https://issues.dlang.org/show_bug.cgi?id=20117

Simen Kjaeraas <simen.kjaras@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |simen.kjaras@gmail.com

--- Comment #1 from Simen Kjaeraas <simen.kjaras@gmail.com> ---
This would perhaps more correctly be filed as 'int does not have opCmp'. Since int does support the <, <=, >=, > operators it kinda, sorta has opCmp, but no member by that name. That's not really an option for Typedef, as it's not a magic type like int. At the same time, we absolutely expect MyInt.init < MyInt.init to compile.

There may be a good argument for adding opCmp and friends to built-in types to make generic code more generic, but removing them from Typedef is no solution.

--

https://issues.dlang.org/show_bug.cgi?id=20117

--- Comment #2 from Simen Kjaeraas <simen.kjaras@gmail.com> ---
Alternatively, Typedef should contain only a single member and use aliases for the rest. Something like this:

struct Typedef(T, T init = T.init, string cookie=null) {
    // Renaming the current Typedef:
    TypedefImpl!(T, init, cookie) _payload;
    alias _payload this;
    static foreach (e; __traits(allMembers, T))
        mixin("alias "~e~" = _payload."~e~";");
}

--

https://issues.dlang.org/show_bug.cgi?id=20117

--- Comment #3 from Atila Neves <atila.neves@gmail.com> ---
But `int` *doesn't* have `opCmp`. I found this by trying to wrap a generic type for Python and implementing the comparison. std.typecons.Typedef completely broke things and I have to special-case for it, because `hasMember!(Typedef, "opCmp")` returns true, but then one can't actually forward to it because it doesn't actually exist.

Incidentally, the reason the bug happens is because `Typedef` uses `Proxy` in its implementation, and `Proxy` defines `opCmp`.

--

https://issues.dlang.org/show_bug.cgi?id=20117

--- Comment #4 from Simen Kjaeraas <simen.kjaras@gmail.com> ---
But `int` *is* comparable using <, <=, >=, and >, so any (non-built-in) type pretending to be an int needs to have `opCmp`. What you're asking for is basically to cripple `Typedef` because you need to write some workaround code. Alternatively, you're asking for what I said in comment 2, but have elected to ignore that.

Actually, even with my `alias this` solution in comment 2, you get two members: `_payload`, and `opAssign`. The former we've defined, the latter the compiler makes for us. `int` does not have `opAssign`, so we can't forward to that, thus again causing possible issues. In fact, `int` doesn't have any members at all, if you ask `__traits(allMembers)`. Thus, any member whatsoever in `Typedef!int` could be problematic in generic code. I hope we can agree that `Typedef!int` needs to have at least one member.

Actually, we can get rid of all fields by (ab)using `align` and some creative
casting:

template Typedef(T) {
    align ((T.sizeof + 1) & -2) // powers of 2 only
    struct Typedef {
        ref auto _payload() {
            import std.typecons : TypedefImpl = Typedef;
            return *cast(TypedefImpl!T*)&this;
        }
        alias _payload this;

        // alias all members of the wrapped type
        static if (__traits(compiles, __traits(allMembers, T)))
            static foreach (e; __traits(allMembers, T))
                mixin("alias _payload."~e~" "~e~";");
    }
}

This still leaves us with a single method (`_payload()`) that I haven't found a
way to get rid of yet. Also, it's hella ugly.

--

https://issues.dlang.org/show_bug.cgi?id=20117

berni44 <bugzilla@d-ecke.de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bugzilla@d-ecke.de
           Hardware|x86_64                      |All
                 OS|Linux                       |All

--

https://issues.dlang.org/show_bug.cgi?id=20117

Iain Buclaw <ibuclaw@gdcproject.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P1                          |P3

--