Boost C++ Libraries: Ticket #4193: Multiple 'using' declarations in multi_index test

attachment set

Chris Jefferson — Wed, 05 May 2010 15:29:45 GMT

attachment → test_modifiers.patch

Simple patch to libs/multi_index/test/test_modifiers.cpp

component changed; owner set

Steven Watanabe — Thu, 06 May 2010 02:40:28 GMT

owner set to Joaquín M López Muñoz
component None → multi_index

Joaquín M López Muñoz — Thu, 06 May 2010 06:10:05 GMT

Hi Chris,

Can you please explain to me why having redundant using declarations is incorrect? I've just tested it in Comeau online and found no problems. For your reference, the test snippet was:

#include <algorithm>
int main()
{
  int x=0,y=0;
  using std::swap;
  swap(x,y);
  using std::swap;
  swap(x,y);
}

anonymous — Thu, 06 May 2010 07:29:09 GMT

Sorry, the reduced test I was using at comeau was not testing what I thought it was.

It is still the case that this code does not compile in clang, because of the repeated std::swaps, but it may be a fault in clang, which could be fixed rather than introducing an #ifdef or patch.

Will report back after further investigation

anonymous — Thu, 06 May 2010 14:23:12 GMT

From Doug Gregor, it appears that clang is right to reject the current code:

Clang is correct; Comeau and g++ are wrong not to diagnose these errors. See C++ [namespace.udecl]p10:

A using-declaration is a declaration and can therefore be used repeatedly

where (and only where) multiple declarations are allowed.

Steven Watanabe — Thu, 06 May 2010 15:07:11 GMT

Huh?

The standard reference seems to imply that

#include <algorithm>
int main() {
    using std::swap;
    using std::swap;
}

is legal iff

int main() {
    void swap(int&, int&);
    void swap(int&, int&);
}

is legal. I would find it very weird if the latter is illegal.

Douglas Gregor — Thu, 06 May 2010 23:34:47 GMT

Oops, thanks Steven! I was convinced that the latter was ill-formed, but I've reviewed the standard again and you are correct. Clang now implements the right semantics, so we should close out this patch as being wrong. Sorry about that!

status changed; resolution set

Douglas Gregor — Fri, 07 May 2010 19:56:41 GMT

status new → closed
resolution → invalid