title: Using NPM layout: docs category: docs order: 12

Using NPM

Npm usage in Figwheel has changed significantly. For reference purposes the original version of this document can be found here

These instructions require org.clojure/clojurescript >= 1.10.773 and com.bhauman/figwheel-main >= 0.2.6.

Read the rest of this document and come back here for a quick reference.

Set the :target compiler option to :bundle. This will cause the compiler to emit an output file file that can be bundled a JavaScript bundler like webpack.

Optionally set the :bundle-cmd compiler option to

{:none ["npx" "webpack" "--mode=development" :output-to 
       "--output-path" :final-output-dir 
	   "--output-filename" :final-output-filename]}

to ensure the output file is bundled after a compile. Figwheel will fill in :output-to, :final-output-dir and :final-output-filename.

Your host page will need to load the final bundled asset.

You will want to make sure that the :output-to file is in the :output-dir directory so the bundler can resolve the assets it requires.

As a rule, if you are using NPM libraries along with the :bundle target, don't use :aot-cache true.

If your are porting a project over to use the :bundle target and NPM libs, use the :clean-outputs Figwheel option.

Relevant Figwheel Options

• :auto-bundle
• :bundle-freq
• :final-output-to

What?

NPM is a package repository for the JavaScript ecosystem. Almost all available JavaScript libraries are packaged, stored, and retrieved via NPM.

We want to use these libraries inside our ClojureScript codebase, but there is some friction because ClojureScript embraced the Google Closure Compiler and its method of declaring libraries, which is quite different than NPM's.

{::comment} We could get into a debate about why ClojureScript designers decided to embrace the less popular ecosystem, but that is largely academic at this point. I will say that the advantages of effortless interactive development via hot-reloading and the amazing capabilities of the Google Closure Compiler's advanced mode are direct benefits of using the GCC's (Google Closure Compiler's) method of defining modules via simple JavaScript object literals. In other words, without the GCC there would most likely not be a Figwheel.

Nevertheless, experiencing friction while importing libraries from the dominant JavaScript ecosystem is a very unfortunate trade-off. {:/comment}

However, with recent changes in the ClojureScript compiler (along with changes in Figwheel) it is now becoming much more straightforward to include NPM modules in your codebase.

The Overview

We are going to utilize a JavaScript bundler like Webpack to bundle up the output of our ClojureScript compiled code to produce a final bundled output file which will contain all the NPM libraries we have required in our ClojureScript code.

So the code is going to go through two steps:

1. compilation by the ClojureScript compiler to an intermediate file
2. bundled together with its NPM dependencies to a final output file that we will load into the browser

It's important to remember that the under the :bundle target the output of the ClojureScript compiler is not loadable by the browser it has to be bundled first.

During development under optimizations :none the bundled output file will only contain the NPM libraries and small amount of ClojureScript boot code that will in turn load our compiled ClojureScript code. So there exists a bundle file and then a bunch of individually compiled ClojureScript namespaces that the bundled boot script will need to load.

It's important to remember that during development the bundled file still depends on the other ClojureScript output files and will not work on its own.

When we deploy to production we will be compiling a single ClojureScript artifact using :simple or :advanced optimizations. This single ClojureScript artifact will then be bundled along with its NPM dependencies into a single bundle that can be deployed on its own.

Getting started with NPM libraries into your project

We are going to assume you are starting from the base example.

I'm going to use npm for this example but if you prefer yarn go ahead and use that. It doesn't really matter for this.

There are four steps that we are going to follow to add some libraries to our hello-world.core project.

1. Initialize NPM in our project by adding a package.json file
2. Install Webpack to bundle the needed dependencies into a single JS file.
3. Add the needed libraries as NPM dependencies.
4. Configure our ClojureScript build to use the bundle generated by Webpack.

Initialize NPM

We will need to initialize npm for our hello-world project.

One way to do this is to use the npm init command. You can of course just create the file from scratch but npm init is faster.

Make sure you are in the root directory of the project and execute:

$ npm init -y

This will create a package.json file in the root directory of your project along side your dev.cljs.edn file.

Install Webpack

We will need webpack to bundle our application along with our moment dependency to make it available to our ClojureScript code.

Install webpack and webpack-cli:

$ npm add --save-dev webpack webpack-cli

This add webpack and webpack-cli as development dependencies in your package.json. It will also download them to a node_modules directory in your project.

Add the needed libraries

Let's say we want to use the moment library in our application.

We'll use npm to add a moment dependency to our package.json file in the usual manner:

$ npm add moment

This should download and install the moment library along with its dependencies if it has any.

