> one in which you can express a "loop exactly n times" construct directly
The problem with this advice is that there are so many things that the processor may be interested in knowing declaratively, rather than having to claw it back from the semantics of your application somehow, that the problem has become totally unmanageable. And even if we could manage it, the next generation of processor microcodes may be completely uninterested in what we've carefully encoded into your programs are actually want some other information instead.
I don't think the problems with that advice are singular. Perhaps the most ultimately-relevant is that looping exactly N times without a "ceremonial" counter is really only very rarely useful. Truly, how often do you see the canonical `for (i = 0; i < N; i++)` pattern in C with no references to `i` anywhere in the loop body? (Hint: if there are any, it's not "ceremonial".) Blindly repeating the exact same sequence of code just isn't something you want to do very often.
To address the issue of "okay, in what languages could we do this anyway?": taken in full mathematical generality, such a capability would necessarily require bignum support, which cuts down your options significantly. I'm not sure if that was actually the intent, but if we instead accept the limitation of "up to some largeish power of two", constructing a macro to do this "directly" in C would also be quite easy. So...(shrug)
I agree that it's rare, but aren't those exactly the cases where this kind of optimization is relevant? In the cases where you're using i and it could overflow, the optimization couldn't be applied anyway.
Premature optimization is the root of all evil - a lot of programs perform more than adequately when just written in a way that directly expresses programmer intent. In cases where you need to push the very limits of performance, you'll have to add extra processor-specific information (or at least information based on educated guesses about how processors in general behave) - I don't think there's any way to avoid that. But if you're including extra information in the source that clutters up developer intent and doesn't help the processor optimize, something has gone very wrong.
The problem with this advice is that there are so many things that the processor may be interested in knowing declaratively, rather than having to claw it back from the semantics of your application somehow, that the problem has become totally unmanageable. And even if we could manage it, the next generation of processor microcodes may be completely uninterested in what we've carefully encoded into your programs are actually want some other information instead.