Skip to content

Latest commit

 

History

History
431 lines (270 loc) · 30.6 KB

Getting-Started-frontend.md

File metadata and controls

431 lines (270 loc) · 30.6 KB

Modern Frontend Web Development Guide

Author: Greg Slepak

This is a guide for web designers and developers, useful for general understanding, and is also tailored specifically to how we're using these techniques in Group Income (Simple Edition).

This guide assumes you know how to use git and GitHub!

If you already fully grok modern web dev and want to get started immediately, skip ahead to Frontend Workflow.

And if you'd like to help out with this project, get in touch!

"A Pox On Modern Web Development!"

Web development used to be simple. Back in the day, all that was needed to make a website was some HTML, CSS, and maybe some Photoshop or a sprinkling of "JavaScript". If you wanted to get fancy, maybe some PHP.

Those days are long over.

Today, web development has mostly forgotten about HTML and CSS. Instead, it's SASS this and JSX that. We've got "Components", and people don't even believe in HTML files anymore. Web frameworks seem to drop out of the sky as if it were christmas every day... What's a seasoned web designer to do?

Don't worry, friend! This document will explain how we got here, and by the end of it you'll understand why modern web development is the way it is and how to pretend as if you didn't care about HTML files either!

What Happened To Web Development?

In a word, apps. From its inception the "web" and "apps" were on a collision course, and most of us didn't realize it. Apps increasingly needed to talk to various web services and would increasingly store much of their data online.

Meanwhile, the tools that were used to create websites became more advanced as the web became more vital and companies began competing to attract visitors using the latest in fancy web technology. This technology was used for one primary purpose: creating user interfaces. And it became really good at it.

At some point the web collectively realized that the web standards known as HTML/CSS/JavaScript (and web browsers) had advanced to such a point that they were now beginning to seriously compete with applications. At the same time, application developers were looking at these tools and discovered that they could use them to create applications just as well!

Moreover and most critically, these fantastic UI/UX tools had AMAZING cross-platform support! A breakthrough occurred through the union of two technologies: Node.js and the web browser. Node.js gave JavaScript access to the Desktop OS itself, and thus the web merged with apps.

Various cross-platform frameworks started popping up promising to slash development time and costs while maintaining a high-quality experience across all platforms, even mobile devices!

And thus applications like Slack and Atom were born.

With this came radical shifts in how web development was done. Now it was necessary to "get organized!" Applications needed to run tens, even hundreds of thousands of lines of JavaScript, tossing around HTML and CSS all over the place, and do this all super-efficiently. These New Apps wanted UIs that could be loaded over any modern web browser over the Internet, as well as "Apps" that were just bundled versions of those websites. They also no longer wanted to have to rewrite the wheel for everything, so that led to the rise of Components—reusable chunks of HTML/CSS/JavaScript for doing one thing (like displaying a photo album) that could be dropped in to any project as easy as: <photoalbum></photoalbum>

React and Vue.js are examples of frameworks that support the creation of such components.

Rollup, Browserify, and Webpack are module systems and bundlers for efficiently organizing and loading the various resources these New Apps need.

The rise of the "single page app" (SPA)

To those new to "modern web development", one of the quirkiest aspects of it is its use (or more accurately, non-use) of HTML files.

Indeed, the modern web app uses just a single index.html file for the entire "website"!

If you visit one of these "single-page app" sites and view-source, you'll see practically nothing, just a mostly empty HTML file that can look like this:

<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="utf-8">
    <title>Site</title>
  </head>
  <body>
    <div id="app"></div>
    <script src="static/build.js"></script>
  </body>
</html>

The browser, meanwhile, will show a bunch of stuff that doesn't match that HTML!

What's going on?

<script src="static/build.js"></script>

This line loads a "javascript bundle", a file that will load everything else that's necessary to display the website, and the JavaScript will be used to asynchronously load HTML, CSS, image files, etc. and swap out the existing document structure for the new stuff.

You can see what the loaded HTML actually is by using your browser's web dev tools (you can bring them up by right-clicking on a portion of the page and choosing Inspect Element).

When links are clicked, or a new "page" is visited, a piece of JavaScript called a client-side router will do several things:

  • It will load the requested document/component/page asynchronously using an AJAX (XMLHttpRequest aka XHR)
  • It will swap out the old HTML and javascript for the new HTML and JavaScript
  • If necessary, it will programmatically change the website's location bar to display a new URL to give the appearance of a normal "page visit"!
  • It will also programmatically update the browser's history so that the back/forward buttons work normally as user's expect.

