sigslot::observer issue when using a pointer

[Sebastian Redinger]

Hi. First of all thanks for this great library.

I think I have found a issue while using sigslot::observer. (Reproducable in MSVC2015 and MSVC2019)

This is my example code:

struct A : sigslot::observer {
    void my_slot() { ; }
};

sigslot::signal<> my_signal;

// Example 1
{
    A a;
    my_signal.connect(&A::my_slot, &a); // <-- Call's overload with observer
}
// Works as expected: my_slot is not called
my_signal();

// Example 2
A* aptr2 = new A;
A& aptr2derefed = *aptr2;
my_signal.connect(&A::my_slot, &aptr2derefed); // <-- Call's overload with observer
delete aptr2;
// Works as expected: my_slot is not called
my_signal();

// Example 3
A* aptr = new A;
my_signal.connect(&A::my_slot, aptr); // <-- Call's overload without observer
delete aptr;

// my_slot gets called even though aptr is already deleted
my_signal();

Example 1 and 2 work as I expect and my_slot does not get called in example 3 however a differnet overload
of connect gets called, where !trait::is_observer_v<Ptr> is checked in std::enable_if. This is an issue because
after aptr is deleted and my_signal gets emited my_slot gets called on a dangling pointer.
It could be a misuse case on my side but I think it might be an error in is_observer_v.
When you add a std::remove_reference_t to is_observer_v the issue seems to be fixed but I'm not sure if this causes any
unwanted side effects.

This is my proposed fix for this issue:

template <typename T>
constexpr bool is_observer_v = std::is_base_of<::sigslot::detail::observer_type,
    std::remove_pointer_t<std::remove_reference_t<T>>>::value;

[Pierre-Antoine Lacaze]

Fixed as per your suggestion, thanks!