The std::abort() , std::quick_exit() , and std::_Exit() functions are used to terminate the program in an immediate fashion. They do so without calling exit handlers registered with std::atexit() and without executing destructors for objects with automatic, thread, or static storage duration. How a system manages open streams when a program ends is implementation-defined [ ISO/IEC 9899:1999 ]. Open streams with unwritten buffered data may or may not be flushed, open streams may or may not be closed, and temporary files may or may not be removed. Because these functions can leave external resources, such as files and network communications, in an indeterminate state, they should be called explicitly only in direct response to a critical error in the application. (See ERR50-CPP-EX1 for more information.)

The std::terminate() function calls the current terminate_handler function, which defaults to calling std::abort() .

The C++ Standard defines several ways in which std::terminate() may be called implicitly by an implementation [ ISO/IEC 14882-2014 ]:

1. When the exception handling mechanism, after completing the initialization of the exception object but before activation of a handler for the exception, calls a function that exits via an exception ([except.throw], paragraph 7)
   
   • See ERR60-CPP. Exception objects must be nothrow copy constructible for more information.
2. When a throw-expression with no operand attempts to rethrow an exception and no exception is being handled ([except.throw], paragraph 9)
3. When the exception handling mechanism cannot find a handler for a thrown exception ([except.handle], paragraph 9)
   • See ERR51-CPP. Handle all exceptions for more information.
4. When the search for a handler encounters the outermost block of a function with a noexcept-specification that does not allow the exception ([except.spec], paragraph 9)
   • See ERR55-CPP. Honor exception specifications for more information.
5. When the destruction of an object during stack unwinding terminates by throwing an exception ([except.ctor], paragraph 3)
   • See DCL57-CPP. Do not let exceptions escape from destructors or deallocation functions for more information.
6. When initialization of a nonlocal variable with static or thread storage duration exits via an exception ([basic.start.init], paragraph 6)
   
   • See ERR58-CPP. Handle all exceptions thrown before main() begins executing for more information.
7. When destruction of an object with static or thread storage duration exits via an exception ([basic.start.term], paragraph 1)
   • See DCL57-CPP. Do not let exceptions escape from destructors or deallocation functions for more information.
8. When execution of a function registered with std::atexit() or std::at_quick_exit() exits via an exception ([support.start.term], paragraphs 8 and 12)
9. When the implementation’s default unexpected exception handler is called ([except.unexpected], paragraph 2)
   Note that std::unexpected() is currently deprecated.
10. When std::unexpected() throws an exception that is not allowed by the previously violated dynamic-exception-specification , and std::bad_exception() is not included in that dynamic-exception-specification ([except.unexpected], paragraph 3)
    
11. When the function std::nested_exception::rethrow_nested() is called for an object that has captured no exception ([except.nested], paragraph 4)
    
12. When execution of the initial function of a thread exits via an exception ([thread.thread.constr], paragraph 5)
    • See ERR51-CPP. Handle all exceptions for more information.
13. When the destructor is invoked on an object of type std::thread that refers to a joinable thread ([thread.thread.destr], paragraph 1)
14. When the copy assignment operator is invoked on an object of type std::thread that refers to a joinable thread ([thread.thread.assign], paragraph 1)
    
15. When calling condition_variable::wait() , condition_variable::wait_until() , or condition_variable::wait_for() results in a failure to meet the postcondition: lock.owns_lock() == true or lock.mutex() is not locked by the calling thread ([thread.condition.condvar], paragraphs 11, 16, 21, 28, 33, and 40)
16. When calling condition_variable_any::wait() , condition_variable_any::wait_until() , or condition_variable_any::wait_for() results in a failure to meet the postcondition: lock is not locked by the calling thread ([thread.condition.condvarany], paragraphs 11, 16, and 22)

In many circumstances, the call stack will not be unwound in response to an implicit call to std::terminate() , and in a few cases, it is implementation-defined whether or not stack unwinding will occur. The C++ Standard, [except.terminate], paragraph 2 [ ISO/IEC 14882-2014 ], in part, states the following:

In the situation where no matching handler is found, it is implementation-defined whether or not the stack is unwound before std::terminate() is called. In the situation where the search for a handler encounters the outermost block of a function with a noexcept-specification that does not allow the exception, it is implementation-defined whether the stack is unwound, unwound partially, or not unwound at all before std::terminate() is called. In all other situations, the stack shall not be unwound before std::terminate() is called.