Client-side routers are the secret sauce behind SPAs, and while most "modern frontend web frameworks" will ship with their own, you can find plenty of standalone ones, or even code your own.

Why??

Two main reasons:

  1. Managing web servers is a PITA.
  2. Web apps can become huge, and bundlers like rollup/browserify/webpack can make it simpler to manage dependencies. At the same time, many of these tools and frameworks (especially webpack and anything more complicated than Vue.js), add lots of totally unnecessary complexity that 95% of front-end developers don't need, especially in an HTTP2 world.

It used to be that servers would render HTML (using a server-side templating language and/or programming language like PHP) for each page that is visited. Everyone remembers putting this in their HTML (right?):

<? include($_SERVER['DOCUMENT_ROOT'] . '/includes/header.php'); ?>

Well, as convenient as that was (to avoiding repeating the same HTML all over the place), it required there be a server that did all the heavy-lifting, assembling each one of these pages and sending the HTML back to clients. That's problematic for several reasons:

  • It puts more load on the server for something that clients (web browsers) could do themselves
  • More importantly, it requires that someone run and maintain one of these PHP servers

Well, with client-side rendering and single-page apps, you can now put all of that logic into the client-side javascript and ship most of your website, except the database (essentially) as static files that can be hosted anywhere! Just toss them onto a CDN (or multiple CDNs!) and be done with it!

The database portion continues to live on the server and now just provides a RESTful API.

Pros/Cons

Cons

  • Some people might find this setup confusing and difficult to work with as it requires a lot of machinery (like Browserify/Webpack).
  • Only some search engines support indexing such websites web apps, because they require JavaScript to display the page.
  • Similarly, won't work for users who disable JavaScript.

Pros

  • Reduces reliance on servers. This is a Good Thing™ both from a performance standpoint as well as a general Internet POV. Now the server just acts as an API and does little else.
  • You can distribute your entire website/app as static files! Just chuck your index.html and app.js file (and any other static resources) onto any CDN (even IPFS!) and you're good to go. No need to worry about managing a server! You can even create a purely P2P app this way with "no server"!
  • For large web apps it makes it much easier to dynamically load (and unload) only those resources/components/widgets that the application currently needs.

Architecture Stack Overview

These are the technologies this project uses. You don't need to understand everything about them, but you must at least read the descriptions below to understand what they are. They are listed in order of lowest-level to highest-level:

Node.js - JavaScript Environment

Node.js is the platform that all the other tech we use builds upon. It's an environment for running javascript outside of the browser, and it has been used to do almost everything, from creating web servers, to creating tools and utilities, to powering standalone applications like Slack.

NPM - Node Package Manager

Used for all kinds of JavaScript projects (backend and/or frontend). Any sort of module/library/framework you can think of is stored here. Yes, even jQuery. ;)

Grunt - Task Runner

For doing things like:

  • Setting up a local development environment with servers to view your website, power a backend API, and automatically refreshing your browser anytime files are modified
  • Watching files for changes and performing actions automatically (like syntax-checking/"linting" them, combining them, copying them, etc.)
  • Converting files from one format into another (bundling .vue files into .js, etc.)
Hapi.js - Node.js-powered HTTP(S) Server

Powers the backend API that the frontend talks to. It's responsible for managing the database of groups and users, and together with the frontend it uses cookies to authenticate users.

Rollup - Tree-Shaking Bundler

Note: previously we used browserify, but have since switched to Rollup because vueify is no longer being supported, and rollup is a far more advanced bundler that uses tree-shaking to reduce the size of bundles. What follows below is the previous text that applied to browserify, but note that even browserify transforms are supported by rollup if you use the rollup-plugin-browserify-transform.

Makes it possible to write front-end code in the same way Node.js backend code is written. Instead of including a library like jQuery using a <script> tag, you npm install jquery --save it, and then var $ = require('jquery') it. Browserify is much more than that however, it also:

  • Is designed be extended by "plugins" and "transforms"
  • The transforms make it possible to require any file format in JavaScript (like.vue files!) and bundle it all together into a single javascript bundle.
  • It can act as a pre-processor, so that when files are read in their contents can be transformed in the resulting JS bundle. This includes obvious things like minifying JavaScript, but also can include string-replacement, etc.
Babel - Tomorrow's JavaScript Features—Today!

