Null Pointer Micromanagement


    12.1.35 • Public • Published

    gnv logo


    Easily build standalone CLI, browser, and Node packages all in one workspace using the latest syntax.

    Build the smallest and fastest JS packages for multiple targets with no runtime dependencies and full backwards compatibility.

    All of your package's dependencies will be baked into the compiled output for each target, so package.json will contain no dependencies or devDependencies fields and npm install will only need to download what is in the dist/ directory.


    This package has no mandatory dependencies aside from Rollup and Closure Compiler (and related plugins). gnv create sets up GitHub repositories by default using hub, but this can be disabled with --no-github flag.

    Note: This package relies on POSIX commands.

    Linux & MacOS should work out of the box. Use Windows Subsystem for Linux if a real operating system is not available.


    It is recommended to install these.

    Package Description
    hub A CLI for managing GitHub repositories developed by GitHub. Install
    (Optional: Use --no-github flag to disable.)

    npm (global)

    To use gnv, first install it with:

    npm i -g gnv

    The gnv CLI will then be functional, but you will not actually be able to bundle anything without Closure Compiler and Rollup, as mentioned above. A convenience command is built into the CLI to save time, and you can install the necessary peer dependencies for gnv (or any gnv package) with:

    gnv get-peer-deps

    gnv get-peer-deps

    This will call npm i -g to install the needed global deps and then link them in the workspace with npm link. You could manually install and link the peer deps yourself, but this is much easier.

    I thought gnv packages don't require dependencies?

    Most compiled gnv projects will not have peer dependencies, but gnv itself relies on native Closure Compiler binaries, and it is all around more performant to just include these tools as peerDeps in this case. Rollup will be bundled into gnv, rather than added as a peerDep, soon (stability is the goal right now).


    You can create, build, and develop gnv workspaces with the gnv CLI.


    Excludes get-peer-deps.

    Package Description
    add [options] <pkgs...> Add the given packages as gnv development dependencies.
    build [options] Build this workspace and run tests when finished. Final output will be in dist/.
    clean Clean the gnv workspace.
    create [options] <name> Create a new gnv workspace and push to GitHub. Use <organization/name> to create for an organization. Requires hub package or -ng flag.
    develop Start developing, and rebuild dev bundles when changes are made to the workspace.
    install [options] [directory] Install all dependencies in [directory]/package.json. Defaults to working directory.
    publish [level] Publish this package to NPM using npm publish. Removes dev CLI from package.json to prevent installation issues and bumps the version by level. Defaults to patch.
    remove [options] <pkgs...> Remove the given packages from gnv development dependencies.
    test Run mocha tests.

    Workspace structure

    A gnv project contains:

    • a .gnv folder that contains configuration information and Rollup scripts
    • a lib/ directory for storing library code
    • an exports/ folder for exposing data to export targets

    The structure of the actual codebase generally looks like:

    ├── commands.js
    └── templates.js
    ├── cli.js
    ├── node.js
    └── universal.js

    Where bolded file indicates an executable file. If you need a concrete example, this project itself is a gnv workspace, and its skeleton is printed when the gnv command is called inside a gnv workspace, as seen in the gnv get-peer-deps example above.

    Think of lib/ as a sandbox - it is where all internal logic will go, and the exports/ folder is where endpoints are exposed for each target. Exports will typically import different things from lib/ and then expose different information for different targets at exports/[target].js.

    For instance, gnv's create() function in lib/commands.js is exposed to the CLI at exports/cli.js for initializing new projects at the command line. It is also exported for usage in Node programs at exports/node.js, and can then be required and imported in the form import { create } from 'gnv' in third party Node packages.

    Creating a new workspace

    Creating a new gnv workspace with gnv create [organization]/name will initialize a GitHub repository with a lib/ and an exports/ directory and install it to the current directory (as a submodule, if inside a git repo). gnv projects are like modules in that they have exports for different targets, which will create corresponding outputs in dev/ and dist/ directories during the build process.

    Building a workspace

    Building the workspace with gnv build will create development and distribution folders. The modules in dev are simply Rollup outputs, and they build very quickly as they do not resolve Node modules. The outputs in dist are the final compiled outputs of a project, and will resolve all third party dependencies and Closure Compile the outputs.

    Here is what the generated dev and dist directories look like:

    ├── cli.cjs
    ├── cli.mjs
    ├── node.cjs
    ├── node.mjs
    ├── universal.cjs
    └── universal.mjs
    ├── cli.cjs
    ├── cli.mjs
    ├── exe.js
    ├── node.cjs
    ├── node.mjs
    ├── universal.cjs
    └── universal.mjs

    Reserved build targets

    There are three reserved targets, which will build with special assumptions. For instance, cli.js and its compiled output will both be set to be executable. Non-reserved targets at exports/[target].js will all build with the same default settings.

