https://issues.dlang.org/show_bug.cgi?id=19430

--- Comment #7 from Stanislav Blinov <stanislav.blinov@gmail.com> ---
What's going on is (pseudo-code) this:

c = S(S.init).opAssign(S.sum(1, 2));

The opAssign is implicitly generated.

Two instances are constructed in-place, one is 'main.s', the other is the
argument to opAssign, no copies are made, no postblits are called.
But then the argument of opAssign is getting destructed, and decrements count
to 0.

So 0 is the correct output here.

For it to be 1, you *need* an explicit opAssign:

void opAssign(S rhs)
{
    if (ptr && *ptr)
        atomicOp!"+="(*ptr, 1);
}

If it were a C++ std::shared_ptr kind of reference count (i.e. not a static counter), then you'd need this kind of opAssign:

void opAssign(S rhs)
{
   import std.algorithm.mutation : swap;
   swap(ptr, rhs.ptr);
}

--

https://issues.dlang.org/show_bug.cgi?id=19430

--- Comment #8 from Stanislav Blinov <stanislav.blinov@gmail.com> ---
Working example:

https://run.dlang.io/is/dnxQNx

--

https://issues.dlang.org/show_bug.cgi?id=19430

Илья Ярошенко <ilyayaroshenko@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |RESOLVED
         Resolution|---                         |LATER

--- Comment #9 from Илья Ярошенко <ilyayaroshenko@gmail.com> ---
Ok, requires more research

--

https://issues.dlang.org/show_bug.cgi?id=19430

--- Comment #10 from Stanislav Blinov <stanislav.blinov@gmail.com> ---
To be fair, calling assignment on an instance that hasn't been yet constructed (i.e. constructor didn't return) isn't the best of ideas. I know Phobos is doing this in places, but it's something that should really, really be avoided.

--

https://issues.dlang.org/show_bug.cgi?id=19430

Илья Ярошенко <ilyayaroshenko@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|LATER                       |INVALID

--

https://issues.dlang.org/show_bug.cgi?id=19430

--- Comment #11 from Илья Ярошенко <ilyayaroshenko@gmail.com> ---
I am starting to dig deeper since I have more code that uses RC with D.
The first one issue is https://issues.dlang.org/show_bug.cgi?id=19774
The second one related to generic `=` operator, still needs to be reduced.

--