Short for loop design (for i n { ... })

[bvssvni]

Dyon supports a short For loop for counters starting at 0 and incremented until it is greater or equal to a pre-evalutated expression:

for i len(list) { println(list[i]) }

This For loop is approximately 2.9x faster when running on AST than the equivalent traditional For loop:

n := len(list)
for i := 0; i < n; i += 1 { println(list[i]) }

Inferring range by indexing

The expression len(list) can be inferred by the index list[i] in the body:

for i { println(list[i]) }

When nesting loops this way, you have to order items in a list in the same order, for example:

for i, j, k {
    println(list[i][j][k]) // i, j, k must be used in the same order
}

However, as long as you don't depend on the previous indices, you can write it any way you like:

sum i, j {
    list[i] - list[j]
}

Specify range

With index start and end:

for i [2, len(list)) { println(list[i]) }

Packed loops

When nesting loops of same kind, you can pack them together by separating the indices with ",":

for i, j, k {
     println(list[i][j][k])
}

You can also write ranges in packed version, like for i n, j [i+1, n) { ... }.

Examples

Computing output for a neural network:

fn run__tensor_input(tensor: [[[f64]]], input: [f64]) -> {
    input := input
    for i {
        input = sift j {
            sigmoid(∑ k {
                tensor[i][j][k] * input[k]
            })
        }
    }
    return clone(input)
}

Compute energy for a system of N physical bodies:

fn energy(bodies: [{}]) -> f64 {
    n := len(bodies)
    return ∑ i n {
        bodies[i].vel · bodies[i].vel * bodies[i].mass / 2.0 -
        bodies[i].mass * ∑ j [i+1, n) {
            bodies[j].mass / |bodies[i].pos - bodies[j].pos|
        }
    }
}

Set all weights in a neural network to random values:

fn randomize__tensor(mut tensor: [[[f64]]]) {
    for i, j, k {
        tensor[i][j][k] = random()
    }
}

Motivation

This is designed for:

• Reduce typing
• Reduce bugs
• Improve readability of code