must provide a strong exception-safety guarantee. If it knows that T's move constructor doesn't throw, it can just move every element. Otherwise it must copy all Ts. This has been described in detail in this post.

---

When can I realistically except to observe a performance improvement after using noexcept? In particular, give an example of code for which a C++ compiler is able to generate better machine code after the addition of noexcept.

Um, never? Is never a time? Never.

noexcept is for compiler performance optimizations in the same way that const is for compiler performance optimizations. That is, almost never.

noexcept is primarily used to allow "you" to detect at compile-time if a function can throw an exception. Remember: most compilers don't emit special code for exceptions unless it actually throws something. So noexcept is not a matter of giving the compiler hints about how to optimize a function so much as giving you hints about how to use a function.

Templates like move_if_noexcept will detect if the move constructor is defined with noexcept and will return a const& instead of a && of the type if it is not. It's a way of saying to move if it is very safe to do so.

In general, you should use noexcept when you think it will actually be useful to do so. Some code will take different paths if is_nothrow_constructible is true for that type. If you're using code that will do that, then feel free to noexcept appropriate constructors.

In short: use it for move constructors and similar constructs, but don't feel like you have to go nuts with it.

---

In Bjarne's words (The C++ Programming Language, 4th Edition, page 366):

Where termination is an acceptable response, an uncaught exception will achieve that because it turns into a call of terminate() (§13.5.2.5). Also, a noexcept specifier (§13.5.1.1) can make that desire explicit. Successful fault-tolerant systems are multilevel. Each level copes with as many errors as it can without getting too contorted and leaves the rest to higher levels. Exceptions support that view. Furthermore, terminate() supports this view by providing an escape if the exception-handling mechanism itself is corrupted or if it has been incompletely used, thus leaving exceptions uncaught. Similarly, noexcept provides a simple escape for errors where trying to recover seems infeasible. double compute(double x) noexcept; { string s = "Courtney and Anya"; vector tmp(10); // ... } The vector constructor may fail to acquire memory for its ten doubles and throw a std::bad_alloc. In that case, the program terminates. It terminates unconditionally by invoking std::terminate() (§30.4.1.3). It does not invoke destructors from calling functions. It is implementation-defined whether destructors from scopes between the throw and the noexcept (e.g., for s in compute()) are invoked. The program is just about to terminate, so we should not depend on any object anyway. By adding a noexcept specifier, we indicate that our code was not written to cope with a throw.

---

There are many examples of functions that I know will never throw, but for which the compiler cannot determine so on its own. Should I append noexcept to the function declaration in all such cases?

noexcept is tricky, as it is part of the functions interface. Especially, if you are writing a library, your client code can depend on the noexcept property. It can be difficult to change it later, as you might break existing code. That might be less of a concern when you are implementing code that is only used by your application.

If you have a function that cannot throw, ask yourself whether it will like stay noexcept or would that restrict future implementations? For example, you might want to introduce error checking of illegal arguments by throwing exceptions (e.g., for unit tests), or you might depend on other library code that could change its exception specification. In that case, it is safer to be conservative and omit noexcept.

On the other hand, if you are confident that the function should never throw and it is correct that it is part of the specification, you should declare it noexcept. However, keep in mind that the compiler will not be able to detect violations of noexcept if your implementation changes.

For which situations should I be more careful about the use of noexcept, and for which situations can I get away with the implied noexcept(false)?

There are four classes of functions that should you should concentrate on because they will likely have the biggest impact:

move operations (move assignment operator and move constructors) swap operations memory deallocators (operator delete, operator delete[]) destructors (though these are implicitly noexcept(true) unless you make them noexcept(false))

These functions should generally be noexcept, and it is most likely that library implementations can make use of the noexcept property. For example, std::vector can use non-throwing move operations without sacrificing strong exception guarantees. Otherwise, it will have to fall back to copying elements (as it did in C++98).

This kind of optimization is on the algorithmic level and does not rely on compiler optimizations. It can have a significant impact, especially if the elements are expensive to copy.

When can I realistically expect to observe a performance improvement after using noexcept? In particular, give an example of code for which a C++ compiler is able to generate better machine code after the addition of noexcept.

The advantage of noexcept against no exception specification or throw() is that the standard allows the compilers more freedom when it comes to stack unwinding. Even in the throw() case, the compiler has to completely unwind the stack (and it has to do it in the exact reverse order of the object constructions).

In the noexcept case, on the other hand, it is not required to do that. There is no requirement that the stack has to be unwound (but the compiler is still allowed to do it). That freedom allows further code optimization as it lowers the overhead of always being able to unwind the stack.

The related question about noexcept, stack unwinding and performance goes into more details about the overhead when stack unwinding is required.

I also recommend Scott Meyers book "Effective Modern C++", "Item 14: Declare functions noexcept if they won't emit exceptions" for further reading.

---

There are many examples of functions that I know will never throw, but for which the compiler cannot determine so on its own. Should I append noexcept to the function declaration in all such cases?

When you say "I know [they] will never throw", you mean by examining the implementation of the function you know that the function will not throw. I think that approach is inside out.

It is better to consider whether a function may throw exceptions to be part of the design of the function: as important as the argument list and whether a method is a mutator (... const). Declaring that "this function never throws exceptions" is a constraint on the implementation. Omitting it does not mean the function might throw exceptions; it means that the current version of the function and all future versions may throw exceptions. It is a constraint that makes the implementation harder. But some methods must have the constraint to be practically useful; most importantly, so they can be called from destructors, but also for implementation of "roll-back" code in methods that provide the strong exception guarantee.

---

Here is a simple example to illustrate when it could really matter.

#include <iostream>
#include <vector>
using namespace std;
class A{
 public:
  A(int){cout << "A(int)" << endl;}
  A(const A&){cout << "A(const A&)" << endl;}
  A(const A&&) noexcept {cout << "A(const A&&)" << endl;}
  ~A(){cout << "~S()" << endl;}
};
int main() {
  vector<A> a;
  cout << a.capacity() << endl;
  a.emplace_back(1);
  cout << a.capacity() << endl;
  a.emplace_back(2);
  cout << a.capacity() << endl;
  return 0;
}

Here is the output

0
A(int)
1
A(int)
A(const A&&)
~S()
2
~S()
~S()

If we remove the noexcept in the move constructor, here is the output

0
A(int)
1
A(int)
A(const A&)
~S()
2
~S()
~S()

The key difference is A(const A&&) vs A(const A&&). In the second case, it has to copy all the values using the copy constructor. VERY INEFFICIENT!!

---