The project Pharo Language Server is an implementation of the Language Server Protocol proposed by Microsoft and implemented by several IDE such as emacs, eclipse, Intellij Idea, VSCode... The project GitHub repository includes:

1. An abstract layer that can be extended to create a new dedicated language server (for example: for the Java Programming Language)
2. A Pharo Language Server implementation that works well with the pharo-vscode plugin.
3. A Debug Adapted protocol implementation for Pharo (that needs another documentation page and is not discussed here).

In this documentation page, we present quickly the protocol, how one can download and install the Pharo Language Server project, its structure, and how to extends it.

• Language Server Protocol
• Pharo Language Server Installation
• Pharo Language Server Structure
  • Package architecture
  • Server class architecture
  • Starting the server
    • Extract request
    • Handle request
  • Method of the protocol
• Extending the Abstract Language Server to implement a new one

Language Server Protocol

The language server protocol consists of enabling communication between several IDE and language servers. Thus, the IDE is a client and the language server is a server. An IDE can interact with serveral servers at the same time.

The bellow sequence diagram present the start of the project

sequenceDiagram
    participant VSCode as VSCode (Client IDE)
    participant Pharo as Pharo (Language Server)
    activate VSCode
    VSCode->>Pharo: Start Pharo
    activate Pharo
    VSCode->>Pharo: What is the port of Language Server Protocol?
    Pharo-->>VSCode: It's 40520!
    activate Pharo
    VSCode->>+Pharo: Initialized?
    Pharo->>-VSCode: Initialized!
    VSCode->>+Pharo: This is my capabilities. What are yours?
    Pharo->>-VSCode: Capabilities!
    loop Client ask for feature | Example:
        VSCode->>+Pharo: What about completion?
        Pharo->>-VSCode: Complete this text with this snippet
    end
    VSCode->>Pharo: I'm done
    Pharo->>VSCode: OK bye!
    deactivate VSCode
    deactivate Pharo

In short, the client (IDE) opens the server. Then, it creates a connection with the client, for instance using a socket. Finally, client and server exchanges information following the protocol.

Pharo Language Server Installation

Installing the Pharo Language server is made easy thanks to a Pharo baseline. After downloading a Pharo 10 image, you can install the project using the following script:

1Metacello new
2  githubUser: 'badetitou' project: 'Pharo-LanguageServer' commitish: 'v3' path: 'src';
3  baseline: 'PharoLanguageServer';
4  load

To do so, after launching a Pharo image, open a playground with Ctrl+O+W, copy and paste the above code snippet, and execute it.

Pharo Language Server Structure

In the following, we describe the strcuture and the main classes of the Pharo Language Server project. We only describe the Pharo code and not the one that can be present in the client extension (used in the IDEs).

Package architecture

The all project is included inside the package PharoLanguageServer. The package is then split into tags.

• Uncategorized contains the core of the server
• Document and TonelUtils contain the pharo code representation for the Pharo implementation of the Language Server Protocol (i.e. if you want to code in Pharo from VSCode)
• Handler contains some override of JRPC Pharo implementation to ease its usage by the Pharo Language Server Protocol
• Structure- contains all the structure send and received by the server. Structures are grouped depending on the protocol they are related to (e.g. PLSDocumentSymbol is linked to retriving symbols using the protocol)

Server class architecture

The two main classes of interest are PLSAbstractServer and PLSServer. PLSAbstractServer contains all the logic of a server implementing the Language Server Protocol. PLSServer is the class implementing the Language Server Protocol in the case of using it for the Pharo programming language.

Each class follow the same coding convention/architecture for the method protocols.

• starting contains the main method start used when starting the server and the required method to resolved message that are sent and received.
• lsp- contains the methods to implement the protocol
• pls- contains extension of the protocol (e.g. configuring the debug mode thought the protocol, or accessing to new specific features).

Starting the server

In this subsection we give some technical description of the starting method of the server.

The server is started by calling the method PLSAbstractServer>>start

1PLSAbstractServer >> start
2    self debugMode ifFalse: [ PLSUIManager withPLSServer: self ].
3    self initializeStreams.
4    lastId := 0.
5    process := [ 
6        [ serverLoop ] whileTrue: [ 
7            | request |
8            request := self extractRequestFrom: clientInStream.
9            ('Request: ' , request) recordDebug.
10            self handleRequest: request toClient: clientOutStream ] ]
11            forkAt: Processor lowIOPriority
12            named: 'JRPC TCP connection'

First, if the debug mode is not activated, we update the UIManager of pharo to use ours instead. For now, it only enables opening popup in the IDEs instead of the Pharo image.

Then, we initialize the stream that will be used for the communication.

