No Data Races

A lambda, and so each task, gets its own copy of count. Changing the copy doesn't change the outside one, and the compiler warns about it:

fn main() {
    let count = 0
    const _ = [1, 2, 3].parallel_map(n => {
        count += 1
        n
    })
    print(count)
}

warning: `count` is a copy inside this lambda: the change doesn't reach the outside. Give the value back instead, or use `Shared` to share it (spec 3.3)0

That is what makes data races impossible: there is nothing two tasks share by accident. To count across tasks, share the value on purpose with Shared, which locks for you:

fn main() {
    const visits = Shared(0)
    const _ = [1, 2, 3, 4, 5].parallel_map(n => {
        visits.update(v => v + 1)
        n
    })
    print(visits.get())
}

5

In most languages, two threads adding to one counter can lose updates, a bug that shows up only sometimes. In Kavach the only shared value is a Shared, and its update takes the lock itself: there is no "remember to lock". See Shared State.