Streaming nested STL containers doesn't work

[wthrowe]

#include <array>
#include <iostream>
#include <vector>

#include "Utilities/StdHelpers.hpp"

void f() {
  // std::vector<int> x{};  // fine                                             
  // std::array<int, 1> x{};  // fine                                           
  std::vector<std::array<int, 1>> x{};  // error
  std::cout << x;
}

/home/wthrowe/spectre/src/Utilities/PrintHelpers.hpp:42:63: error: call to function 'operator<<' that is neither visible in the template definition nor found by argument-dependent lookup
      [](std::ostream& os, const ForwardIt& it) noexcept { os << *it; });
                                                              ^
/home/wthrowe/spectre/src/Utilities/StdHelpers.hpp:81:3: note: in instantiation of function template specialization 'sequence_print_helper<__gnu_cxx::__normal_iterator<const std::array<int, 1> *, std::vector<std::array<int, 1>, std::allocator<std::array<int, 1> > > > >' requested here
  sequence_print_helper(os, std::begin(v), std::end(v));
  ^

[nilsdeppe]

I think the fix is adding a using ::operator<<; inside those lambdas that do the streaming. Would you be willing to try that?

[wthrowe]

I get "error: no member named 'operator<<' in the global namespace".

I've noticed that the code is sensitive to the order of includes. In my original code snippet, including "Utilities/StlStreamDeclarations.hpp" before "Utilities/StdHelpers.hpp" makes everything work, but including it after does not help.

[nilsdeppe]

Sorry for the delay... Yea, I can see why this would be include order dependent... I don't really have a great idea other than the super sketchy (and technically UB) of moving the operator<< for STL containers into the std namespace. I'll need to think more about it. Maybe we can convince ADL to delay lookup long enough that all includes have been parsed?

[wthrowe]

It seems like this can mostly be worked around using MakeString (and probably get_output). Catch provides a way to override its "{?}" output using a CATCH_CONFIG_FALLBACK_STRINGIFIER, but I haven't found a way to make it work yet.

[nilsdeppe]

That sounds like a much better approach :) It's nice that catch has a stringifier (as long is it works :P )