Configure our ClojureScript build

Everything we have done up until this point is very similar to what we would normally do if we were using NPM and Webpack for a simple JavaScript project.

In the dev.cljs.edn file we'll add the following config:

{:main hello-world.core
 :target :bundle
 :bundle-cmd {:none ["npx" "webpack" "--mode=development" :output-to 
                     "--output-path" :final-output-dir
					 "--output-filename" :final-output-filename]}}	

Understanding the above configuration is important, so I'm going to explain each part.

The :target compiler option is set to :bundle to instruct the ClojureScript compiler to produce an output file that can be bundled by a JavaScript bundler like Webpack.

The :bundle-cmd compiler option is set to

{:none ["npx" "webpack" "--mode=development" :output-to 
        "--output-path" :final-output-dir
		"--output-filename" :final-output-filename]}}

This provides the ClojureScript compiler with a command that it can use to bundle the intermediate output of the compiler into its final bundled form.

Figwheel adds some additional functionality to the :bundle-cmd. It interpolates the keywords :output-to, :final-ouitput-dir and :final-output-filename into the command. In this case the :output-to is going to be replaced by the default :output-to path ./target/public/cljs-out/dev/main.js. The :final-output-dir is replaced by the path part of :output-to which is ./target/public/cljs-out/dev/. The :final-output-filename defaults to the filename of of :output-to with a _bundle added before the extension or main_bundle.js.

Or stated more simply in this case:

• :output-to is replaced with ./target/public/cljs-out/dev/main.js
• :final-output-dir is replaced with ./target/public/cljs-out/dev
• :final-output-filename is replaced with main_bundle.js

If you supply your own :output-to cljs compiler option, it will be used instead of the default.

After the ClojureScript compiler is finished compiling it will call the :bundle-cmd to bundle up the output.

In this case it will call:

$ npx webpack --mode=development ./target/public/cljs-out/dev/main.js --output-path ./target/public/cljs-out/dev --output-filename main_bundle.js

This will bundle up the main.js file and pull in the moment dependency along with it.

Now let's modify our source file to use moment so that we can make sure that things are working.

Edit the src/hello_world/core.cljs to look like:

(ns hello-world.core
  (:require [moment]))

(js/console.log moment)
(println (str "Hello there it's "
              (.format (moment) "dddd")))

Run the build

OK now that we've setup everything up we can run the build.

$ clojure -m figwheel.main -b dev -r

The browser and REPL should launch as usual:

Except now you can see one additional line in the output which notifies us that the bundle command was called.

And if you look at the dev tools console of the browser window that just popped open you should see similar output printed as below verifying that we were able to use moment successfully.

Well we successfully used an NPM package from our ClojureScript code. Now you can npm add other JavaScript NPM packages and use them from ClojureScript

Simplifying with :auto-bundle

If you just need to get up and running quickly the :auto-bundle Figwheel option will set up all the default options that we configured above.

So this configuration is equivalent to the above configuration:

^{:auto-bundle :webpack}
{:main hello-world.core}

When enabled :auto-bundle will set :target to :bundle.

When choosing :webpack it will set :bundle-cmd to:

{:none ["npx" "webpack" "--mode=development" :output-to 
        "--output-path" :final-output-dir
		"--output-filename" :final-output-filename]
 :default ["npx" "webpack" "--mode=production" :output-to 
           "--output-path" :final-output-dir
		   "--output-filename" :final-output-filename]}

and when choosing :parcel it will set :bundle-cmd to:

{:none ["npx" "parcel" "build" :output-to
        "--out-dir" :final-output-dir
        "--out-file" :final-output-filename
        "--no-minify"]
 :default ["npx" "parcel" "build" :output-to
           "--out-dir" :final-output-dir
           "--out-file" :final-output-filename]}

These :bundle-cmd configurations are merged with any :bundle-cmd configurations in your .cljs.edn file so that you can override either the default ones.

:auto-bundle also adds an important config to :closure-defines in your compiler options when not using :optimizations :none.

It adds:

{:closure-defines {...
                   cljs.core/*global "window"}}

Configuration Tips

Don't specify :output-to

Don't specify the :output-to compiler option. Figwheel will take care of this for you and normally it's just an intermediate file. You are probably much more interested in the output of the bundler.

If you must specify where the output of the compiler is sent use :output-dir and Figwheel will specify an [:output-dir]/main.js file for you.

