WoofWare.Myriad.Plugins 2.1.17

There is a newer version of this package available.
See the version list below for details.
dotnet add package WoofWare.Myriad.Plugins --version 2.1.17
NuGet\Install-Package WoofWare.Myriad.Plugins -Version 2.1.17
This command is intended to be used within the Package Manager Console in Visual Studio, as it uses the NuGet module's version of Install-Package.
<PackageReference Include="WoofWare.Myriad.Plugins" Version="2.1.17" />
For projects that support PackageReference, copy this XML node into the project file to reference the package.
paket add WoofWare.Myriad.Plugins --version 2.1.17
#r "nuget: WoofWare.Myriad.Plugins, 2.1.17"
#r directive can be used in F# Interactive and Polyglot Notebooks. Copy this into the interactive tool or source code of the script to reference the package.
// Install WoofWare.Myriad.Plugins as a Cake Addin
#addin nuget:?package=WoofWare.Myriad.Plugins&version=2.1.17

// Install WoofWare.Myriad.Plugins as a Cake Tool
#tool nuget:?package=WoofWare.Myriad.Plugins&version=2.1.17


NuGet version GitHub Actions status License file

Project logo: the face of a cartoon Shiba Inu, staring with powerful cyborg eyes directly at the viewer, with a background of stylised plugs.

Some helpers in Myriad which might be useful.

These are currently somewhat experimental, and I personally am their primary customer. The RemoveOptions generator in particular is extremely half-baked.

If you would like to ensure that your particular use-case remains unbroken, please do contribute tests to this repository. The ConsumePlugin assembly contains a number of invocations of these source generators, so you just need to add copies of your types to that assembly to ensure that I will at least notice if I break the build; and if you add tests to WoofWare.Myriad.Plugins.Test then I will also notice if I break the runtime semantics of the generated code.

Currently implemented:

  • JsonParse (to stamp out jsonParse : JsonNode -> 'T methods);
  • JsonSerialize (to stamp out toJsonNode : 'T -> JsonNode methods);
  • RemoveOptions (to strip option modifiers from a type).
  • HttpClient (to stamp out a RestEase-style HTTP client).
  • GenerateMock (to stamp out a record type corresponding to an interface).
  • CreateCatamorphism (to stamp out a non-stack-overflowing catamorphism for a discriminated union).


Takes records like this:

type InnerType =
        [<JsonPropertyName "something">]
        Thing : string

/// My whatnot
type JsonRecordType =
        /// A thing!
        A : int
        /// Another thing!
        B : string
        [<System.Text.Json.Serialization.JsonPropertyName "hi">]
        C : int list
        D : InnerType

and stamps out parsing methods like this:

/// Module containing JSON parsing methods for the InnerType type
module InnerType =
    /// Parse from a JSON node.
    let jsonParse (node: System.Text.Json.Nodes.JsonNode) : InnerType =
        let Thing = node.["something"].AsValue().GetValue<string>()
        { Thing = Thing }
namespace UsePlugin

/// Module containing JSON parsing methods for the JsonRecordType type
module JsonRecordType =
    /// Parse from a JSON node.
    let jsonParse (node: System.Text.Json.Nodes.JsonNode) : JsonRecordType =
        let D = InnerType.jsonParse node.["d"]

        let C =
            node.["hi"].AsArray() |> Seq.map (fun elt -> elt.GetValue<int>()) |> List.ofSeq

        let B = node.["b"].AsValue().GetValue<string>()
        let A = node.["a"].AsValue().GetValue<int>()
        { A = A; B = B; C = C; D = D }

You can optionally supply the boolean true to the attribute, which will cause Myriad to stamp out an extension method rather than a module with the same name as the type. This is useful if you want to reuse the type name as a module name yourself, or if you want to apply multiple source generators which each want to use the module name.

What's the point?

System.Text.Json, in a PublishAot context, relies on C# source generators. The default reflection-heavy implementations have the necessary code trimmed away, and result in a runtime exception. But C# source generators are entirely unsupported in F#.

This Myriad generator expects you to use System.Text.Json to construct a JsonNode, and then the generator takes over to construct a strongly-typed object.


