[adamc]

I thought the second comment was interesting:

>I believe the flaw in the argument is that the higher-level languages are actually more constrained, because they have made too many broad promises and are unable to tell when they can violate them without ill effect, even though it is obvious to the programmer. We might think that your example above would be a perfect case in which we could parallelize the loop... but what if sum() had side effects, such that it would produce the wrong result if not evaluated in order? What if sum() didn't have any directly detectable side effects, but called another function which called another function which did... sometimes, from code in a runtime eval? The compiler has no way of knowing this, so it has to assume the worst, and this is why it ends up being unable to make all the optimizations that one thinks it "should" be able to make. The human looks at it and says "duh, it's called 'sum'. If it has side effects, I'm going to smack somebody up the side of the head" and proceeds to optimize in the expected ways.<

[mjuyhjui]

Which is exactly why Fortran compilers are still faster than C++.