Published Tags

Writing a VSCode extension with Fable 2.1


This post is part of the F# Advent Calendar in English 2018. I’m very grateful to be a part of this and I hope you will enjoy this one as much as I have enjoyed the others.

Today I would like to show you how you can create VS Code extension from scratch using Fable. Along the way we write some code that will extend the Markdown syntax.

Project setup

Code extension

To get started we will follow the general approach of scaffolding a new extension. Using the yeoman generator we will scaffold a new JavaScript extension.

vs code

Adding Fable

In order to add Fable we need two main pieces: a couple of npm modules and some NuGet packages.
First, we will add them with yarn.

PS> cd fable-markdown-extension

PS> yarn add -D “fable-compiler” “fable-splitter”

We won’t be using webpack as we don’t really need a bundle in this scenario. When a plugin is installed all JavaScript is executed in a node environment. All files are on disk so there is no need to bundle in order to spare some network requests.

Next we need a F# project. This can be a simple netstandard classlib.

PS> dotnet new classlib -lang F# -o src -n AdventExtension

This created a src folder with a Library.fs file in it. Let us try and compile this to JavaScript using fable-splitter.
In our package.json we can add a new script called build.

  "scripts": {
    "postinstall": "node ./node_modules/vscode/bin/install",
    "build": "fable-splitter --config splitter.config.js"

And as you would expect we will also need a splitter.config.js file.
The configuration looks like this:

const path = require("path");

function resolve(relativePath) {
  return path.join(__dirname, relativePath);

module.exports = {
  entry: resolve("src/AdventExtension.fsproj"),
  outDir: resolve("out"),
  allFiles: true

Execute by running yarn build. This is the equivalent of running npm run build. Yarn just doesn’t need the run word.
We see that all the files in the project are compiled to the out folder as specified in splitter.config.js.

Now let us rename Library.fs and change it to Extension.fs, compile again and try to run our extension.

The main property of our package.json is pointing to ./extension, so we will need to change that into out/Extension.js.
However if we look at the existing extension.js file it seems to expose an activate function.

// The module 'vscode' contains the VS Code extensibility API
// Import the module and reference it with the alias vscode in your code below
const vscode = require('vscode');

// this method is called when your extension is activated
// your extension is activated the very first time the command is executed
function activate(context) {

    // Use the console to output diagnostic information (console.log) and errors (console.error)
    // This line of code will only be executed once when your extension is activated
    console.log('Congratulations, your extension "fable-markdown-extension" is now active!');

    // The command has been defined in the package.json file
    // Now provide the implementation of the command with  registerCommand
    // The commandId parameter must match the command field in package.json
    let disposable = vscode.commands.registerCommand('extension.sayHello', function () {
        // The code you place here will be executed every time your command is executed

        // Display a message box to the user
        vscode.window.showInformationMessage('Hello World!');

exports.activate = activate;

// this method is called when your extension is deactivated
function deactivate() {
exports.deactivate = deactivate;

So we will need this in our Fable code as well.
Changing Extension.fs to

module AdventExtension.Extension

let activate _ =
    printfn "Fable extension is activated!"

should do the trick.

At this point we have had to compile for a couple times already. It would be easier if we could compile continuously.
This can be achieved by means of another fable-splitter, so let’s add a new build script watch:

    "scripts": {
        "postinstall": "node ./node_modules/vscode/bin/install",
        "build": "fable-splitter --config splitter.config.js",
        "watch": "yarn run build --watch"

yarn watch

Our compiled code looks like

import { toConsole, printf } from "./fable-library.2.1.8/String";
export function activate(_arg1) {
  toConsole(printf("Fable extension is activated!"));

and we can launch the extension in code by hitting the play button in the Debug panel.
However in our current setup we want our extension to run when VS Code starts.

Change the activationEvents to * in the package.json.

  "activationEvents": [

and start debugging.


This opens a new instance of vscode and should contain our extension.

Yikes, this isn’t really doing anything at all!
We even see a crash when we toggle the developer tools.

crash in logs

What code is telling us here is that it doesn’t know how to deal with the import and export keywords.
I believe this whole thing is running in a node context, so we should be using module.exports instead.

One small Babel plugin can do this for us. We need to download it with yarn and tell fable-splitter to use it as well.