    Target Filename Description
    exports/cli.js A CLI that is hooked into the bin field of package.json and will execute when your package is called from the command line.
    exports/node.js A module which will be exposed to Node, i.e. what what users will import or require from inside a NodeJS script.
    exports/universal.js A module which does not rely on any Node dependencies and can be run in any environment (browser, Node, etc.). Primarily for browser bundles and pure vanilla JS programs which will benefit from the maximum level of compression offered by the Closure Compiler. The executable (see below) will be generated from the universal target output.

    How does the CLI export work?

    The bin field of package.json points to dist/cli.cjs and uses the commander package by default to provide an interactive command line interface.

    gnv will generate cli.js with an example program when a new workspace is created with gnv create my-project-name, and it will also initialize a GitHub repository (or submodule, if you're in a git directory already) unless passed the -ng, --no-github or -ns, --no-submodule flags respectively.

    You will spend most of your time working running my-project-name-dev rather than my-project-name, as the development version of the CLI will run the source code in exports/, whereas the release version will point to the compiled output at dist/cli.cjs.

    CLI development

    If you want to test live changes to the exports/cli.js file, a development version of the CLI is linked to my-package-dev:

    $ my-package-dev
    Logging without rebuilding!

    Create a new project (linking to bin is done automatically):

    gnv create my-package # use --no-github to skip GitHub integration 

    Enter the directory and build the project:

    cd my-package && gnv build

    Call your built program (./dist/cli.cjs):

    $ my-package
    Hello world!

    Check exports/cli.js to see what's happening behind the scenes.

    How does the Node export work?

    Node will import or require the appropriate compiled export file (dist/node.mjs or dist/node.cjs respectively) in dist/. The following settings allow for this:

    • .mjs and .cjs extensions on built files for explicit specification of ES/CJS modules
    • "main": "dist/node.cjs" in package.json for pre-13.2.0 compatibility
    • "exports": { ... } in package.json to specify CJS and ESM export locations for post-13.2.0 import/require support
    • "type": "module in package.json so ES6 import/export are available by default, i.e. for inside lib/

    How does the universal export work, and what's exe.js?

    The executable output is generated from the universal target. While universal specifies exports that can be import'd and require'd, exe.js is the most compressed version of this module which exports nothing and aims to only produce the same side effects as the exports/universal.js, removing any unused code, mangling variable names, recursive inlining, and so on.

    For example:

    exports/universal.js ->

     * Expose data from `lib/`.
    export * from '../lib/TestAB.js';
    export * from '../lib/TestC.js';
    export { default as TestDefault } from '../lib/TestDefault.js';
    export { TEST_STRING } from '../lib/templates.js';
     * This is a side effect that won't get removed due to dead code elimination.
    const a = 10;
    console.log(`a is ${a}`);

    -> dist/universal.cjs

    All exports from universal are exposed.

    'use strict';var $jscomp=$jscomp||{};$jscomp.scope={};$jscomp.createTemplateTagFirstArg=function(b){return b.raw=b};$jscomp.createTemplateTagFirstArgWithRaw=function(b,c){b.raw=c;return b};Object.defineProperty(exports,"__esModule",{value:!0});var TestA=function(){console.log("Test A!")},TestB=function(){console.log("Test B!")},TestC=function(){console.log("Test C!")},TestDefault=function(){console.log("Default checking in!")},TEST_STRING="HELLO WORLD!",a=10;console.log("a is "+a);

    -> dist/exe.js

    All dead code is removed, only executed side effects are included. Full Closure Compiler optimization.

    console.log("a is 10");

    How do I use third party modules if I want to keep my package.json free of dependencies?

    We're used to doing yarn add [pkg] / npm install [pkg] when we need to use a new third party module in our project. However, since our outputs do not depend on anything, we want to avoid adding any standard dependencies to package.json. The only time dependencies are needed are if we're actively developing in the workspace, or running the dev version of the CLI, or otherwise executing source code directly (which will contain import and/or require statements). In this case, we can bootstrap our workspace and download all needed dependencies with gnv boot. The dev CLI source at exports/cli.js will then be functional and ready for debugging.

    How do I add/install development dependencies?

