On Fri, Jul 8, 2011 at 7:10 PM, Chris Frey <cdf...@foursquare.net> wrote:
> On Thu, Jul 07, 2011 at 11:18:17PM +0200, Kai Sterker wrote:
>> but it may well be triggered by our mingling of "normal"
>> STL and TR1 headers.
>
> Are you referring to this, in base/hash_map.h?
>
> namespace std { using namespace tr1; }
>
> I suspect that's not quite safe. :-)
>
> I don't have gcc 4.6 handy, but I'd be curious what the results would
> be if the above line was commented out, and -std=c++0x or -std=gnu++0x
> was added to the command line.
I haven't found a binary gcc 4,6 package for Ubuntu 11.04, so I could
not really test either. With gcc 4.5, removing the supect line and
using one of the -std options didn't go well, however. So for now I
just changed the defines of hash_map and hash_set to
#define hash_map tr1::unordered_map
#define hash_set tr1::unordered_set
That way, stuff compiled with gcc 4.5 without the namespace trickery,
and I hope it will help with gcc 4.6 too. All in all, this is going to
get a bit messy, however. With c++0x, unordered_map and unordered_set
will move to namespace std. So we'll end up with more #ifdefs in that
header. OTOH, starting with gcc 4.4 (I believe), using
hash_map/hash_set produces a lot of deprecation warnings and I didn't
think it wise to just suppress them.
Maybe the better approach would be to do something like here
http://trac.wxwidgets.org/ticket/9532
But that would have to be replicated for the CMake build too, whereas
the current implementation works independent from the build system
used.
Anyway. please give it a try, Shirish, and let me know if it fixes
compilation for you.
Kai
_______________________________________________
Adonthell-general mailing list
Adonthell-general@nongnu.org
https://lists.nongnu.org/mailman/listinfo/adonthell-general