This source generator is enough for what I first wanted to use it for. However, there is far more that could be done.

  • Make it possible to give an exact format and cultural info in date and time parsing.
  • Make it possible to reject parsing if extra fields are present.
  • Generally support all the System.Text.Json attributes.

For an example of using both JsonParse and JsonSerialize together with complex types, see the type definitions and tests.


Takes records like this:

[<WoofWare.Myriad.Plugins.JsonSerialize true>]
type InnerTypeWithBoth =
        Thing : string
        ReadOnlyDict : IReadOnlyDictionary<string, Uri list>

and stamps out modules like this:

module InnerTypeWithBoth =
    let toJsonNode (input : InnerTypeWithBoth) : System.Text.Json.Nodes.JsonNode =
        let node = System.Text.Json.Nodes.JsonObject ()

            node.Add (("it's-a-me"), System.Text.Json.Nodes.JsonValue.Create<string> input.Thing)

            node.Add (
                (fun field ->
                    let ret = System.Text.Json.Nodes.JsonObject ()

                    for (KeyValue (key, value)) in field do
                        ret.Add (key.ToString (), System.Text.Json.Nodes.JsonValue.Create<Uri> value)

                ) input.Map


As in JsonParse, you can optionally supply the boolean true to the attribute, which will cause Myriad to stamp out an extension method rather than a module with the same name as the type.

The same limitations generally apply to JsonSerialize as do to JsonParse.

For an example of using both JsonParse and JsonSerialize together with complex types, see the type definitions and tests.


Takes a record like this:

type Foo =
        A : int option
        B : string
        C : float list

and stamps out a record like this:

module Foo =
    type Short =
            A : int
            B : string
            C : float list

What's the point?

The motivating example is argument parsing. An argument parser naturally wants to express "the user did not supply this, so I will provide a default". But it's not a very ergonomic experience for the programmer to deal with all these options, so this Myriad generator stamps out a type without any options, and also stamps out an appropriate constructor function.


This generator is far from where I want it, because I haven't really spent any time on it.

  • It really wants to be able to recurse into the types within the record, to strip options from them.
  • It needs some sort of attribute to mark a field as not receiving this treatment.
  • What do we do about discriminated unions?


Takes a type like this:

type IPureGymApi =
    [<Get "v1/gyms/">]
    abstract GetGyms : ?ct : CancellationToken -> Task<Gym list>

    [<Get "v1/gyms/{gym_id}/attendance">]
    abstract GetGymAttendance : [<Path "gym_id">] gymId : int * ?ct : CancellationToken -> Task<GymAttendance>

    [<Get "v1/member">]
    abstract GetMember : ?ct : CancellationToken -> Task<Member>

    [<Get "v1/gyms/{gym_id}">]
    abstract GetGym : [<Path "gym_id">] gymId : int * ?ct : CancellationToken -> Task<Gym>

    [<Get "v1/member/activity">]
    abstract GetMemberActivity : ?ct : CancellationToken -> Task<MemberActivityDto>

    [<Get "v2/gymSessions/member">]
    abstract GetSessions :
        [<Query>] fromDate : DateTime * [<Query>] toDate : DateTime * ?ct : CancellationToken -> Task<Sessions>

and stamps out a type like this:

/// Module for constructing a REST client.
module PureGymApi =
    /// Create a REST client.
    let make (client : System.Net.Http.HttpClient) : IPureGymApi =
        { new IPureGymApi with
            member _.GetGyms (ct : CancellationToken option) =
                async {
                    let! ct = Async.CancellationToken

                    let httpMessage =
                        new System.Net.Http.HttpRequestMessage (
                            Method = System.Net.Http.HttpMethod.Get,
                            RequestUri = System.Uri (client.BaseAddress.ToString () + "v1/gyms/")

                    let! response = client.SendAsync (httpMessage, ct) |> Async.AwaitTask
                    let response = response.EnsureSuccessStatusCode ()
                    let! stream = response.Content.ReadAsStreamAsync ct |> Async.AwaitTask

                    let! node =
                        System.Text.Json.Nodes.JsonNode.ParseAsync (stream, cancellationToken = ct)
                        |> Async.AwaitTask

                    return node.AsArray () |> Seq.map (fun elt -> Gym.jsonParse elt) |> List.ofSeq
                |> (fun a -> Async.StartAsTask (a, ?cancellationToken = ct))

            // (more methods here)

What's the point?