    Package Description
    gnv add pkg[@latest] Adds a package to package.json's gnvDependencies field.
    gnv add -P pkg[@latest] Adds a peer dependency to package.json's peerDependencies field.
    gnv boot Installs all gnvDependencies in the local node_modules/ without touching package.json using npm i --no-save, and installs all global peerDependencies with npm i -g --no-save, links all global peer dependencies into the local node_modules/ with npm link --no-save.

    Development dependencies for a gnv workspace can be added with gnv add my-dependency, which installs the package globally with npm i -g --no-save and then links in the workspace with npm link. It is also added to custom field in package.json called gnvDependencies, which will be installed along with peerDependencies on a call to gnv boot (which just calls a bootloader script that installs all of the needed development dependencies specified in package.json).

    This workflow prevents us from needing to re-install the same dependencies over and over in multiple folders, while assuring the dev build of a workspace freeze will be consistent over time. The one caveat to this is that gnv add pkg@1.0.0 does not accept semantic versioning, only explicit versions or tags (i.e. gnv add pkg@latest).

    You can always add dependencies locally rather than globally and ship the package as-is, if your priorities are different. gnv will not modify package.json except to bump the version and remove (then re-add) the development CLI from package.json's bin field on gnv publish.

    How will my exports work with require and import?

    The goal of gnv workspaces is full CJS/ESM interop, and require and import should both expose your exports as expected thanks to Rollup. This allows for distraction-free usage as seen below:

    ES6 | namedImportTest.mjs [PASSING]

    import { TestA, TestB, TestC, TestDefault } from '../dist/universal.mjs';

    CJS | namedImportTest.cjs [PASSING]

    const { TestA, TestB, TestC, TestDefault } = require('../dist/universal.cjs');

    An additional small twist that gnv adds is, because it explicitly only ever compiles exports, there will never really be a situation where the default export would be empty by default, so if a default export is not defined, all named exports are exported as the default export. This is a little ugly in the source, and, using the gnv project itself as an example, looks like:

    Generated development bundle at dev/node.mjs

    // expose as named exports
    export { callCompiler, checkInsideProject, compile, create, devCompile, develop, displayProjectInfo, initialize };
    // expose all as default export
    export default { callCompiler, checkInsideProject, compile, create, devCompile, develop, displayProjectInfo, initialize };

    This can be overridden by setting a default export manually. A large motivation of this is so that the form import pkg from ... can be used instead of import * as pkg from ..., and it allows for predictable coding patterns:

    ESM import

    // default (all)
    import gnv from './dev/node.mjs';
    // named
    import { callCompiler } from './dev/node.mjs';

    CJS require

    // default (all)
    const gnv = require('./dev/node.cjs');
    // named
    const { callCompiler } = require('./dev/node.cjs');

    Running in Node to compare outputs:

    $ node test.cjs

      gnv: {
        callCompiler: [AsyncFunction: callCompiler],
        checkInsideProject: [Function: checkInsideProject],
        compile: [AsyncFunction: compile],
        create: [AsyncFunction: create],
        devCompile: [AsyncFunction: devCompile],
        develop: [AsyncFunction: develop],
        displayProjectInfo: [Function: displayProjectInfo],
        initialize: [AsyncFunction: initialize]
      callCompiler: [AsyncFunction: callCompiler]

    $ node test.mjs

      gnv: {
        callCompiler: [AsyncFunction: callCompiler],
        checkInsideProject: [Function: checkInsideProject],
        compile: [AsyncFunction: compile],
        create: [AsyncFunction: create],
        devCompile: [AsyncFunction: devCompile],
        develop: [AsyncFunction: develop],
        displayProjectInfo: [Function: displayProjectInfo],
        initialize: [AsyncFunction: initialize]
      callCompiler: [AsyncFunction: callCompiler]

    Everything works as expected! The default and named exports for CJS and ESM outputs are functionally identical.

    Aren't global NPM installs bad?

    No. The most popular article explaining why it might be a bad idea to install packages globally offered the following reasoning:

    The obvious short answer is that your project depends on them. If your project depends on a package, it should be documented in package.json so that you can guarantee that it is installed when someone types npm install. Otherwise, you’ll need to add extra steps in your README file to inform anyone else who clones your project that they need to install each of your global dependencies as well.

    However, since our dependencies are rolled into the compiled output, this is no longer a concern, and we can save the user a step and ship them only the code that they will actually be executing. A call to gnv boot will allow a developer to install the global and local dependencies needed to run the source without adding them to package.json's standard dependency fields.


    npm i gnv

    DownloadsWeekly Downloads






    Unpacked Size

    676 kB

    Total Files


    Last publish


    • ctjlewis