[v3] policy-based data structures, round two

[Paolo Carlini]

Hi Benjamin, Ami, all,

>tested on x86/linux
>tested on x86_64/linux
>tested on powerpc64/linux
>  
>
Looking at testresults (Janis, Andreas) and my own local tests, most of 
the regression tests fail at runtime on 64-bit machines. For example, on 
ia64, hash_data_map_rand.exe fails like this:

Program received signal SIGSEGV, Segmentation fault.
pb_ds::test::twister_rand_gen::get_unsigned_long_imp 
(this=0x6000000000006d08)
    at 
/usr/src/paolo/gcc-head/gcc/libstdc++-v3/testsuite/util/rng/twister_rand_gen.cc:182
182                   m_a_mt[kk] = m_a_mt[kk+(mers_m-mers_n)] ^ (y >> 1) 
^ m_a_mag01[y&  1];

Apparently, all those tests are using a built-in random number 
generator, I suspect it's not ok wrt 64-bit unsigned long... Currently 
I'm working on something else (tr1/random, funny ;) but if nobody jumps 
in I can soon provide more info...

Thanks,
Paolo.