kernel-output

Kernel Output

Render kernel messages in an OuputArea.

• Code structure
• Initializing a Kernel Session
• OutputArea and Model
• Asynchronous Extension Initialization
• Make it Run

In this example, you will see how to initialize a kernel, execute code and display the rendered output. The OutputArea class used by JupyterLab to render an output area under a notebook cell or in the console will be reused.

Essentially, OutputArea will render the data that comes as a reply to an execute message sent to an underlying kernel. Under the hood, the OutputArea and the OutputAreaModel classes act similarly to the KernelView and KernelModel classes defined in the Kernel Messaging example.

Code structure

The code is split into two parts:

1. the JupyterLab plugin that activates all the extension components and connects them to the main JupyterLab application via commands, launcher and menu items,
2. a panel that contains the extension logic and UI elements to interact with it.

The first part is contained in the index.ts file and the second in panel.ts.

In the following sections, the logic will be first described. It is followed by the creation of the visual element.

Initializing a Kernel Session

To interact with a kernel, you can create a SessionContext object (see the documentation). Here it is stored in the private _sessionContext variable:

// src/panel.ts#L86-L86

private _sessionContext: SessionContext;

A SessionContext handles a single kernel session. The session itself (not yet the kernel) is started with these lines:

// src/panel.ts#L36-L40

this._sessionContext = new SessionContext({
  sessionManager: manager.sessions,
  specsManager: manager.kernelspecs,
  name: 'Kernel Output'
});

The private session variable is exposed as read-only for other users through a getter method:

// src/panel.ts#L64-L66

get session(): ISessionContext {
  return this._sessionContext;
}

Once you have created a session, the associated kernel can be initialized with this line:

// src/panel.ts#L50-L61

void this._sessionContext
  .initialize()
  .then(async value => {
    if (value) {
      await sessionContextDialogs.selectKernel(this._sessionContext);
    }
  })
  .catch(reason => {
    console.error(
      `Failed to initialize the session in ExamplePanel.\n${reason}`
    );
  });

When a session has no predefined preferred kernel, a dialog will request the user to choose a kernel to start. Conveniently, this can also be an already existing kernel, as you will see later.

The following two methods ensure the clean disposal of the session when you close the panel.

// src/panel.ts#L68-L71

dispose(): void {
  this._sessionContext.dispose();
  super.dispose();
}

// src/panel.ts#L81-L84

protected onCloseRequest(msg: Message): void {
  super.onCloseRequest(msg);
  this.dispose();
}

OutputArea and Model

The SimplifiedOutputArea class is a Widget, as described in the widget example. It has the ability to display the results of a notebook cell execution. You can instantiate it with a new OutputAreaModel; this is class containing the data to show:

// src/panel.ts#L42-L46

this._outputareamodel = new OutputAreaModel();
this._outputarea = new SimplifiedOutputArea({
  model: this._outputareamodel,
  rendermime: rendermime
});

SimplifiedOutputArea provides a static method execute that sends some code to a kernel through a ISessionContext (see documentation). And then it displays the result in the specific SimplifiedOutputArea object you created:

// src/panel.ts#L73-L79

execute(code: string): void {
  SimplifiedOutputArea.execute(code, this._outputarea, this._sessionContext)
    .then((msg: KernelMessage.IExecuteReplyMsg) => {
      console.log(msg);
    })
    .catch(reason => console.error(reason));
}

The SimplifiedOutputArea.execute function receives at some point a response message from the kernel which says that the code was executed (this message does not contain the data that is displayed). When this message is received, .then is executed and prints this message to the console.

To display the SimplifiedOutputArea Widget you need to add it to your panel with:

// src/panel.ts#L48-L48

this.addWidget(this._outputarea);

The last step is to add the panel to the JupyterLab main area.

Asynchronous Extension Initialization

index.ts contains the code to initialize the extension.

First, it is a good practice to unify the extension commands into one namespace at the top of the file:

// src/index.ts#L21-L25

namespace CommandIDs {
  export const create = 'kernel-output:create';

  export const execute = 'kernel-output:execute';
}

You can then add the commands to the palette and the menu by iterating on a list:

// src/index.ts#L105-L109

// add items in command palette and menu
[CommandIDs.create, CommandIDs.execute].forEach(command => {
  palette.addItem({ command, category });
  exampleMenu.addItem({ command });
});

To create a new client session, the service manager must be obtained from the JupyterLab application:

// src/index.ts#L54-L54

const manager = app.serviceManager;

To launch the panel, you need to wait for the service manager to be ready. Then once the panel is created and its session is ready, it can be added to the JupyterLab main area:

// src/index.ts#L58-L69

let panel: ExamplePanel;

/**
 * Creates a example panel.
 *
 * @returns The panel
 */
async function createPanel(): Promise<ExamplePanel> {
  panel = new ExamplePanel(manager, rendermime);
  shell.add(panel, 'main');
  return panel;
}

Make it Run

To make the output area useful, you will request the user to enter a statement to be executed by the kernel. Then you will send it to your panel for execution and display:

// src/index.ts#L83-L103

commands.addCommand(CommandIDs.execute, {
  label: 'Contact Kernel and Execute Code',
  caption: 'Contact Kernel and Execute Code',
  execute: async () => {
    // Create the panel if it does not exist
    if (!panel) {
      await createPanel();
    }
    // Prompt the user about the statement to be executed
    const input = await InputDialog.getText({
      title: 'Code to execute',
      okLabel: 'Execute',
      placeholder: 'Statement to execute'
    });
    // Execute the statement
    if (input.button.accept) {
      const code = input.value;
      panel.execute(code);
    }
  }
});

Any advanced output can be displayed by the SimplifiedOutputArea widget, from raw text to interactive ipywidgets. In the GIF below, it is assumed you have a variable, named df, in your python kernel that contains a pandas dataframe. By entering the variable name in the input dialog, it will be displayed in the example panel.

Where to Go Next

This example makes use of input dialogs. To know which input dialogs are available have a look at the documentation.