Explicitly use nest namespace to avoid compiler confusion

[hakonsbm]

When compiling with -std=c++17, the compiler confuses the apply() in nest.h with std::apply() from <tuple>. This happens with both GCC and Clang. However, there doesn't seem to be any place where we say that the std namespace should be used in that context. The calls are therefore changed to explicitly use the nest namespace, which fixes the problem and therefore fixes #2252.

As a side-note, it's possible to reproduce the behaviour without NEST, with this reproducer.

#include <tuple>
#include <vector>

namespace foo
{

void
bar()
{
}

std::vector< double >
apply( const std::vector< double >&, const std::vector< double >& )
{
  return {};
}

void
run()
{
  std::vector< double > vec;

  bar();                  // OK
  apply( vec, vec );      // Tries to use std::apply() from <tuple>, which causes an error with the arguments
  foo::apply( vec, vec ); // OK
}
} // end of namespace foo

int
main()
{
  foo::run();
}

[heplesser]

@hakonsbm Thanks for the quick fix! But some numerics seem to have changed (last bit of float), we probably should change the test to use something like x == pytest.approx(x_expected).

[ikitayama]

@hakonsbm if you like, I am happy to bring your reproducer to the LLVM community's attention. As I am using the main branch of LLVM.

[hakonsbm]

@heplesser I've updated the test, and all tests pass now.

@ikitayama Sure, go ahead.

[heplesser]

@hakonsbm if you like, I am happy to bring your reproducer to the LLVM community's attention. As I am using the main branch of LLVM.

Since the problem also occurs with gcc 11 when compiling with -std=c++17, could you also open a ticket on https://gcc.gnu.org/bugzilla?

[stinebuu]

Looks good to me! 👍