Re: [Patch] Implement std::experimental::variant

[Tim Shen <timshen at google dot com>]

On Mon, May 16, 2016 at 6:02 AM, Jonathan Wakely wrote:
> Those constructors should be constexpr because in each case the
> alternative being constructed is 'literal' which has a constexpr
> constructor. I believe this is the hardest part of 'variant' to
> implement, so it would be good to know if your current approach is
> fundamentally incompatible with this requirement before it's too late
> to change it.

Good catch :). I believe that my current approach is fundamentally
incompatible with this requirement.

I current have two approaches to implement variant:
1) recursive union
2) placement-newed raw buffer

For 1), the following code example fails to compile:
  struct literal {
      constexpr literal() = default;
  };

  struct nonliteral {
      nonliteral() { }
  };

  union A {
      constexpr A() : a{} {}

      literal a;
      nonliteral b;
  };

  constexpr A a{};

The potential solution is to change the standard definition of
constexpr to allow the above case.


For 2), constexpr constructor body cannot have placement new. I found
some discussion here
<https://groups.google.com/a/isocpp.org/forum/#!topic/std-discussion/_s7vi9pOhfY>,
but it seems to be complicated to implement & get standardized.

Do you have any other apporaches to make this work? Notice that the
"variadic multiple inheritance" appraoch doesn't work here, because it
aggregates data, instead of creating a data union.


The third option is to change the variant proposal to require all
alternatives to be literal type in order to have a literal variant
type. It's a safe and forward-compatible move in standardization's
perspective.

I'm not sure if in the future "constexpr" variable may not imply const
anymore, for example:
  constexpr int a = 0;  // not constexpr const int a = 0;
  constexpr int iota() { return a++; }

If this may be the future, then a literal variant variant should
definitely hold literal alternatives only (so that it supports
mutation at compile-time for changing to all alternatives).


-- 
Regards,
Tim Shen