FxDart 101 › Section 3 · Transforming › scan

scan

A lazy running accumulation — like reduce, but it yields every intermediate value instead of only the last.

Lecture

scan is reduce/fold with its intermediate steps exposed: instead of collapsing an iterable down to one final value, it emits every running accumulation, including the seed itself as the first value. That first-value-is-the-seed detail matters — scan(f, 0, [1, 2, 3]) yields four values (0, then three running sums), not three.

scan1 is the unseeded variant, ported from FxTS's scan(f, iterable) overload (no seed argument). It uses the first element of the iterable as the initial accumulator and yields it immediately, then keeps folding the rest — mirroring reduce's relationship to fold. On an empty iterable, scan1 has no first element to seed with, so it yields nothing at all. Note there is no chain method for scan1 (only scan is on Fx/FxAsync) — call it data-first: scan1(f, iterable).

Both are lazy: nothing runs until you pull. On the async side, scanAsync/scan1Async still fold one step at a time in order (each step needs the previous result), so .concurrent(n) doesn't parallelize the fold itself — but it does let an upstream fetch stage run concurrently, as long as the accumulator function itself stays cheap. See Demo 2.

Demo 1 · Basics — scan and scan1

Demo 2 · Async, with concurrency upstream

The fold itself stays sequential, but the fetch that feeds it doesn't have to:

Try it yourself

Exercise: use scan to produce a running total of steps, seeded at 0.