_GLIBCXX_DEBUG causing runtime failure when --order rand

[thomasfanell]

Description

If _GLIBCXX_DEBUG is defined during compilation, the following runtime error occur when the test executable is invoked with --order rand:

/usr/include/c++/7/debug/safe_container.h:83:
Error: attempt to self move assign.

Objects involved in the operation:
    sequence "this" @ 0x0x55d377712c78 {
      type = __gnu_debug::_Safe_container<std::__debug::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, __gnu_debug::_Safe_sequence, true>;
    }
Aborted (core dumped)

Steps to reproduce

Code

// main.cpp
#define CATCH_CONFIG_RUNNER
#include "catch.hpp"
TEST_CASE("A", "[bug]") { SECTION("a") { REQUIRE(true); } }
TEST_CASE("B", "[bug]") { SECTION("b") { REQUIRE(true); } }
TEST_CASE("C", "[bug]") { SECTION("c") { REQUIRE(true); } }
TEST_CASE("D", "[bug]") { SECTION("d") { REQUIRE(true); } }
TEST_CASE("E", "[bug]") { SECTION("e") { REQUIRE(true); } }
TEST_CASE("F", "[bug]") { SECTION("f") { REQUIRE(true); } }
int main(int argc, char* argv[])
{
    return Catch::Session().run(argc, argv);
}

Compile and run

• g++-7 -std=c++14 -g -D_GLIBCXX_DEBUG main.cpp && ./a.out --rng-seed time --order rand
• clang++-5.0 -std=c++14 -g -D_GLIBCXX_DEBUG main.cpp && ./a.out --rng-seed time --order rand

Extra information

• Catch version: v2.4.0 (v2.2.2 fails too)
• Operating System: Ubuntu 18.04 LTS
• Compiler+version: g++ 7.3.0, clang 5.0.1

[horenmar]

As far as I can tell, this is a bug in libstdc++. Depending on your PoV, it is either bug in the debug checker, or in its implementation of std::shuffle. Either way, I consider this a 3rd party bug that is not worth a workaround.

[thomasfanell]

It seems you're right. I must admit I managed to completely miss that SO post. I suppose we'll have to redirect __glibcxx_check_self_move_assign for now. Thanks for the link and my apologies for wasting your time.

[horenmar]

Not a problem.