Full name of submitter (unless configured in github; will be published with the issue): Richard Smith

Reference (section label): [basic.life]/1.5

Link to reflector thread (if any):

Issue description:

[basic.life]/1.5 says:

The lifetime of an object o of type T ends when [...] the storage which the object occupies [...] is reused [...]

but what does that mean? In llvm/llvm-project#61562 we have two possible interpretations:

1. In an expression new (p) T(x), the lifetime of *p ends when p is returned from void* operator new(size_t, void*), before x is evaluated.
2. In an expression new (p) T(x), the lifetime of *p ends the moment the constructor body starts executing, after x is evaluated.

Note that interpretation (2) only works when the initialization is performed by a constructor or if T is a non-class type, and we must use interpretation (1) for aggregate initialization because the evaluation of x can directly initialize a part of the resulting object.

Suggested resolution:

To me, it seems that the most consistent and simplest model is that the lifetime of *p ends the moment that p is returned from an operator new function invoked by a new-expression, and in particular, the object no longer exists in the evaluation of the new-initializer.

Alternatively, we could say that the the lifetime ends as late as possible -- the moment anything actually starts initializing the new object (perhaps, when a constructor starts running or a value is stored for the object or any of its subobjects). That's probably going to be challenging for constant interpreters to model, though.