It's important to remember that all the output-files still have to be accessible from the browser (i.e. served by the web-server) when developing, as these files are not included in the output bundle.

Using :final-output-to

:final-output-to is a Figwheel option, NOT a ClojureScript compiler option. Place it in the metadata section of your build config.

If you don't want to use the default location and need to specify a specific location for your build's final bundled asset its probably best to supply a :final-output-to config option.

You don't need to specify :final-output-to if you are not using the built in Figwheel REPL host page, Extra-Mains, or Auto-testing.

However, its helpful for Figwheel to know where the final bundled asset of your build is located. If the location is known then Figwheel can provide you a REPL without having to create and index.html page. Figwheel can also munge the name of the :final-output-to to create bundles for Extra-Mains and Auto-testing. Both :final-output-dir and :final-output-filename are derived from :final-output-to.

Using :bundle-cmd

The :bundle-cmd compiler option is not required. It specifies what to do with the a compiled ClojureScript output file.

You can skip using :bundle-cmd entirely and this would require that you run the bundler manually AFTER your files have been initially compiled.

This can be unpleasant because it's rather nice to run a Figwheel build command and have the browser window pop open along with a running REPL.

When you omit the :bundle-cmd you will need to launch figwheel first. The browser will pop open and display your custom index.html application page but it will be broken because the bundled JavaScript isn't available yet. Next, you will need to run your bundler and reload the browser. At this point, everything should be up and running fine. However, this is not the best experience and why using the :bundle-cmd is helpful.

Another reason the :bundle-cmd is helpful is to provide Figwheel a template of a command that can create bundles for your build. If you create a :bundle-cmd with the keywords :output-to and :final-output-to, Figwheel will be able to reuse that command with different parameters to create slightly different bundles for things like Extra-Mains and Auto-testing.

When filling in the :bundle-cmd template Figwheel also replaces the :final-output-dir and :final-output-filename template keywords (this helps with commands that require them). It obtains these values from the :final-output-to value.

Figwheel by default only runs the :bundle-cmd after the first compile, this avoids the incurring the latency of bundling on every single file change. This can significantly slow down hot-reloading depending on your set up. You can change this with the :bundle-freq Figwheel option.

Using :bundle-freq

The :bundle-freq Figwheel option controls how often a the :bundle-cmd is called.

It has three settings :once, :always, and :smart:

• :once - bundles only once after the initial build compile. This helpful if you want to launch a webpack watch command on the command line after you start your build.
• :always - bundles after every ClojureScript compile. This is fine when you are starting out and bundle times are short. This however increases the total compile time and can slow down your hot-reload time.
• :smart - re-bundles only when your :output-to or npm_deps.js changes thus trying to only bundle when you have included a new node dependency in your CLJS code. This will cover most cases where you need to rebundle but is no where near a complete as launching a webpack watcher.

The default is :once which is the most conservative setting.

I think that using :smart is probably the best choice if you are not using a webpack watcher.

Keep in mind that a hot-reload will NOT pick up changes from a re-bundling. You have to refresh the browser to get those changes. So if you add a node dependency and it gets re-bundled into your JS bundle you won't see it until you reload the browser.

Using a webpack.config.js

It's important to note that you can still use a webpack.config.js to specify various configurations for your bundle. The CLI options simply override the configuration in the webpack.config.js file. If an input file is specified on the command line it will simply override any entry supplied in the Webpack config. If an -o is supplied it will any output supplied in the Webpack config as well.

This frees you to configure your Webpack bundle as you need.

How NPM support works

The motivation is that we want to have a JavaScript bundler process our ClojureScript output so that it can bring in and bundle all of our NPM dependencies. So we have two output files the output file from ClojureScript and the final bundled output from a JavaScript bundler like Webpack or Parcel.

When you use the :bundle target, the ClojureScript compiler does just this. We're going to examine how this is accomplished.

Let's look at an example default output file when you don't supply a :target option and thus get the default browser target output.

window.CLOSURE_UNCOMPILED_DEFINES = {"figwheel.repl.connect_url":"ws:\/\/localhost:9500\/figwheel-connect?fwprocess=29ee5f&fwbuild=devy"};
window.CLOSURE_NO_DEPS = true;
if(typeof goog == "undefined") document.write('<script src="/cljs-out/devy/goog/base.js"></script>');
document.write('<script src="/cljs-out/devy/goog/deps.js"></script>');
document.write('<script src="/cljs-out/devy/cljs_deps.js"></script>');
document.write('<script>if (typeof goog == "undefined") console.warn("ClojureScript could not load :main, did you forget to specify :asset-path?");</script>');
document.write('<script>goog.require("figwheel.core");</script>');
document.write('<script>goog.require("figwheel.main");</script>');
document.write('<script>goog.require("figwheel.repl.preload");</script>');
document.write('<script>goog.require("devtools.preload");</script>');
document.write('<script>goog.require("figwheel.main.system_exit");</script>');
document.write('<script>goog.require("process.env");</script>');
document.write('<script>goog.require("hello_world.core");</script>');

In the code above you can see the standard ClojureScript bootstrap code that loads your application into the browser (along with the figwheel preloads).

