RFC FS-1031 - Mixing ranges and values to construct sequences

[edgarfgp]

• Suggestion: Mixing ranges and values to construct sequences fslang-suggestions#1031
• Proposed Implementation Allow mixing ranges and values to construct sequences dotnet/fsharp#18670

[brianrourkeboll]

Question 1 from out of band discussion: should if … else … (or match … with … -> … | … -> …) with heterogeneous implicit yield/yield! be supported?

[if p then 1 else 1..10]

Right now, an explicit yield! of a sub-collection is allowed:

[if p then yield 1 else yield! [1..10]]

[if p then 1 else yield! [1..10]]

[brianrourkeboll]

I think that we would want to require explicit yield!—

[if p then 1 else yield! 1..10]

—just as we do not allow

[if p then 1 else [1..10]]

but require instead

[if p then 1 else yield! [1..10]]

Or we could only allow bare ranges at the top level and avoid this question altogether.

[brianrourkeboll]

...And even if they're not mixed? These (with implicit yields) are currently allowed:

let f p q r =
    [
        if p then
            if q then 3
        if p then
            if q then
                if r then 3
    ]

let f x (|P|_|) (|Q|_|) (|R|_|) =
    [
        match x with
        | P _ -> 3
        | _ -> ()

        match x with
        | P y ->
            match y with
            | Q z ->
                match z with
                | R -> 3
                | _ -> ()
            | _ -> ()
        | _ -> ()
    ]

Should their equivalents with bare ranges be allowed and have implicit yield! inserted?

let f p q r =
    [
        if p then
            if q then 1..10
        if p then
            if q then
                if r then 1..10
    ]

let f x (|P|_|) (|Q|_|) (|R|_|) =
    [
        match x with
        | P _ -> 1..10
        | _ -> ()

        match x with
        | P y ->
            match y with
            | Q z ->
                match z with
                | R -> 1..10
                | _ -> ()
            | _ -> ()
        | _ -> ()
    ]

[edgarfgp]

I would be inclined to say Yes l, as it would be consistent with how we treat yield! [ 1..10 ]

[brianrourkeboll]

Question 2 from out of band discussion: should explicit yield! 1..10 be supported?

[yield! 1..10]

We currently require an explicit specification of a collection/computation expression builder

[yield! [1..10]]

seq { yield! seq { 1..10 } }

[brianrourkeboll]

(I think it would make sense to allow this, if it is not allowed in dotnet/fsharp#18670 already.)

[edgarfgp]

Agreed. Added support for it in prototype PR

[charlesroddie]

An example of why you would want to do this would be useful. So far it seems speculative: someone might want to create disjoint sets of intervals and points. Is this about typing in sequences representing things that stop and restart and therefore consist of disjoint intervals?