PS> yarn add -D “@babel/plugin-transform-modules-commonjs” “@babel/core”

splitter.config.js becomes

const path = require("path");

function resolve(relativePath) {
  return path.join(__dirname, relativePath);

module.exports = {
  entry: resolve("src/AdventExtension.fsproj"),
  outDir: resolve("out"),
  babel: {
    plugins: ["@babel/plugin-transform-modules-commonjs"]
  allFiles: true

The JavaScript output now looks like

"use strict";

Object.defineProperty(exports, "__esModule", {
  value: true
exports.activate = activate;

var _String = require("./fable-library.2.1.8/String");

function activate(_arg1) {
  (0, _String.toConsole)((0, _String.printf)("Fable extension is activated!"));

and our extension now runs!

extension runs

F# dependencies

Ok, after some trial and error we have the basics in place. However, a little intelliSense would not hurt.
There is a NuGet package for that: Fable.Import.VSCode.

We could install it into our fsproj with dotnet add, but I would prefer to download it with paket instead.
Paket can easily be added to a project using:

PS> dotnet tool install —tool-path “.paket” Paket —add-source

This downloads the .NETCore version of Paket into a .paket folder. This should be familiar to most of us.
Next we enter .paket\paket.exe init.

Open the paket.dependencies file and add

framework: netstandard2.0

# nuget Fable.Import.Jest // Use when ready for Fable2
github nojaf/fable-jest:fable2 fable/Bindings.fs
github nojaf/fable-jest:fable2 fable/Exports.fs
github nojaf/fable-jest:fable2 fable/Matchers.fs
nuget FSharp.Core redirects:force
nuget Fable.Import.VSCode
nuget Fable.Parsimmon

It will download VS Code bindings, Jest bindings and Fable.Parsimmon. I’ll explain later why we need those.
Then, create a paket.references file next to the fsproj.


File:Bindings.fs Jest
File:Exports.fs Jest
File:Matchers.fs Jest

Run .paket\paket.exe install, which should add everything on .NET side.

Those last npm packages

To wrap up the setup of the project, we still need a few npm packages.

yarn add -D “jest” “markdown-it”

That should be it. Thanks for sticking around if you are still here 😊.
Now we can finally start coding.


I feel like we have been through a lot already, and I haven’t really revealed the master plan of all this.
I mentioned that we would extend Markdown. In this blogpost we will add two new features:

Whenever we type :fable: in our Markdown file, we would like to the Fable logo to show in our preview.

Fable text

If we add text between two 🐉 emoji, it should rendered in a different font.

Both features should also be highlighted in the editor itself.

Test-driven development

Now that we know what we want, we can start thinking about solving the problem.
Without considering VS Code we know that we will need a way to detect:fable: and text between 🐉.

What better way of parsing text than using Fable.Parsimmon?
If you have any experience with FParsec, the API is very similar.

In order to have a short feedback loop, we can easily do some TDD instead of trying things out in the extension itself.

We already installed Jest and the accompanying F# bindings. Note that these bindings were installed with paket through github instead of nuget. This has to do with the fact that Fable.Jest doesn’t work 100% with Fable 2.X.

My fork of Fable.Jest contains a workaround for this.
Here you can already see that using paket instead of NuGet was a smart move.

Unit testing in Fable is described to some extent in the FAQ. In our case everything is almost fully set up. We need to change the test script in our package.json.

"scripts": {
    "postinstall": "node ./node_modules/vscode/bin/install",
    "build": "fable-splitter --config splitter.config.js",
    "watch": "yarn run build --watch",
    "test": "jest"

Add a file that matches the following convention *.test.fs to our project, and the *.test.js should be picked up by Jest.
Parser.test.fs with a simple unit test should be enough for now.

module AdventExtension.Tests.Parser

open Fable.Import.Jest
open Fable.Import.Jest.Matchers

test "simple Jest test should be picked up" <| fun () ->
    "42" == "the answer of the Universe"

Execute with

PS> yarn build

PS> yarn test

Coolio, we can unit test now! The way I like to proceed now is having two console windows. One to run fable-splitter with yarn watch, and other to run tests with yarn test --watchAll.