The motivating example is again ahead-of-time compilation: we wish to avoid the reflection which RestEase does.



RestEase is complex, and handles a lot of different stuff.

  • If you set the BaseAddress on your input HttpClient, make sure to end with a trailing slash on any trailing directories (so "blah/foo/" rather than "blah/foo"). We combine URIs using UriKind.Relative, so without a trailing slash, the last component may be chopped off.
  • Parameters are serialised naively with toJsonNode as though the JsonSerialize generator were applied, and you can't control the serialisation. You can't yet serialise e.g. a primitive type this way (other than String); all body parameters must be types which have a suitable toJsonNode : 'a -> JsonNode method.
  • Deserialisation follows the same logic as the JsonParse generator, and it generally assumes you're using types which JsonParse is applied to.
  • Anonymous parameters are currently forbidden.

There are also some design decisions:

  • Every function must take an optional CancellationToken (which is good practice anyway); so arguments are forced to be tupled.
  • The [<Optional>] attribute is not supported and will probably not be supported, because I consider it to be cursed.


Takes a type like this:

type IPublicType =
    abstract Mem1 : string * int -> string list
    abstract Mem2 : string -> int

and stamps out a type like this:

/// Mock record type for an interface
type internal PublicTypeMock =
        Mem1 : string * int -> string list
        Mem2 : string -> int

    static member Empty : PublicTypeMock =
            Mem1 = (fun x -> raise (System.NotImplementedException "Unimplemented mock function"))
            Mem2 = (fun x -> raise (System.NotImplementedException "Unimplemented mock function"))

    interface IPublicType with
        member this.Mem1 (arg0, arg1) = this.Mem1 (arg0, arg1)
        member this.Mem2 (arg0) = this.Mem2 (arg0)

What's the point?

Reflective mocking libraries like Foq in my experience are a rich source of flaky tests. The Grug-brained developer would prefer to do this without reflection, and this reduces the rate of strange one-in-ten-thousand "failed to generate IL" errors. But since F# does not let you partially update an interface definition, we instead stamp out a record, thereby allowing the programmer to use F#'s record-update syntax.


  • You may supply an isInternal : bool argument to the attribute. By default, we make the resulting record type at most internal (never public), since this is intended only to be used in tests; but you can instead make it public with [<GenerateMock false>].


Takes a collection of mutually recursive discriminated unions:

[<CreateCatamorphism "MyCata">]
type Expr =
    | Const of Const
    | Pair of Expr * Expr * PairOpKind
    | Sequential of Expr list
    | Builder of Expr * ExprBuilder

and ExprBuilder =
    | Child of ExprBuilder
    | Parent of Expr

and stamps out a type like this:

type ExprCata<'Expr, 'ExprBuilder> =
    abstract Const : Const -> 'Expr
    abstract Pair : 'Expr -> 'Expr -> PairOpKind -> 'Expr
    abstract Sequential : 'Expr list -> 'Expr
    abstract Builder : 'Expr -> 'ExprBuilder -> 'Expr

type ExprBuilderCata<'Expr, 'ExprBuilder> =
    abstract Child : 'ExprBuilder -> 'ExprBuilder
    abstract Parent : 'Expr -> 'ExprBuilder

type MyCata<'Expr, 'ExprBuilder> =
        Expr : ExprCata<'Expr, 'ExprBuilder>
        ExprBuilder : ExprBuilderCata<'Expr, 'ExprBuilder>

module ExprCata =
    let runExpr (cata : MyCata<'ExprRet, 'ExprBuilderRet>) (x : Expr) : 'ExprRet =
        failwith "this is implemented"

    let runExprBuilder (cata : MyCata<'ExprRet, 'ExprBuilderRet>) (x : ExprBuilder) : 'ExprBuilderRet =
        failwith "this is implemented"

What's the point?

