Install NPM dependencies that run in browser without Browserify, Webpack
github.comHTTP2 solves the loading multiple dependencies in parallel part of the equation, but as far as I can see, the browser would still have to download each dependency before they can resolve their transitive dependencies, so wouldn't this suffer latency issues due to requiring as many sequential client-server roundtrips as the max depth of your dependency tree?
SystemJS & JSPM has been around for quite a while now, and the goals are similar to this project (to enable loading of any NPM dependency as native ES modules in the browser). They solve the roundtrip problem using an import map that's generated by JSPM to pre-resolve all transitive dependencies, so they can be loaded by the browser up front with maximum parallelism.
However, this project specifically calls out the lack of a need for an import map in its description as an advantage, so I'm wondering if they have some alternative solution to this problem.
Also somewhat related: the project maintainers for SystemJS and JSPM also provide a CDN at https://jspm.io/ serving jspm packaged ES modules for all npm packages, so you can load them as native ES modules with a module script tag or native dynamic import() call in supported browsers. But they specifically mention that this CDN is meant for quick prototyping/experimentation, and not suitable for production use due to latency issues, and recommend bundling workflows for production.
I've been following the SystemJS and JSPM projects for quite a while now, and I feel it's a much better solution for module loading than webpack and the likes that hack together their own non-interoperable proprietary module loader rather than supporting and advancing the official ES module loading standard. But unfortunately that's where the community is at, and SystemJS and JSPM has never been able to reach that critical mass of essential dev tooling that webpack & others have enjoyed for quite some time (the major road block I ran into the last time I tried it was the lack of a good hot module replacement story for development targeting the latest SystemJS 2.0 environment).
> the browser would still have to download each dependency before they can resolve their transitive dependencies.
The endgame of that train of thought is something like https://mariusgundersen.net/module-pusher/
Indeed, and it can even be done with HTTP/1: https://jacquesmattheij.com/the-several-million-dollar-bug/
tl;dr: a smart server can start pushing dependencies as soon as the request for a module comes in.
Even with HTTP/1, the browser will handle a response that came before the matching request.
This has been the biggest thing holding back wide adoption of native ES modules: every single one of your dependencies has had to be a proper, browser-compatible module. Very exciting.
I'm very curious how you could possibly avoid code bloat with this. Installing hundreds of megabytes of dependencies is fine server side, but won't this approach lead to massive JS downloads?
Yeah, bundlers have dead code analysis and throw some of the unused code out.
You can look for packages that have zero dependencies :)
This is what I wished for when I began using Web Components, and it is super awesome +1
The author was on the Polymer core team.
Wow, they claim that this is equal to or better than most apps that are bundled into a single bundle file due to the cache benefits!
I never would have guessed they could manage that and keep it this easy to use!
How common is the es module entry point in major packages?
If I understand correctly, they say this is only the case when you have less than 1000 flies:
Our app is certainly over that, and I suspect most apps quickly grow to this (most of the files are dependencies). I'm not sure where the 50 moderate number comes from (we have an unfortunate dearth of information when it comes to "real world test cases" -- it seems common for bundlers to just check whether it performs better for the app they were working on at the time), but I would be surprised if most the 50 file threshold is maintained for very long for highly dynamic sites (vs., say, a company brand site or a blog where this may be the case forever? Again, I honestly have no idea).@pika/web's installation most closely matches the study's moderate, "50 file" bundling strategy. Jung's post found that for HTTP/2, "differences among concatenation levels below 1000 [small files] (50, 6 or 1) were negligible.""Our app is certainly over that [1000 files]...I suspect most apps quickly grow to this"
I've not had a need to do much JavaScript since the days when you just directly used a few external libraries via <script>, so this just sounds nuts to me.
I suppose it's probably common, and the various packers make it work fine, but wow.
The current JS ecosystem leans heavily on bundlers and packers to enable this kind of workflow, and IMO it's a good thing that we do.
For example, lodash (a somewhat common set of utility functions and general helpers) is a single package, but internally it's broken up into hundreds of individual packages which lean on the dependency module system to resolve among themselves, deduplicate, and enable the ability to include only a few of them (and their dependencies) if you only need those in a project.
Many libraries are setup like this, because there's not much of a reason not to. Adding complicated functionality to an existing package is often worse for everyone than just creating a new package which implements what you want and setting it as a dependency.
As a counterpoint: I have a reasonably complex app that gets data via WebSocket pub/sub and dashboards it, does things like stream logs from remote devices, open up SSH consoles, draw canvas diagram with network map etc. It’s split up into lots of component files and I think I still have less than 50 files. So no, I would not say 1000 file app is normal (which is not to say they are not doing something grand that justifies it).
With my app size I can get away with no Webpack, just a Makefile that links together required file list into a template file, which is then included into the page template on the backend.
https://www.pikapkg.com/ will show you which packages export modules.
There is apparently more than 40.000 packages that are es module ready, that is quite a lot, more than I imagined!
This is brilliant for my efforts to just get started with THREE.js. I gave up due to module dependencies first time around, now I can actually get over that hurdle and get on to what I had planned to do with THREE.js maybe to do things the proper, optimised way with Browserify once I have some feasibility study done on my localhost where I don't care about it being more than a rough prototype.
Easiest way is to just use the pre-built version of THREE, and throw it on your page via script tag include.
Then it’s available in the window.THREE object, it’s not the “most correct” way but it’s great for prototyping
Thanks for your tip, I tried that but then I wanted to use some lathe type thing to make a shape and got into module hell. I am trying to not distract myself with having fun with that right now as I only want to make an 'undocumented feature' along the lines of embedding a flight simulator in a spreadsheet program, but it pleases me that I can get on it later to cook something fun up.
Does this work with commonjs modules or not? There's so few tc39 modules it's almost not worth considering otherwise.
It does!
> By default, @pika/web will install all package.json dependencies with an ES "module" entrypoint to the web_modules/ directory. @pika/web is able to handle any legacy Common.js/Node.js transitive dependencies in your dependency tree, however the top-level dependencies (the ones that you import directly) must have a "module" entrypoint.
I was excited but it looks like it'll break Typescript's module resolution. I'm sure someone will come up with something in time, but this is the same story over and over with web development this side of 2016. Nothing works together.
This is why we need a JS standard library...
Way to take advantage of HTTP/2. Really cool idea, and at a glance it looks like a reasonable implementation. Stoked to give this a try! :)
Sadly, there is people that can't use yet http/2