When we are working with NPM, we want the ClojureScript compiler to resolve all of our references to NPM libraries in our ns declarations and then add them into this file somehow so we can call webpack on it to import them.

For example when we have a namespace like above:

(ns hello-world.core
  (:require [moment]))

We want ClojureScript to make a note of that and then emit something that can bundled and that will process and include the moment NPM library.

With this in mind let's look at the main output-to file when we use the :bundle target.

import {npmDeps} from "./npm_deps.js";
window.CLOSURE_UNCOMPILED_DEFINES = {"figwheel.repl.connect_url":"ws:\/\/localhost:9500\/figwheel-connect?fwprocess=87057f&fwbuild=dev","cljs.core._STAR_target_STAR_":"bundle"};
window.CLOSURE_NO_DEPS = true;
if(typeof goog == "undefined") document.write('<script src="/cljs-out/dev/goog/base.js"></script>');
document.write('<script src="/cljs-out/dev/goog/deps.js"></script>');
document.write('<script src="/cljs-out/dev/cljs_deps.js"></script>');
document.write('<script>if (typeof goog == "undefined") console.warn("ClojureScript could not load :main, did you forget to specify :asset-path?");</script>');
document.write('<script>goog.require("figwheel.core");</script>');
document.write('<script>goog.require("figwheel.main");</script>');
document.write('<script>goog.require("figwheel.repl.preload");</script>');
document.write('<script>goog.require("devtools.preload");</script>');
document.write('<script>goog.require("figwheel.main.system_exit");</script>');
document.write('<script>goog.require("process.env");</script>');
document.write('<script>goog.require("hello_world.core");</script>');
window.require = function(lib) {
   return npmDeps[lib];
}

The first line is new. Let's look at it:

import {npmDeps} from "./npm_deps.js";

OK so we are importing an npm_deps.js file. Let's look at its contents as well:

module.exports = {
  npmDeps: {
    "moment": require('moment')  }
};

So, the npm_deps.js file is a JavaScript bundler ready file that was generated by the CLJS compiler. The compiler looked at all of our ns declarations and resolved the NPM libraries by looking at the node_modules directory.

Now when we run a bundler on our main.js output file the bundler will resolve the npm_deps.js file and then will resolve and include all NPM libs we are using.

It's important to note that the main.js output file is importing the ./npm_deps.js file relatively so both these files need to be in the same directory. ClojureScript will let you break this easily if you supply an :output-to option that isn't in the output directory.

Now let's look at the last lines of our main.js output file.

window.require = function(lib) {
   return npmDeps[lib];
}

These lines shim require so that it requiring a NPM library in ClojureScript works correctly.

Further is we look at the output of compiling the hello_world/core.cljs file you will see this:

// Compiled by ClojureScript 1.10.775 {:target :nodejs}
goog.provide('hello_world.core');
goog.require('cljs.core');
hello_world.core.node$module$moment = require('moment');
console.log(hello_world.core.node$module$moment);
cljs.core.println.call(null,["Hello there it's ",cljs.core.str.cljs$core$IFn$_invoke$arity$1(hello_world.core.node$module$moment.call(null).format("dddd"))].join(''));

//# sourceMappingURL=core.js.map

The interesting line is where a scoped local reference of the moment is created.

hello_world.core.node$module$moment = require('moment');

The :bundle target uses the current functionality of the :nodejs target which emits these requires. And while this works fine for Node we need to shim require to support it in the browser.

Keep in mind this is all only true when we are developing in :optimizations :none. When we want to create a single compiled artifact to deploy for production, using :simple or :advanced optimization mode ClojureScript will put together a single output file with the requires in it. Then the bundler will resolve and replace these top requires as it normally does.