Recursing over a tree is not easy to get right, especially if you want to avoid stack overflows. Instead of writing the recursion many times, it's better to do it once, and then each time you only plug in what you want to do.


  • Mutually recursive DUs are supported (as in the example above). Every DU in a recursive type Foo... and Bar... knot will be given an appropriate cata, as long as any one of those DUs has the [<CreateCatamorphism>] attribute.
  • There is limited support for records and for lists.
  • There is extremely brittle support for generics in the DUs you are cata'ing over. It is based on the names of the generic parameters, so you must ensure that generic parameters with the same name have the same meaning across the various cases in your recursive knot of DUs. (If you overstep the bounds of what this generator can do, you will get compile-time errors, e.g. with generics being constrained to each other's values.) See the List tests for an example, where we re-implement FSharpList<'a>.


I am not at all convinced of the correctness of this generator, and I know it is very incomplete (in the sense that there are many possible DUs you could write for which the generator will bail out). I strongly recommend implementing the identity catamorphism for your type and using property-based tests (as I do) to assert that the correct thing happens. Feel free to raise GitHub issues with code I can copy-paste to reproduce a case where the wrong thing happens (though I can't promise to look at them).

  • This is a particularly half-baked generator which has so far seen no real-world use. It likely has a bunch of 80/20 low-hanging fruit remaining, but it also likely has impossible problems to solve which I don't know about yet.
  • Only a very few kinds of DU field are currently implemented. For example, this generator can't see through an interface (e.g. the kind of interface one would use to implement the crate pattern to represent a GADT), so the generated cata will simply grant you access to the interface (rather than attempting to descend into it to discover recursive references). You can't nest lists deeply. All sorts of other cases are unaddressed.
  • This generator does not try to solve the "exponential diamond dependency" problem. If you have a case of the form type Expr = | Branch of Expr * Expr, the cata will walk into both Exprs separately. If the Exprs happen to be equal, the cata will nevertheless traverse them individually (that is, it will traverse the same Expr twice). Your type may represent a DAG, but we will always effectively expand it into a tree of paths and operate on each of the exponentially-many paths.

Detailed examples

See the tests. For example, PureGymDto.fs is a real-world set of DTOs.

How to use

  • In your .fsproj file, define a helper variable so that subsequent steps don't all have to be kept in sync:
  • Take a reference on WoofWare.Myriad.Plugins.Attributes (which has no other dependencies), to obtain access to the attributes which the generator will recognise:
        <PackageReference Include="WoofWare.Myriad.Plugins.Attributes" Version="2.0.2" />
  • Take a reference (with private assets, to prevent these from propagating to your own assembly) on WoofWare.Myriad.Plugins, to obtain the plugins which Myriad will run, and on Myriad.Sdk, to obtain the Myriad binary itself:
        <PackageReference Include="WoofWare.Myriad.Plugins" Version="$(WoofWareMyriadPluginVersion)" PrivateAssets="all" />
        <PackageReference Include="Myriad.Sdk" Version="0.8.3" PrivateAssets="all" />
  • Point Myriad to the DLL within the NuGet package which is the source of the plugins:
      <MyriadSdkGenerator Include="$(NuGetPackageRoot)/woofware.myriad.plugins/$(WoofWareMyriadPluginVersion)/lib/net6.0/WoofWare.Myriad.Plugins.dll" />

Now you are ready to start using the generators. For example, this specifies that Myriad is to use the contents of Client.fs to generate the file GeneratedClient.fs:

