Address testing guide nits

[steveklabnik]

Fixes #19861

/cc @huonw

[alexcrichton]

r? @huonw

[huonw]

This benchmark is still incorrect: the text implies that this new benchmark is then fixed (i.e. accurately benchmarking the time it takes to xor 1000 uints), but it's not. It looks like we may need a new benchmark, or stronger black-boxing:

let mut n = 1000;
// `n` is unchanged, butthe optimiser is forced to assume that `n` may be mutated
test::black_box(&mut n);

b.iter(|| {
    test::black_box(range(0u, n).fold(...));
})

(or the version that does the range.fold without a semicolon.)

I haven't tested that, but one should see numbers on the order of the 375 ns/iter that it was before (the exact values isn't too important and varies based on the specific computer, but it should certainly be hundreds of nanoseconds, not ones).

[steveklabnik]

I'd like to finish this PR up, did you ever come up with a good replacement?

[huonw]

Oh, whoops, forgot to write it down:

    b.iter(|| {
        let mut n = 1000_u32;
        test::black_box(&mut n); // pretend to modify `n`
        range(0, n).fold(0, |a, b| a ^ b)
    })

[steveklabnik]

@huonw added the example. what do you think?