Do not explicitly or implicitly call std::quick_exit() , std::abort() , or std::_Exit() . When the default terminate_handler is installed or the current terminate_handler responds by calling std::abort() or std::_Exit() , do not explicitly or implicitly call std::terminate() . Abnormal process termination is the typical vector for denial-of-service attacks .

The std::exit() function is more complex. The C++ Standard, [basic.start.main], paragraph 4, states:

Terminating the program without leaving the current block (e.g., by calling the function std::exit(int) (17.5)) does not destroy any objects with automatic storage duration (11.4.6). If std::exit is called to end a program during the destruction of an object with static or thread storage duration, the program has undefined behavior.

You may call std::exit() only in a program that has not yet initialized any objects with automatic storage duration.

Noncompliant Code Example

In this noncompliant code example, the call to f() , which was registered as an exit handler with std::at_exit() , may result in a call to std::terminate() because throwing_func() may throw an exception.

#include <cstdlib>
void throwing_func() noexcept(false);
void f() { // Not invoked by the program except as an exit handler.
  throwing_func();
int main() {
  if (0 != std::atexit(f)) {
    // Handle error
  // ...

Compliant Solution

In this compliant solution, f() handles all exceptions thrown by throwing_func() and does not rethrow.

#include <cstdlib>
void throwing_func() noexcept(false);
void f() { // Not invoked by the program except as an exit handler.
  try {
    throwing_func();
  } catch (...) {
    // Handle error
int main() {
  if (0 != std::atexit(f)) {
    // Handle error
  // ...

Exceptions

ERR50-CPP-EX1: It is acceptable, after indicating the nature of the problem to the operator, to explicitly call std::abort() , std::_Exit() , or std::terminate() in response to a critical program error for which no recovery is possible, as in this example.

#include <exception>
void report(const char *msg) noexcept;
[[noreturn]] void fast_fail(const char *msg) {
  // Report error message to operator
  report(msg);
  // Terminate
  std::terminate();
void critical_function_that_fails() noexcept(false);
void f() {
  try {
    critical_function_that_fails();
  } catch (...) {
    fast_fail("Critical function failure");

The assert() macro is permissible under this exception because failed assertions will notify the operator on the standard error stream in an implementation-defined manner before calling std::abort() .

Risk Assessment

Allowing the application to abnormally terminate can lead to resources not being freed, closed, and so on. It is frequently a vector for denial-of-service attacks .

Automated Detection

Related Vulnerabilities

Search for other vulnerabilities resulting from the violation of this rule on the CERT website .

Related Guidelines

Bibliography

First, fork() is POSIX not ISO C++, so it is out-of-scope of this rule :) And this fact does interfere with satisfactorily answering your question.

Depending on your platforms fork() implementation, it may indeed be safe to terminate a forked child process that is executing the same code, but has no destructors to invoke. But I can't say for sure without official POSIX diocumentation about how fork() works with C++ destructors.

That is weird...I would expect that fork() should have dropped you off in the switch statement, and not re-executed the "Before fork" statement.

I thought that calling fork() inside switch might explain the behavior, but it persists even if we hoist the fork() outside the switch. This problem also exists in C. And gcc vs clang doesn't matter.

POSIX IEEE Std 1003.1-2017 says wrt fork():
Both processes shall continue to execute from the fork() function.

I figured out what is happening. Your C++ example works as expected if you flush the output:

...
std::cout << "Before fork" ;
std::cout.flush();
switch(fork()) {
...

My skim of fork(3) suggests that unflushed buffered data can get duplicated.

Sigh. This is what happens when I try a coding puzzle at 4pm Friday :/

This is indeed what's going on, but your workaround kind of misses the point. Just like it's important to not skip cleanup/flushes, it's also important to not run them early/twice, and there's not really a feasible way to ensure the latter in a forked child other than _Exit , since in a real program, you don't know every single buffer there is to flush ahead of time, and in some cases, even if you did there's nothing you can do about it.

Consider this scenario: a constructor creates a temporary file, and the corresponding destructor deletes it. The constructor runs, the program forks, and the destructor runs in the child. Now the file is gone, but the parent still needs it.

That's the problem with fork()...it shares everything whether the child process needs it or not.

