Re: std::tr1::unordered_set problem in GCC 4.1.2

From:

Howard Hinnant <howard.hinnant@gmail.com>

Newsgroups:

comp.lang.c++.moderated

Date:

Sun, 8 Apr 2007 17:14:56 CST

Message-ID:

<howard.hinnant-CA4A33.17581408042007@johnf2.biosci.ohio-state.edu>

In article <hNWdnSoVZfLfoYTbnZ2dnUVZ_vKunZ2d@giganews.com>,
Pete Becker <pete@versatilecoding.com> wrote:

Howard Hinnant wrote:

Your class actually looks like:

class Traxboard {
    public:
      Traxboard();
      bool operator==(const Traxboard& another) const;
      bool operator&();
      int getHash() const;
};

The gcc code is taking the address of a Traxboard and you're returning a
bool. gcc wants a Traxboard* when it says &(some expression yielding an
lvalue Traxboard).

Which is, nevertheless, a gcc bug, since the key type for an unordered
container is allowed to overloaed operator&. Of course, that's not a
good idea, and should probably not be allowed. It wreaks havoc unless
library code that needs an address carefully works around it.

I believe I disagree with you on both points:

1. The current standard requires &t to yield the address of t.

2. This requirement is unnecessary and should be removed for C++0X.
Containers should use allocator::address(t), and that should work along
the lines of boost::address_of.

References:

20.1.3 [lib.copyconstructible] in C++03 ([copyconstructible]) in N2134.

http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1856.html#20.1.4
%20-%20Copy%20construction

-Howard

--
[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]