81

When creating a new instance of a MyClass as an argument to a function like so:

class MyClass
{
  MyClass(int a);
};    

myFunction(MyClass(42));

Does the standard make any guarantees on the timing of the destructor?

Specifically, can I assume that it is going to be called before the next statement after the call to myFunction() ?

1

4 Answers 4

125

Temporary objects are destroyed at the end of the full expression they're part of.

A full expression is an expression that isn't a sub-expression of some other expression. Usually this means it ends at the ; (or ) for if, while, switch etc.) denoting the end of the statement. In your example, it's the end of the function call.

Note that you can extend the lifetime of temporaries by binding them to a const reference. Doing so extends their lifetime to the reference's lifetime:

MyClass getMyClass();

{
  const MyClass& r = getMyClass(); // full expression ends here
  ...
} // object returned by getMyClass() is destroyed here

If you don't plan to change the returned object, then this is a nice trick to save a copy constructor call (compared to MyClass obj = getMyClass();), in case return value optimization was not being applied. Unfortunately it isn't very well known. (I suppose C++11's move semantics will render it less useful, though.)

Sign up to request clarification or add additional context in comments.

8 Comments

+1 nice answer. Tho i think you mean ) instead of } for if, while, switch etc? if(f(string())) { ... } surely won't make the string survive until } :)
You said it yourself correctly: "A full expression is an expression that isn't a sub-expression of some other expression." and in the while, if, switch cases we have a full expression between the ( and ) because the statement they appear in is not an expression by any means. In your interpretation, the end of ; would not be a full-expression-end at all, because there is a larger "expression" around it if you are within a while/if or switch. I think it's easy to see that this makes no sense :)
@Johannes: I changed it. <sigh> After so many years of C++ you still learn something every week.
I didn't realize you could extend the life of a temporary object using a const reference. Good to know!
In C++11, an rvalue reference also extends the lifetime of a temporary. However, I would not worry about using either of these for performance reasons. I would expect any optimizing compiler to apply copy / move elision here, making the two versions compile to identical code (capture by value or bind to reference).
|
29

Everyone has rightly cited 12.2/3 or similar, which answers your question:

Temporary objects are destroyed as the last step in evaluating the full-expression that (lexically) contains the point where they were created.

I find it amusing that over the next page in my printing of the standard, 12.2/4 says:

There are two contexts in which temporaries are destroyed at a different point than the end of the full-expression.

Neither of them applies to your example, they both relate to the use of temporaries in initializers. But it does go to show that you have to keep your wits about you when dealing with a tricky beast like the C++ standard.

2 Comments

It would be awesome if that second quote was intentionally aligned such that it starts on one page and ends on the next. :-)
Wow, thanks for mentioning this "exception". Temporaries in initializers was precisely the case I was bitten by.
10

The standard does indeed offer guarantees - from section 12.2/5:

A temporary bound to a reference parameter in a function call (5.2.2) persists until the completion of the full expression containing the call

However, in your code it is not clear if the parameter is being passed by reference or by value, though at some point a copy constructor which does take a reference will be used.

1 Comment

Any sections about whether the calling function or the called function calls the destructor of the by-value parameter? Checking disassembly of VS2013 (clang or gcc unknown), the called function calls the destructor before returning. It is understandable for stdcall because the callee would pop the parameters before returning, but it surprised me for cdecl (understandable to avoid code though).
2

In section 12.2, Temporary Objects, clause 3, the ANSI/ISO C standard states: "... Temporary objects are destroyed as the last step in evaluating the full-expression that (lexically) contains the point where they were created."

This is closely related to the concept of Sequence Points. When a sequence point is reached, all side-effects of expressions are guaranteed to have taken place.

1 Comment

There was (at the time of this answer) a sequence point at the end of each full-expression, but I don't see how that implies "closely related". Yes, the destruction of temporary objects could have side-effects, so there is a relation to the regulation of side-effects (a.k.a. "sequence points"), but that's not a particularly close relation. At least not in the sense that understanding one helps you understand the other.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.