I think there should be a POSIX recommendation or rule saying "do not let unbuffered data persist when forking a child process, if both parent and child subsequently perform read or write on that file descriptor", as the results are almost never what the developer wanted. That rule would address your example of a temporary file being created in a parent process and destroyed in a child process.

Such a recommendation/rule wouldn't be practical to follow in non-toy programs, since libraries often do that and don't give you a way to release anything without breaking what they're doing. (Side note: I'm also concerned that this rule itself is getting impractical to follow, now that for most programs it basically means "never call exit ".)

WRT my fork recommendation, it would be a rule in the C POSIX standard. There is no C++/POSIX standard and we don't have the funds to create a good one. I personally have no idea how destructors and fork() interact...if a fork() call means destructors get called both in the parent & child threads, then you probably should never call fork() in a C++ program.

This rule is trying to comply with RAII...to make sure all destructors are called on normal termination (at the end of main(). That is, if a dtor was not invoked then by definition it was not a 'normal' exit :). Perhaps this rule now needs an exception saying "it is ok to call abort() or terminate() if your program detects a security flaw such that worse things than a crash can result if the program does not terminate immediately".

I like this article by Raymond Chen: https://devblogs.microsoft.com/oldnewthing/20120105-00/?p=8683

The building is being demolished. Don’t bother sweeping the floor and emptying the trash cans and erasing the whiteboards. And don’t line up at the exit to the building so everybody can move their in/out magnet to out. All you’re doing is making the demolition team wait for you to finish these pointless housecleaning tasks. Okay, if you have internal file buffers, you can write them out to the file handle. That’s like remembering to take the last pieces of mail from the mailroom out to the mailbox. But don’t bother closing the handle or freeing the buffer, in the same way you shouldn’t bother updating the “mail last picked up on” sign or resetting the flags on all the mailboxes.

IMO, if something that a destructor does matters even if the program is exiting (like flushing an output buffer), then the program/library should register an atexit handler to make sure it gets done, but for most destructors, they're just deallocating memory that's about to be deallocated anyway, so I don't see harm in skipping them.

FWIW, when working on the C++ guidelines, I explicitly elected to not cover POSIX. POSIX is a superset of the C standard and is designed such that a conflict between it and C means that C wins. But there's not as clear of a relationship to the C++ standard and we didn't have the resources to consider how POSIX and C++ interact because there's a whole lot of surface area to cover there (things like std::filesystem, std::thread, etc). I'd be wary of adding a POSIX section to the CERT C++ standard without loud warnings about its completeness.

I guess I see that as two sides of the same coin. We didn't consider POSIX at all, so the rules are likely harder to use in conjunction with POSIX functionality and there are no rules specifically about POSIX.

FWIW, I don't think that's as bad of a situation as it would be in C – C++ has a lot of improved replacement facilities for POSIX functionality and so it being hard to use some parts of POSIX may not be a bad thing (like pthreads and file system stuff).

These sentences seem misleading:

The std::abort() , std::quick_exit() , and std::_Exit() functions are used to terminate the program in an immediate fashion. They do so without calling exit handlers registered with std::atexit() and without executing destructors for objects with automatic, thread, or static storage duration.

It is acceptable to call a termination function that safely executes destructors and properly cleans up resources, such as std::exit() .

The problem is that they make it sound like std::exit() calls destructors for objects with automatic storage duration, but it actually doesn't:

#include <iostream>
struct foo {
    std::string s;
    foo(std::string s) : s(s) {
        std::cout << "Constructing " << s << std::endl;
    ~foo() {
        std::cout << "Destructing " << s << std::endl;
foo static_foo{"static_foo"};
int main() {
    foo automatic_foo{"automatic_foo"};
    std::exit(0);

Result:

Constructing static_foo
Constructing automatic_foo
Destructing static_foo

But it doesn't just skip the current function's variables: it skips all of them. This program doesn't call automatic_foo 's destructor either, even though your new wording is fine with it:

#include <iostream>
struct foo {
    std::string s;
    foo(std::string s) : s(s) {
        std::cout << "Constructing " << s << std::endl;
    ~foo() {
        std::cout << "Destructing " << s << std::endl;
foo static_foo{"static_foo"};
void f() {
    // this function has not yet initialized any objects with automatic storage duration
    std::exit(0);
int main() {
    foo automatic_foo{"automatic_foo"};
            
Powered by Atlassian Confluence 8.5.14
            
Printed by Atlassian Confluence 8.5.14
            
Report a bug
            
Atlassian News