1initializeStreams
2    | tcpServer |
3    withStdIO ifTrue: [ 
4        clientInStream := Stdio stdin.
5        clientOutStream := Stdio stdout.
6        ^ self ].
7    tcpServer := Socket newTCP.
8    tcpServer listenOn: self port backlogSize: 10.
9    Stdio stdout nextPutAll: tcpServer port asString asByteArray.
10    Stdio stdout flush.
11    serverLoop := true.
12    (tcpServer waitForAcceptFor: 60) ifNotNil: [ :clientSocket | 
13        clientInStream := SocketStream on: clientSocket.
14        clientOutStream := clientInStream.
15        self
16            logMessage: 'Client connected to Server using socket'
17            ofType: PLSMessageType info ]

To do so, we first check if the server ask us to be connected using STDIO or, by default, with socket. For instance, STDIO is used with Eclipse IDE and socket with VSCode.

To configure a socket, we first create a socket with Pharo with an available port. Then, we send to the client IDE the port thanks to STDIO. When connected, we set the clientInStream and clientOutStream with the same socket (since it can be used safely to get and send data).

Once the streams are set, we can perform the main loop that:

1. extract the request from the in stream (extractRequestFrom:)
2. process and send the response to the out stream (handleRequest:toClient:)

1 [ serverLoop ] whileTrue: [ 
2            | request |
3            request := self extractRequestFrom: clientInStream.
4            ('Request: ' , request) recordDebug.
5            self handleRequest: request toClient: clientOutStream ]

Extract request

The extraction of the request is done in two steps. It is implemented in the method extractRequestFrom:.

1extractRequestFrom: stream
2
3    | length startingPoint endPoint result |
4    "data is the current buffer state"
5    length := -1.
6    [ length = -1 and: [ serverLoop ] ] whileTrue: [ 
7        [ data ifEmpty: [ data := (stream next: 25) asString ] ]
8            on: ConnectionTimedOut
9            do: [ self log: 'timeout but still work' ].
10        length := self extractLengthOf: data ].
11    startingPoint := data indexOf: ${.
12    endPoint := data findCloseBracesFor: startingPoint.
13    result := String new: length.
14    "three options"
15    "startingPoint and endPoint are found"
16    (startingPoint ~= 0 and: [ endPoint ~= 0 ]) ifTrue: [ 
17        result := data copyFrom: startingPoint to: endPoint.
18        data := data copyFrom: endPoint + 1 to: data size.
19        ^ result ].
20    startingPoint = 0
21        ifTrue: [ "none were found" 
22            self getDatafromPosition: 1 fromSocket: stream in: result ]
23        ifFalse: [ "only startingPoint is found"
24            (data copyFrom: startingPoint to: data size) withIndexDo: [ 
25                :each 
26                :index | result at: index put: each ].
27            self
28                getDatafromPosition: data size - startingPoint + 2
29                fromSocket: stream
30                in: result ].
31    data := ''.
32    ^ result

1. We extract the length of the complete request. This work is done by extractLengthOf: once we receive first data.
2. Then, based on the known length of incoming data, we extract the full request.

Handle request

Handling a request is also done in several steps. It is implemented in the method handleRequest:toClient:.

flowchart TD
    parseJSON(Check that the JSON is Correct)
    parseJSON --> B{Is it?}
    B ---->|No| E(Return Error)
    B -->|Yes| Dispatch(Dispatch to the method with the correct pragma)
    Dispatch --> Convert(Convert returned object to JSON)
    Convert --> Send(Send answer)

Method of the protocol

Every method of the protocol is implemented as a method in the server. Thoses methods are categorised in protocol following the Language Server Protocol specification. For example, the method relative to Language Server Protocol Hover are inside the protocol lsp - hover.

The Pharo method uses as pragma the requested remote method by the protocol. For example, for hover, it is the method textDocument/hover, so the implementation uses the pragma as follow.

1textDocumentHoverWithPosition: position textDocument: textDocument
2    <jrpc: #'textDocument/hover'>
3    self subclassResponsibility

The pharo method also implement at least as many arguments as the client can send. In the hover example, the specification declares that the sent object is the HoverParams:

1interface TextDocumentPositionParams {
2    /**
3     * The text document.
4     */
5    textDocument: TextDocumentIdentifier;
6
7    /**
8     * The position inside the text document.
9     */
10    position: Position;
11}
12
13
14export interface HoverParams extends TextDocumentPositionParams, WorkDoneProgressParams {
15}

the current version of the server ignore the WorkDoneProgressParams for now

So the method must at least accept attributes with the name: textDocument and position. If the pharo method has more available arguments, they will be filled with nil. If the incoming data have arguments in an incorrect order, the server will sort them first. If the incoming data have an unknow argument, it will be ignore.

The return of the Pharo method must be a PLS structure that implement the method asJRPCJSON. This method will convert the pharo object to be transmitted to the client.

For example, for the PLSServer (see snippet of code below). The dictionary and PLSHover implement asJRPCJSON.

1textDocumentHoverWithPosition: position textDocument: textDocument
2    <jrpc: #'textDocument/hover'>
3    | hover document |
4    document := (self context textItem: (textDocument at: #uri)). 
5    hover := PLSHover new
6        context: self context;
7        source: document;
8        position: position;
9        yourself.
10    ^ { #contents -> hover contents } asDictionary

Extending the Abstract Language Server to implement a new one