Since we are going to parse text, we need a type to capture the result of our parsing.

type ParserResult =
    | Text of string
    | FableFontText of NodeResult<string>
    | FableLogo of NodeResult<string>

We may now have found some text, text between dragons (🐉) or :fable:.
When we have found one or both of the latter two, we would need to know where in our text we have found them.
This is where Parsimmon.node comes in handy, as it returns a NodeResult<'t>.

type TokenPosition =
    { offset: int
      line: int
      column: int }

type NodeResult<'t> =
    { name: string
      value: 't
      start: TokenPosition
      ``end``: TokenPosition }

Next we need a function that takes a string and returns Fable.Parsimmon.ParseResult<ParserResult array>.

let parseText input : ParseResult<ParserResult array> = failwith "Nothing here yet."

And we can quickly write some tests.

test "basic fable font" <| fun () ->
    let input = "🐉vibes🐉"
    let parsed = Parser.parseText input
    parsed.status == true

test "no fable font" <| fun () ->
    let input = "fableless vibes"
    let parsed = Parser.parseText input
    parsed.status == true

test "fable logo is found" <| fun () ->
    let input = "Once upon a :fable:..."
    let parsed = Parser.parseText input

    parsed.status == true

    match parsed.value with
    | [| Text _; FableLogo logo; Text _|] ->
        logo.start.offset == 12
        logo.value.Length == 7
    | _ -> failwith "Expected logo in center"

Notice that == is a shorthand for Fable.Import.Jest.Exports.expect.Invoke("foo").toBe("bar").

To make these tests pass we’ll write some custom parsers, mostly based on regular expressions.

module AdventExtension.Parser

open Fable.Parsimmon
open Fable.Core
open System.Text.RegularExpressions

type ParserResult =
    | Text of string
    | FableFontText of NodeResult<string>
    | FableLogo of NodeResult<string>

// no /g flag
[<Emit("new RegExp($0)")>]
let createRegex pattern : Regex = jsNative

let dragonParser = Parsimmon.str "🐉"

let noDragonParser =
    Parsimmon.regex (createRegex @"[^🐉]+")

let fableFontParser =
    Parsimmon.between dragonParser dragonParser noDragonParser
    |> Parsimmon.node "FableFont"
    |> (ParserResult.FableFontText)

let halfDragon =
    |> Array.head
    |> (string)

let noDragonsParser =
    Parsimmon.satisfy (fun token -> token <> halfDragon && token <> ":")
        |> Parsimmon.atLeastOneOrMany
        |> Parsimmon.concat
        |> (ParserResult.Text)

let fableLogoParser =
    Parsimmon.str ":fable:"
    |> Parsimmon.node "FableLogo"
    |> (FableLogo)

let parseText input : ParseResult<ParserResult array> =
    Parsimmon.choose [
    |> fun parser -> parser.many().parse(input)

I won’t go into detail about exactly how this works but I do want to highlight two things:

Presumably this code is more or less readable, and we can continue 😅.

Highlighting code

Highlighting in VS Code is surprisingly easily. All we need to do is tell code which range should be decorated with which style.

Let’s first create some TextDecorationEditorStyle objects.

module AdventExtension.Extension

open Fable.Core.JsInterop
open Fable.Import.vscode

let createFontOptions color backgroundColor fontWeight : DecorationRenderOptions =
    jsOptions (fun options ->
        options.color <- Some color
        options?fontWeight <- fontWeight
        options.backgroundColor <- backgroundColor

let activate _ =
    let fableFontStyle = window.createTextEditorDecorationType(createFontOptions "#87c5fd" (Some "white") None)
    let fableLogoStyle = window.createTextEditorDecorationType(createFontOptions "#87c5fd" None (Some "bold"))

    printfn "Fable extension activated!"

Now we need to subscribe to certain events.

let activate (context: ExtensionContext) =
    let fableFontStyle = window.createTextEditorDecorationType(createFontOptions "#87c5fd" (Some "white") None)
    let fableLogoStyle = window.createTextEditorDecorationType(createFontOptions "#87c5fd" None (Some "bold"))

    // Capture the active editor.
    let mutable activeEditor = window.activeTextEditor

    let updateDecorations() = failwithf "Nothing here yet."

    let mutable timeoutKey = None;
    let triggerUpdateDecorations _ =
        |> Option.iter (JS.clearTimeout)

        timeoutKey <- Some (JS.setTimeout updateDecorations 500)

    // User is typing
    window.onDidChangeActiveTextEditor.Invoke((fun editor ->
        activeEditor <- Some editor
    |> context.subscriptions.Add

    // Change of document
    workspace.onDidChangeTextDocument.Invoke((fun event ->
        match activeEditor with
        | Some aEditor when (event.document = aEditor.document) ->
        | _ -> ()
    |> context.subscriptions.Add

    |> Option.iter (triggerUpdateDecorations)

    printfn "Fable extension activated!"

