c++ -std= strangeness

[Michael Dickens]

Yet another interesting oddity came up in my debugging today trying to fix CMake to build correctly on OSX 10.5 PPC(32). The issue is in the use of std::lround, which is defined for c++11 and newer only, and, was possibly not included as part of c++11 in GCC[4-6] ... it is apparently included in GCC7 and newer.

This got me going down a rabbit hole with regard to __cplusplus (set via the -std flag) and building code that calls std::lround, so I created a simple test file as follows:
{{{
#if __cplusplus == 201707L
#warning "Using STD c++2a (will be c++20)"
#elif __cplusplus == 201703L
#warning "Using STD c++17 or c++1z (will be c++17)"
#elif __cplusplus == 201402L
#warning "Using STD c++14"
#elif __cplusplus == 201103L
#warning "Using STD c++11"
#elif __cplusplus == 199711L
#warning "Using STD c++98 or c++03"
#endif

#include <cmath>
#include <iostream>

int
main
(void)
{
  std::cout << "lround(+2.3) = " << std::lround(2.3) << std::endl;
  return 0;
}
}}}

Then, I compile this code using various -std=c++XY or gnu++XY ... doesn't really matter for this example. I also use the following compilers:

* clang++ :Apple LLVM version 10.0.0 (clang-1000.11.45.5)
* clang++-mp-7.0 : clang version 7.0.1 (tags/RELEASE_701/final)
* g++-mp-8 : gcc version 8.2.0 (MacPorts gcc8 8.2.0_3)
* g++-mp-9 : gcc version 9.0.1 20190127 (experimental) (MacPorts gcc9 9-20190127_0)

The interesting strangeness is that the clang++ compilers (both Apple and MP) do not generate errors when using -std=c++98 or -std=c++03 ... which is wrong! The #warning printout indicates that the correct __cplusplus is selected, but the build should not succeed! All g++ compilers correctly error out with these -std settings.

This lack of compliance at least by default is troubling! Maybe I'm missing something here? Are there some other flags I should be setting to keep clang from figuring out the c++ standard and using it instead of what I specify -- if this is the issue?

Thanks for any feedback! - MLD