[intro.execution] CWG2431: Destruction of a temp. object bound to a reference is not a full-expression?

[languagelawyer]

http://eel.is/c++draft/intro.execution#def:full-expression:

A full-expression is
— an invocation of a destructor generated at the end of the lifetime of an object other than a temporary object ([class.temporary]), or

Here is an example

void foo()
{
    const auto& rk = Klass{};
    Klass k;
    // ...
    // generated invocation `k.~Klass()` is a full-expression
    // generated invocation `rk.~Klass()` is not a full-expression?
}

I think it is intended to be a full-expression, so the Standard should say something like

other than a temporary object whose lifetime was not extended

[jwakely]

If you have multiple lifetime-extended temporaries, do their destructors produce multiple full-expressions?

{
    const auto& rk = Klass{};
    const auto& rk2 = Klass{};
    // ...
}

The current wording seems to allow the destructor invocations rk.~Klass() and rk2.~Klass() to be part of the same full-expression (though it could be clearer whether that is intended, or whether your reading is correct).

[languagelawyer]

If you have multiple lifetime-extended temporaries, do their destructors produce multiple full-expressions?

If I have multiple non-reference variables, do their destruction produce multiple full-expressions?

[jwakely]

Yes, but the destruction of temporary objects doesn't have to produce multiple full-expressions, so there's no inherent requirement that destruction of objects happens in separate full-expressions.

[languagelawyer]

Ok, the difference could be intended, lets wait what others will say.

[jensmaurer]

The text is clearly wrong (a lifetime-extended temporaries should be treated as-if it were a non-reference variable), but is the fix editorial?

[zygoloid]

Given the ambiguity Jonathan mentioned, I think we should process this as a core issue.

[jensmaurer]

@zygoloid, @jwakely: What would the effective difference be between "one full-expression for destruction of several lifetime-extended temporaries" vs. "one full-expression each"? We can't have temporaries in default arguments for destructors.

[languagelawyer]

@jensmaurer the difference would be observable if a destructor throws an exception.

[jensmaurer]

@languagelawyer, sorry, maybe it's too early in the morning for me, but could you please be more specific?

[languagelawyer]

@jensmaurer if there are several temporary object destructor invocations in one full-expression and one of them throws an exception, it will finish evaluating the rest of the full-expression and could left some destructors not invoced. But if each destructor invocation is a full-expression and one of them thows, the following destructors still would be called during stack unwinding.

[jensmaurer]

@languagelawyer: Such behavior would be at odds with [stmt.jump] p2: Even if a destructor throws an exception, you still need to execute the rest of the destructors, now in the context of stack unwinding [except.ctor].

[languagelawyer]

@jensmaurer hm, indeed. Then I also don't see a difference.