Leaf sections sometimes skipped with Clang+libc++ on Linux

[u3shit]

Description

When compiling with clang+libc++, and an exception was thrown before (even in a different test-case), Catch will sometime execute a test case where all child sections are skipped. The CHECK in the example below should succeed in all cases, yet it fails with clang and libc++ (with expansion 0 > 0). Compiling with clang and gcc's libstdc++, with gcc, or with clang-cl it works fine, so either libc++ implements something differently and catch doesn't like it, or it's simply buggy (previously I've found a bug in libcxxrt's exception handling code while debugging a luajit related issue, so it's possible...).

Steps to reproduce

Here's a minimal example that triggers the problem.

#define CATCH_CONFIG_MAIN
#include <catch.hpp>

TEST_CASE("a")
{
    CHECK_THROWS(throw 3);
}

TEST_CASE("b")
{
    int i = 0;
    SECTION("a") { i = 1; }
    SECTION("b") { i = 2; }
    CHECK(i > 0);
}

Compile it with clang++ -stdlib=libc++. Optimization levels or -std=c++11, -std=c++14 don't matter. If I only run test case b, it works correctly.

Extra information

• Catch version: v1.7.0
• Operating System: Gentoo
• Compiler+version: clang 3.9.1, libcxxrt from git master

[u3shit]

Okay, I'm pretty sure libcxxrt is at fault here: https://github.com/pathscale/libcxxrt/issues/45
Basically std::uncaught_exception() returns true after a rethrow, even when there's no active exceptions.
Closing the issue here.

[horenmar]

We have had similar thing going on with libstdc++ in the past (#352, #271 and possibly others), this is the first time with libc++.

Thank you for the followup and raising issue against libcxxrt.