With the endian discussion, I looked at Boost.Integer to try to figure out why it didn't support long long and __int64 where possible. Since I couldn't see a reason why it didn't, I went ahead and implemented it. The only major interface change is adding additional ValueUpper template parameters to int_(max|min)_value_t and uint_value_t. They default to 0, so no valid old code should be affected. They allow the programmer to use, for example, uint_value_t<0x34567890,0x12> to request a type that can hold 0x1234567890 or int_min_value_t<-0x34567890,-0x12> to request a type that can hold -0x1234567890, which will properly fail if there are no types available over 32 bits (instead of just truncating the constant and giving the wrong result). I also took the opportunity to add int_exact_t templates for cases where an exact number of bits is needed. int_exact_t<32>::exact, for example. If this looks good I'll try to decipher the macros in the test suite to add some new ones. The changes don't cause any regressions from integer_test.cpp under my g++ 3.3.6, 3.4.6, or 4.1.1. In keeping with the rationale and noticing that it works just about all the compilers, I used no partial specialisation or recursion, so I'm hoping that it'll work fine in the more troublesome compilers as well. Unfortunately portage tell me that gcc-2* is to old to work with the rest of my system, so I haven't been able to try. ~ Scott McMurray