Thou shalt have no other gods before the ANSI C standard 1647

CBFalconer

You're confusing machine dependence with abstraction.

Sure, there's lots of code that works fine with variable integer sizes, with the variable specified at compile time (templates, typedef, etc.) or at run time (bigint, etc.). The benefits of this generality often outweigh the costs.

But using register-sized variables (long, etc.) does not produce this generality. A perfect illustration is the same real-world problem of a counter above 2^32 being stored in a 32-bit ``long'' variable. How do we fix this problem? Do we tell the users to change the meaning of ``long'' by throwing away their CPU (or modifying the compiler to break the ABI)? Of course not. The mistake was using ``long'' in the first place.

I should emphasize that this is separate from the costs of machine dependence, namely portability problems. What I'm saying is that your buttignment of benefits to machine dependence is fundamentally wrong. The benefits you're talking about are from abstraction, which is not the same thing.

I should also emphasize that machine dependence is occasionally needed to achieve top speed. But this is an issue only for the small chunks of code that actually matter for speed.

---D. J. Bernstein, buttociate Professor, Department of Mathematics, Statistics, and Computer Science, University of Illinois at Chicago