Notice that the event handlers call triggerUpdateDecorations, which prevents us from re-rendering too much.

// Parsimmon ranges are 1-based, VS Code works with 0-based indexing.
let nodeResultToRange<'t> (nr: NodeResult<'t>) =
    let zeroBasedFloat r = r - 1 |> (float)
    vscode.Range(zeroBasedFloat nr.start.line,
                 zeroBasedFloat nr.start.column,
                 zeroBasedFloat nr.``end``.line,
                 zeroBasedFloat nr.``end``.column)

let updateDecorations() =
    |> Option.iter (fun aEditor ->
        if aEditor.document.languageId = "markdown" then
            let text = aEditor.document.getText()
            let parsed = Parser.parseText text

            if parsed.status then
                |> (fun node ->
                    match node with
                    | ParserResult.FableLogo logo ->
                        Some (fableLogoStyle, nodeResultToRange logo)

                    | ParserResult.FableFontText fableText ->
                        Some (fableFontStyle, nodeResultToRange fableText)

                    | _ ->

                |> Array.choose id
                |> Array.groupBy (fun (style,_) -> style.key)
                |> Array.iter (fun (styleKey, ranges) ->
                    let r =
               snd ranges
                        |> ResizeArray

                    let style = if fableFontStyle.key = styleKey then fableFontStyle else fableLogoStyle

                    aEditor.setDecorations(style, !^ r)

updateDecorations is rather straightforward as well. We map the result of parseText and make ranges out of it. The result looks as follows:

Markdown highlighting in VS Code

Preview mode

When we switch to preview mode of the Markdown file, we don’t really see any changes.

Preview mode before

This can be changed by writing our own markdown-it extension.
To be fair, it took me a while to figure out how this works. Even after reading the documentation, I was still somewhat puzzled.

What markdown-it does is it parses the Markdown content to a token stream.


This is some **markdown text**

Will be parsed to

    "type": "paragraph_open",
    "tag": "p",
    "type": "inline",
    "tag": "",
    "map": [
    "nesting": 0,
    "level": 1,
    "children": [
        "type": "text",
        "tag": "",
        "content": "This is some ",
        "type": "strong_open",
        "tag": "strong",
        "type": "text",
        "tag": "",
        "content": "markdown text",
        "type": "strong_close",
        "tag": "strong",
        "type": "text",
        "tag": "",
    "content": "This is some **markdown text**",
    "type": "paragraph_close",
    "tag": "p",

Note that this is a simplified version. We need to break down any tokens of type:"text" and replace them with our own token type (fable-logo and fable-text).

Once we’ve updated the tree, we can hook into the render engine by adding custom rules.

F# bindings

I found that there are TypeScript bindings on npm. With ts2fable I cooked up some minimal F# bindings which we can use.

Unit tests

Since this is a rather tricky situation, we should add some tests first.

module AdventExtension.Tests.MarkdownPlugin

open Fable.Import.Jest
open AdventExtension.MarkdownPlugin


let md = MarkdownItModule.Exports.Invoke().``use``(fableMarkdownPlugin)

test "fable img is rendered" <| fun () ->
    let input = ":fable:"
    let parsed = md.render input
    expect.Invoke(parsed.StartsWith("<p><img class='fable-logo' src")).toBeTruthy()

test "fable font is rendered" <| fun () ->
    let input = "Over the seas 🐉we shall rise🐉!"
    let parsed = md.render input
    expect.Invoke(parsed.Contains("<span class='fable'>we shall rise</span>")).toBeTruthy()

The plugin function fableMarkdownPlugin looks like

module AdventExtension.MarkdownPlugin

open MarkdownItModule
let fableMarkdownPlugin (md: MarkdownIt) (options: Options) = failwith "Nothing yet"