    <Compile Include="Client.fs" />
    <Compile Include="GeneratedClient.fs">

Myriad Gotchas

  • MsBuild doesn't always realise that it needs to invoke Myriad during rebuild. You can always save a whitespace change to the source file (e.g. Client.fs above), and MsBuild will then execute Myriad during the next build.
  • Fantomas, the F# source formatter which powers Myriad, is customisable with editorconfig, but it does not easily expose this customisation except through the standalone Fantomas client. So Myriad's output is formatted without respect to any conventions which may hold in the rest of your repository. You should probably add these files to your fantomasignore if you use Fantomas to format your repo; the alternative is to manually reformat every time Myriad changes the generated files.
Product Compatible and additional computed target framework versions.
.NET net6.0 is compatible.  net6.0-android was computed.  net6.0-ios was computed.  net6.0-maccatalyst was computed.  net6.0-macos was computed.  net6.0-tvos was computed.  net6.0-windows was computed.  net7.0 was computed.  net7.0-android was computed.  net7.0-ios was computed.  net7.0-maccatalyst was computed.  net7.0-macos was computed.  net7.0-tvos was computed.  net7.0-windows was computed.  net8.0 was computed.  net8.0-android was computed.  net8.0-browser was computed.  net8.0-ios was computed.  net8.0-maccatalyst was computed.  net8.0-macos was computed.  net8.0-tvos was computed.  net8.0-windows was computed. 
Compatible target framework(s)
Included target framework(s) (in package)
Learn more about Target Frameworks and .NET Standard.

NuGet packages

This package is not used by any NuGet packages.

GitHub repositories

This package is not used by any popular GitHub repositories.

Version Downloads Last updated
2.1.27 10 5/24/2024
2.1.26 15 5/24/2024
2.1.25 27 5/24/2024
2.1.24 78 5/20/2024
2.1.23 67 5/20/2024
2.1.22 101 5/6/2024
2.1.21 76 4/30/2024
2.1.20 84 4/29/2024
2.1.19 74 4/29/2024
2.1.18 78 4/22/2024
2.1.17 80 4/17/2024
2.1.16 77 4/16/2024
2.1.15 84 4/16/2024
2.1.14 82 4/15/2024
2.1.13 98 3/19/2024
2.1.12 81 3/11/2024
2.1.11 93 3/4/2024
2.1.10 104 2/26/2024
2.1.9 93 2/26/2024
2.1.8 87 2/25/2024
2.1.7 95 2/25/2024
2.1.6 85 2/25/2024
2.1.5 80 2/19/2024
2.1.4 80 2/19/2024
2.1.3 73 2/18/2024
2.1.2 72 2/18/2024
2.1.1 74 2/17/2024
2.0.9 78 2/14/2024
2.0.8 90 2/13/2024
2.0.7 78 2/13/2024
2.0.6 76 2/13/2024
2.0.5 95 2/12/2024
2.0.4 63 2/7/2024
2.0.3 67 2/7/2024
2.0.2 73 2/7/2024
2.0.1 83 2/7/2024
1.4.15 87 2/6/2024
1.4.14 82 2/6/2024
1.4.13 80 2/6/2024
1.4.12 85 2/6/2024
1.4.11 86 2/6/2024
1.4.10 68 2/5/2024
1.4.9 81 1/30/2024
1.4.8 90 1/29/2024
1.4.7 74 1/29/2024
1.4.6 73 1/29/2024
1.4.5 76 1/28/2024
1.4.4 76 1/28/2024
1.4.3 77 1/26/2024
1.4.2 76 1/26/2024
1.4.1 74 1/26/2024
1.3.5 78 1/25/2024
1.3.4 94 1/15/2024
1.3.3 84 1/15/2024
1.3.2 98 1/8/2024
1.3.1 95 1/8/2024
1.2.3 93 1/3/2024
1.2.2 99 12/31/2023
1.2.1 102 12/30/2023
1.1.15 110 12/30/2023
1.1.14 111 12/30/2023
1.1.13 107 12/30/2023
1.1.12 105 12/30/2023
1.1.11 89 12/30/2023
1.1.10 111 12/29/2023
1.1.9 101 12/29/2023
1.1.8 105 12/29/2023
1.1.7 113 12/29/2023
1.1.6 114 12/29/2023
1.1.5 106 12/29/2023
1.1.4 102 12/29/2023
1.1.3 106 12/29/2023
1.1.2 93 12/28/2023
1.1.1 96 12/28/2023
1.0.6 99 12/28/2023
1.0.5 94 12/28/2023
1.0.4 96 12/27/2023