What is testing.B.Loop (Go 1.24) and why is it preferred over the classic b.N loop?

b.Loop() returns true while the benchmark should keep iterating. It fixes three classic b.N mistakes:

1. Setup cost: the timer is reset on the first call, so expensive setup before the loop is excluded without b.ResetTimer(). Cleanup after the loop is also excluded.
2. Dead-code elimination: the compiler keeps the arguments and results of calls inside a b.Loop body alive, so an inlined pure function can't be optimized away.
3. Setup runs once: with b.N, the whole function runs multiple times (b.N = 1, 100, 10000, ...), repeating setup each time. With b.Loop, the function body is executed once and iteration ramp-up happens inside the loop.

func BenchmarkParse(b *testing.B) {
    input := loadFixture(b) // excluded from timing automatically
    b.ReportAllocs()
    for b.Loop() {
        parse(input) // result kept alive; not eliminated
    }
}

// Old style (still valid, but easy to get wrong):
func BenchmarkParseOld(b *testing.B) {
    input := loadFixture(b)
    b.ResetTimer()
    for range b.N {
        parse(input)
    }
}

Gotchas: call b.Loop in exactly one loop per benchmark and don't mix it with b.N. Don't reference b.N inside a b.Loop benchmark to size things.