Adding support for NULL pointers for several shared pointer methods

[P-p-H-d]

No description provided.

[P-p-H-d]

Question: about the I/O methods. Support could be added but it has to handle it in a way that doesn't interfere with the printing of the internal object, which may be hard.

[BrianSipos]

That support is not important (to me). Documenting preconditions and behaviors with null pointers on the main interfaces is the important thing to me. Allowing null pointer items in container "shared ptr within array/deque/stack/buffer" is helpful to have.

[P-p-H-d]

ThreadSanitizer has detected a lock-order-inversion (potential deadlock) when using _apply or _for_each if the used callback performs some mutex lock (directly or indirectly by calling another shared ptr function).

It cannot be fixed easily.

Probably we need to :

• specify the problem and the exact restriction in theses functions in the specification, without being too restrictive
• add a new set of functions that will avoid this restriction (but perform much slower and consume more memory)

EDIT: Currently we will only specify the problem and the restriction. If needed by some uses, we can implement a new set of functions later.