Butterfly barrier hangs when `arrive_and_drop`ing multiple times from the same thread.

[brycelelbach]

#include <barrier>
#include <iostream>

int main() {
  constexpr std::size_t n = 2;

  std::barrier b(n, [] { std::cout << "phase completed\n"; });

  for (std::size_t i = 0; i < n; ++i)
    b.arrive_and_drop();

  std::cout << "finished arriving and dropping\n";
}

If I compile with -D__BARRIER_NO_BUTTERFLY, it works fine.

This is on the top of trunk (a8ea4b4).

I ran into this in my demo with multiple threads calling arrive_and_drop multiple times each.

[ogiroux]

So that's UB. Not a bug.