Memos

createMemo compute returns a Memo<'T>, a derived value that recomputes when something it read has changed. A memo is lazy: it runs on its first read, and its Status is Uninitialized until then. compute receives the memo's previous value; the examples before this section ignore it with fun _ ->.

Peek returns the last computed value untracked, as stored, even when the memo is stale. Use untrack (fun () -> memo.Value) for a current value without recording a dependency; see Untrack.

Writes leave this unobserved memo stale. Peek prints its stored value; Read refreshes it. Watch the memo's run count as you step through the replay.

Glitch-free

Propagation is glitch-free. In a diamond, where two memos read one signal and an effect reads both memos, the effect runs once per write and always sees both memos at the same write.

let a = createSignal 1
let plusOne = createMemo (fun _ -> a.Value + 1)
let timesTen = createMemo (fun _ -> a.Value * 10)
let seen = ResizeArray ()
createEffect (fun () -> seen.Add (plusOne.Value, timesTen.Value))
// How many values does seen have? What are the values?
a.Value <- 2
// How about now?
a.Value <- 3
// And now?

A memo read by several effects recomputes once per write.

let a = createSignal 1

let shared = createMemo (fun _ -> a.Value * 100)
// 1. How many times has shared run?
createEffect (fun () -> shared.Value |> ignore)
// 2. What about now?
createEffect (fun () -> shared.Value |> ignore)
// 3. and now?
a.Value <- 2
// 4. and now?
a.Value <- 3
// 5. and now?

Pure and owning memos

let owningLog =
    use graph = new Graph ()
    use _ = graph.Activate ()
    let user = createSignal "ada"
    let log = ResizeArray ()

    let greeting =
        createMemoWith (fun _ ->
            let name = user.Value
            onCleanup (fun () -> log.Add $"release {name}")
            $"hello {name}")

    log.Add greeting.Value
    user.Value <- "grace"
    log.Add greeting.Value
    greeting.Dispose ()
    List.ofSeq log

owningLog |> List.iter (printfn "%s")

hello ada
release ada
hello grace
release grace

Construct Memo (graph, compute) directly for a pure memo, and Memo (graph, compute, true) for an owning one. createAsync and createAsyncWith split the same way, and a boundary always owns the nodes its body creates.

An owning computation replaces the nodes its body creates on every re-run. An async value created and read in the same body restarts its flight each time it settles, and never settles. Create it outside and read it inside; see Troubleshooting.

An owning memo's cleanups run untracked, whichever computation read the memo: a signal read by a cleanup never becomes a dependency of the reader. A node created by a cleanup belongs to the memo and is disposed before its next run.

The body runs once per discharge. A cleanup that writes one of the memo's sources and then reads the memo runs the body at that read, and that run is the re-run: the memo holds one run's nodes.

The previous value

compute receives the value the memo last published. Use it to accumulate a total, keep a running maximum, or reuse part of a previous result.

Its type is 'T voption -> 'T:

• ValueNone on the first run.
• ValueSome previous after a value has been published. This is the same value Peek returns.

A total advances once per run, using the inputs read in that run. Several writes in a batch, or before an unobserved memo is read, contribute only their final values.

Compare a single write with two writes in a batch. The total adds 10, then adds only the batch's final 2.

let totals, totalRuns =
    use graph = new Graph ()
    use _ = graph.Activate ()
    let amount = createSignal 5
    let total = createMemo (fun prev -> ValueOption.defaultValue 0 prev + amount.Value)
    let seen = ResizeArray ()
    createEffect (fun () -> seen.Add total.Value)

    amount.Value <- 10

    batch (fun () ->
        amount.Value <- 1
        amount.Value <- 2)

    List.ofSeq seen, total.Runs

printfn "totals seen: %A, runs: %d" totals totalRuns

totals seen: [5; 15; 17], runs: 3

Returning the previous value unchanged triggers an equality cutoff: readers do not re-run just because the memo ran.

let highs, highestRuns =
    use graph = new Graph ()
    use _ = graph.Activate ()
    let reading = createSignal 20

    let highest =
        createMemo (fun prev ->
            match prev with
            | ValueSome best when best >= reading.Value -> best
            | _ -> reading.Value)

    let seen = ResizeArray ()
    createEffect (fun () -> seen.Add highest.Value)

    for r in [ 18; 25; 22; 19 ] do
        reading.Value <- r

    List.ofSeq seen, highest.Runs

printfn "highs seen: %A, memo runs: %d" highs highestRuns

highs seen: [20; 25], memo runs: 5

Equality cutoff

A write equal to the current value stops there: it does not schedule its readers.

A memo that recomputes to an equal value stops downstream bodies from running. Unlike an equal signal write, the upstream change has already marked its readers for check. Those checks bring their dependencies current and resolve clean when no dependency publishes a change. This can happen at any level of a chain. Equality describes the graph-wide policy.

let cutoffRuns, parityRuns =
    use graph = new Graph ()
    use _ = graph.Activate ()
    let count = createSignal 2
    let isEven = createMemo (fun _ -> count.Value % 2 = 0)
    let label = createMemo (fun _ -> if isEven.Value then "even" else "odd")
    let mutable effectRuns = 0
    createEffect (fun () -> label.Value |> ignore; effectRuns <- effectRuns + 1)

    count.Value <- 2 // equal write
    count.Value <- 4 // isEven recomputes to true again
    effectRuns, isEven.Runs

printfn "effect runs: %d, isEven runs: %d" cutoffRuns parityRuns

effect runs: 1, isEven runs: 2

In the map, an equal write stops at count. A write that keeps the parity re-runs isEven, which recomputes to the same value and stops there. A write that flips the parity reaches the effect.

Dynamic dependencies

A memo or effect tracks its reads afresh on every run. If a branch stops reading a source, that source stops triggering the computation. Taking the branch again restores the dependency.

Reading the same source twice in one run records one dependency.

let branchLog =
    use graph = new Graph ()
    use _ = graph.Activate ()
    let useFirst = createSignal true
    let first = createSignal "a"
    let second = createSignal "x"
    let log = ResizeArray ()
    createEffect (fun () -> log.Add (if useFirst.Value then first.Value else second.Value))

    useFirst.Value <- false // now reads `second`
    first.Value <- "b" // not read by the last run: no re-run
    second.Value <- "y"
    useFirst.Value <- true // reads `first` again
    first.Value <- "c"
    List.ofSeq log

printfn "%A" branchLog

["a"; "x"; "y"; "b"; "c"]

Toggle the branch to move the effect's edge between first and second. A write to the source off the branch wakes nothing.

Read a disposed memo

A disposed memo keeps its last computed value. It stops recomputing and stops waking its readers, even if it was stale when disposed.

let disposedValue, disposedRuns =
    use graph = new Graph ()
    use _ = graph.Activate ()
    let source = createSignal 1
    let tenfold = createMemo (fun _ -> source.Value * 10)
    tenfold.Value |> ignore
    source.Value <- 2
    tenfold.Dispose ()
    tenfold.Value, tenfold.Runs

printfn "value after disposal: %d, runs: %d" disposedValue disposedRuns

value after disposal: 10, runs: 1