@microsoft/ts-command-line

ts-command-line

This library helps you create professional command-line tools for Node.js. By "professional", we mean:

• no gotchas for users: Seems obvious, but try typing "npm install --save-dex" instead of "npm install --save-dev" sometime. The command seems to execute successfully, but it doesn't save anything! The misspelled flag was silently ignored. This lack of rigor plagues many familiar NodeJS tools and can be confusing and frustrating. For a great user experience, a command line tool should always be strict about its syntax.

• no gotchas for developers: Many command-line libraries store their parsed data in a simple JavaScript hash object. This is convenient for small projects. But suppose a large project has many different source files that define and read parameters. If you try to read data['output-dir'] when it wasn't defined, or if you misspell the key name, your tool will silently behave as if the parameter was omitted. And is data['max-count'] a string or a number? Hard to tell! We solve this by modeling each parameter kind as a real TypeScript class.

• automatic documentation: Some command-line libraries treat the --help docs as someone else's job. ts-command-line requires each every parameter to have a documentation string, and will automatically generate the --help docs for you. If you like to write long paragraphs, no problem -- they will be word-wrapped correctly. [golf clap]

• structure and extensibility: Instead of a simple function chain, ts-command-line provides a "scaffold" pattern that makes it easy to find and understand the command-line implementation for any tool project. The scaffold model is generally recommended, but there's also a "dynamic" model if you need it. See below for examples.

Internally, the implementation is based on argparse and the Python approach to command-lines. Compared to other libraries, ts-command-line doesn't provide zillions of custom syntaxes and bells and whistles. Instead it aims to be a simple, consistent, and professional solution for your command-line tool. Give it a try!

Some Terminology

Suppose that we want to parse a command-line like this:

widget --verbose push --force --max-count 123

In this example, we can identify the following components:

• The tool name in this example is widget. This is the name of your Node.js bin script.
• The parameters are --verbose, --force, and --max-count.
• The currently supported parameter kinds include: flag (i.e. boolean), integer, string, choice (i.e. enums), and string list.
• The value "123" is the argument for the --max-count integer parameter. (Flags don't have arguments, because their value is determined by whether the flag was provided or not.)
• Similar to Git's command-line, the push token is called an action. It acts as sub-command with its own unique set of parameters.
• The --verbose flag is a global parameter because it precedes the action name. It affects all actions.
• The --force flag is an action parameter because it comes after the action name. It only applies to that action.

Scaffold Model

If your tool uses the scaffold model, you will create subclasses of two abstract base classes: CommandLineParser for the overall command-line, and CommandLineAction for each action.

Continuing our example from above, suppose we want to start with a couple simple flags like this:

widget --verbose push --force

We could define our subclass for the "push" action like this:

class PushAction extends CommandLineAction {
  private _force: CommandLineFlagParameter;

  public constructor() {
    super({
      actionName: 'push',
      summary: 'Pushes a widget to the service',
      documentation: 'Your long description goes here.'
    });
  }

  protected onExecute(): Promise<void> { // abstract
    return BusinessLogic.doTheWork(this._force.value);
  }

  protected onDefineParameters(): void { // abstract
    this._force = this.defineFlagParameter({
      parameterLongName: '--force',
      parameterShortName: '-f',
      description: 'Push and overwrite any existing state'
    });
  }
}

Then we might define the parser subclass like this:

class WidgetCommandLine extends CommandLineParser {
  private _verbose: CommandLineFlagParameter;

  public constructor() {
    super({
      toolFilename: 'widget',
      toolDescription: 'The widget tool is really great.'
    });

    this.addAction(new PushAction());
  }

  protected onDefineParameters(): void { // abstract
    this._verbose = this.defineFlagParameter({
      parameterLongName: '--verbose',
      parameterShortName: '-v',
      description: 'Show extra logging detail'
    });
  }

  protected onExecute(): Promise<void> { // override
    BusinessLogic.configureLogger(this._verbose.value);
    return super.onExecute();
  }
}

To invoke the parser, the application entry point will do something like this:

const commandLine: WidgetCommandLine = new WidgetCommandLine();
commandLine.execute();

When we run widget --verbose push --force, the PushAction.onExecute() method will get invoked and then your business logic takes over.

Testing out the docs

If you invoke the tool as "widget --help", the docs are automatically generated:

usage: widget [-h] [-v] <command> ...

The widget tool is really great.

Positional arguments:
  <command>
    push         Pushes a widget to the service

Optional arguments:
  -h, --help     Show this help message and exit.
  -v, --verbose  Show extra logging detail

For detailed help about a specific command, use: widget <command> -h

For help about the push action, the user can type "widget push --help", which shows this output:

usage: widget push [-h] [-f]

Your long description goes here.

Optional arguments:
  -h, --help   Show this help message and exit.
  -f, --force  Push and overwrite any existing state

Dynamic Model

The action subclasses provide a simple, recognizable pattern that you can use across all your tooling projects. It's the generally recommended approach. However, there are some cases where we need to break out of the scaffold. For example:

• Actions or parameters may be discovered at runtime, e.g. from a config file
• The actions and their implementations may sometimes have very different structures

In this case, you can use the DynamicCommandLineAction and DynamicCommandLineParser classes which are not abstract (and not intended to be subclassed). Here's our above example rewritten for this model:

// Define the parser
const commandLineParser: DynamicCommandLineParser = new DynamicCommandLineParser({
  toolFilename: 'widget',
  toolDescription: 'The widget tool is really great.'
});
commandLineParser.defineFlagParameter({
  parameterLongName: '--verbose',
  parameterShortName: '-v',
  description: 'Show extra logging detail'
});

// Define the action
const action: DynamicCommandLineAction = new DynamicCommandLineAction({
  actionName: 'push',
  summary: 'Pushes a widget to the service',
  documentation: 'More detail about the "push" action'
});
commandLineParser.addAction(action);

action.defineFlagParameter({
  parameterLongName: '--force',
  parameterShortName: '-f',
  description: 'Push and overwrite any existing state'
});

// Parse the command line
commandLineParser.execute(process.argv).then(() => {
  console.log('The action is: ' + commandLineParser.selectedAction!.actionName);
  console.log('The force flag is: ' + action.getFlagParameter('--force').value);
});

You can also mix the two models. For example, we could augment the WidgetCommandLine from the original model by adding DynamicAction objects to it.

Further reading

The API reference has complete documentation for the library.

Here are some real world GitHub projects that illustrate different use cases for ts-command-line:

• @microsoft/rush
• @microsoft/api-extractor
• @microsoft/api-documenter