Inside this function, we need to replace the tokens and add custom render rules.

module AdventExtension.MarkdownPlugin

open MarkdownItModule
open Fable.Core
open AdventExtension.Parser

[<Emit("new $0.Token($1,$2,$3)")>]
let createToken state name tag nesting: Token = jsNative

[<Emit("$0[$1] = $2")>]
let addRenderRule rules name fn: unit = jsNative

let fableMarkdownPlugin (md: MarkdownIt) (options: Options) =
    // Phase one: replace text tokens to custom tokens
    let replaceTokens (state: State) =
        let parseTextToken (parentToken: Token) (textToken: Token) =
            let textContent = textToken.content
            let parsedContent = parseText textContent
            let updatedChildren =
                if not parsedContent.status then
                    |> (fun node ->
                        match node with
                        | ParserResult.Text text ->
                            let t = createToken state "text" "" 0
                            t.content <- text

                        | ParserResult.FableLogo _ ->
                            createToken state "fable-logo" "img" 0

                        | ParserResult.FableFontText fontText ->
                            let t = createToken state "fable-font" "span" 0
                            t.content <- fontText.value.Replace("🐉", "")

            parentToken.children <- new ResizeArray<Token>(updatedChildren)

        let rec parseToken (parentToken: Token) (token: Token) =
            if token.children <> null && not(Seq.isEmpty token.children) then
                |> Seq.iter (fun childToken ->
                    if childToken.``type`` = "inline" then
                        parseToken token childToken
                        parseTextToken token childToken

            if token.``type`` = "text" then
                parseTextToken parentToken token

        |> Seq.filter (fun t -> t.``type`` = "inline" && t.children <> null && not (Seq.isEmpty t.children))
        |> Seq.iter (fun token ->
            |> Seq.iter (parseToken token)

    md.core.ruler.push("fable", replaceTokens)

This code might not look like much, but we are traversing the tree, replacing the children of a parent node that contains text.

Our unit tests remain unaffected, as the render part will just ignore the fable-logo and fable-text tokens and render them as regular text.

Part two determines how our own nodes should render:

[<Emit("$0[$1] = $2")>]
let addRenderRule rules name fn: unit = jsNative

let fableMarkdownPlugin (md: MarkdownIt) (options: Options) = ...
    // Phase one ...

    // Phase two
    let renderFableLogo tokens idx = "<img class='fable-logo' src='data:image/png;base64,iVBORw0KGgoAAAANSUhEU...' />"

    let renderFableText (tokens: Token array) (idx: int) =
        let token = tokens.[idx]
        sprintf "<span class='fable'>%s</span>" (token.content.Replace("🐉", ""))

    addRenderRule md.renderer.rules "fable-logo" (new System.Func<Token array,int,string>(fun tokens idx -> renderFableLogo tokens idx))
    addRenderRule md.renderer.rules "fable-font" (new System.Func<Token array,int,string>(fun tokens idx -> renderFableText tokens idx))

We need to wrap our render functions inside a System.Func<_,_,_> in order to preserve compatibility with JavaScript.
The reason for this is that Fable compiles these into functions that can be curried, and markdown-it doesn’t like this.

Once more our tests show green, and the times has come to shove this into our VS Code extension.

Tag on code

To see all this in code we need to extend the contributes section in our package.json

  "contributes": {
    "markdown.markdownItPlugins": true,
    "markdown.previewStyles": [

Create a fable.css file

// this works in code, I was somewhat surprised by this. Should be a local url in real world scenario.
@import url(',600i');

.fable {
    font-family: 'Josefin Slab', serif;
    font-weight: 600;
    font-style: italic;
    color: #87c5fd;
    font-size: 18px;

.fable-logo {
    display: inline-block;
    width: 24px;

And make the final change to Extension.fs by adding

    // Add render plugin
    createObj [
        "extendMarkdownIt" ==> (fun md -> md?``use``(MarkdownPlugin.fableMarkdownPlugin))

at the end.

Final outcome:

preview after

The Source

You can find all code on my github.


Final words

I hope you enjoyed this blogpost and it all makes sense to you. If you have any suggestions or questions, please leave a comment.

Yours truly,

Photo by unsplash-logoMassimo Rinaldi