Problems with cached compiled assets

Looking at the previous example if we hadn't been using the :bundle target had been using the Cljsjs cljsjs/moment library instead of the NPM based moment library the output of compiling hello_world/core.cljs would have instead been the following:

// Compiled by ClojureScript 1.10.775 {}
goog.provide('hello_world.core');
goog.require('cljs.core');
goog.require('moment');
hello_world.core.global$module$moment = goog.global["moment"];
console.log(hello_world.core.global$module$moment);
cljs.core.println.call(null,["Hello there it's ",cljs.core.str.cljs$core$IFn$_invoke$arity$1(hello_world.core.global$module$moment.call(null).format("dddd"))].join(''));

//# sourceMappingURL=core.js.map

The major difference is one the line:

hello_world.core.global$module$moment = goog.global["moment"];

which when compiled with an NPM moment library looks like this:

hello_world.core.node$module$moment = require('moment');

When transitioning from Cljsjs libraries to NPM based libraries the way that moment is required changes. This can become a problem while we are actively converting our code to use NPM libraries.

The caching of compiled code can wreak havoc on this process. The AOT caching that happens when you set :aot-cache true caches compiled libraries globally.

If you compile a version of reagent the depends on Cljsjs based react and it gets cached to the global AOT-cache, then the compiled reagent lib in the cache will have been specialized to use a Cljsjs based react.

If you then create a new project that has a reagent dependency that tries to meet its react dependency from NPM it will fail if you are using :aot-cache true. It will fail because you aren't providing the library via Cljsjs but the cached version of reagent is expecting it.

This same caching problem also occurs with code that's been compiled to your local target directory. A compiled reagent library could be specialized to either NPM or Cljsjs.

This becomes even more confusing if you are accidentally providing both the Cljsjs lib and the NPM lib. It's possible that your code will appear to work but may potentially be fetching the JavaScript library from the wrong source.

Bottom line

As a rule, if you are using NPM libraries along with the :bundle target don't use :aot-cache true. The default value of :aot-cache is false in Figwheel. However the default value of :aot-cache is true when you use cljs.main.

Also, if your are porting a project over to use the :bundle target and NPM libs, use the :clean-outputs Figwheel option. The clean outputs option will delete all of your compiled artifacts everytime you start a Figwheel build.

If you run into a situation where your Cljsjs lib or your NPM lib doesn't resolve you should look at the compiled output file and verify that the file is being required correctly as you may have a caching problem.

Compiling for production

When you want to compile your final output file you will use an :optimizations mode other than :none.

The thing to remember is that the ClojureScript compiler is going to outputing a single JavaScript file and then the bundler is going to resolve all of the JavaScript requires present in that file in order to ultimately create a JavaScript bundle that you can deploy.

This is different process to what happens during development (under :optimizations :none). During developement bundle only resolves the requires in the :output-to file and the npm_deps.js file.

This is an important difference because if there is an errant require in your code that is called conditionally it won't cause a problem during development because require is shimmed, however the bundle process is resolving requires statically. So if there is a (js/require ...) for a library that isn't in node_modules or is only available in the Nodejs runtime (but not in the browser) your bundler will throw an error about a missing library.

Keeping this in mind we should remember to not dynamically require NPM libraries that are only available on the backend, or are not currently stored in node_modules.

When compiling for production we should also, make sure that you aren't including Cljsjs libraries that are being provided for by NPM libraries. We don't want to be including libraries twice.

For example, we need to exclude react dependencies from reagent to prevent bundling to copies of react into your final application.

An example of excluding react from reagent:

{:deps {org.clojure/clojurescript {:mvn/version "1.10.773"}
        reagent {:mvn/version "0.10.0" :exclusions [cljsjs/react cljsjs/react-dom]}}}

We also need to make sure we include cljs.core/*global* "window" in our :closure-defines (I.e.:closure-defines {cljs.core/*global* "window" ...}). This is required when we aren't compiling in :optimizations :none. Figwheel adds this automatically when using the :auto-bundle option.

Troubleshooting

Bad bundle command

The most likely thing that will happen is you will have a bad :bundle-cmd.

• Please check that the command that is logged is the command that you expect.
• comment out the :bundle-cmd option, run the figwheel build and then run the webpack command from your terminal/shell environment to see what errors are showing up

The relationship between the :output-to file and npm_deps.js

The :output-to file emitted by the :bundle target imports the npm_deps.js file. This can only work when the :output-to file is in the same directory as the npm_deps.js file. I.E. the :output-to file has to be in the :output-dir for the build.

{::comment} TODO

• advanced
• talk about the combination of bundling :once and the launching a watcher
• maybe discouraging the :output-to is a bad idea, perhaps just encourage the "_bundle" convention. {/:comment}