The JavaScript language is always evolving. Oftentimes new features will be "solidified" by the standards bodies but won't yet be available in either web browser or Node.js environments. That's where Babel comes in. It lets you use tomorrow's by automatically rewriting files with new syntax into older syntax that browsers (and/or node.js) support.

  • Babel can be used through a variety of tools, including Browserify! Vue.js templates automatically use Babel, so you can use the latest JS features within them.
  • Like Browserify, Babel also supports plugins. In fact we use one called transform-inline-environment-variables to automatically replace environment variables like process.env.NODE_ENV with their corresponding values (both on the frontend and the backend).
Vue.js - Modern Frontend Component Framework

A frontend web framework that takes some inspiration from React, but is simply better in many ways (TLDR: just as powerful and far simpler):

In the section What Vue.js is good for (and not) we discuss its role and when/where/how to use it.

Riot.js is another (very similar) fantastic framework for working with single-page-apps (SPAs). Both are great, I just came across Vue.js first.

Honorable Mentions: Webpack, Gulp, React, Riot.js, Rollup

We've chosen (for now at least) to use Browserify over Webpack, Grunt over Gulp, and Vue.js over React, even though these are all perfectly fine tools in the modern web development toolkit. Briefly, our reasoning:

  • Although Webpack has features that Browserify does not, it is also far more difficult to get up and running with, and I don't think it provides web-based implementations of Node.js's API the way Browserify does. For now at least, Browserify seems to be serving our needs just fine, and it's unlikely that we'll switch away from it.
  • The reasons for choosing Vue.js over React are mentioned above.
  • Grunt and Gulp appear to be about evenly matched, and although Gulp seems to be hip and has a nifty design, Grunt is also perfectly well designed too and works just as well. We were already experienced with Grunt so we went with it. Plus, we like its emphasis on configuration over code and its native support for template strings. Either tool will serve you well.
  • Riot.js is very similar to Vue.js, perhaps even more minimalist. I might have used it instead had I come across it first. Both are great!
  • (Update: we've now switched to Rollup! 😂) Rollup is an extremely efficient bundler that takes advantage of ES6 modules to bundle only the functions you use and their dependencies (instead of entire modules). Because of this it produces smaller bundles than either Browserify or Webpack (currently). Webpack 2 plans to also implement the tooling necessary to do this, and you can use Rollup with Browserify via the rollupify transform.

Frontend Workflow

For those new to "modern web development", we have a section above to quickly bring you up to speed, and you should read that first: "A Pox On Modern Web Development!"

How do I get set up / just run the site?

The instructions below are for *NIX systems like OS X and Linux.

Pre-reqs:

  1. Install node.js if you haven't already. On a Mac the recommended way to install Node is via Homebrew via: brew install node
  2. Install the grunt command line tool via: npm install -g grunt-cli

Once you have npm, and grunt, clone this repo using git, and within the group-income-simple directory run these two commands:

npm install
grunt dev

If all went well you should be able to visit: http://localhost:8000


What does npm install do?

It installs the dependencies that this project relies on and places them into the node_modules/ folder. The list of dependencies is retrieved from the package.json file, and a list of the exact versions installed is stored in the package-lock.json file. We've set up grunt to verify you always have the latest dependencies installed, but you do need to run this command at least once.


What does grunt dev do?

A lot of things. Its output will tell you exactly what it does, but the general idea is:

  • It runs the backend API server whose "start" file is: backend/index.js
  • It creates a dist/ folder and copies and compiles files from within the frontend/ folder into it
  • It launches a web server to let you see the site
  • It watches files for changes, recompiling them and refreshing the browser anytime most files in frontend/ are changed, and re-running the backend API server whenever files in backend/ are changed

What files do I edit?

The frontend/ folder contains the frontend code.

Here are the important files and folders within frontend/:

  • index.html - This is the "entry point" for the frontend and the only HTML file in the entire web app. Everything starts with this file being served. It loads the dist/app.js bundle that gets created by Browserify, and that file contains the contents of all of the .js, .vue, and .ejs files.
    • Note: This file is purposefully minimal! You will rarely need to modify this file. If we modify it, it might only be to make it even smaller by removing any unnecessary <script> and <link> tags within it. This file is the "header and footer" that gets applied to every "page" of our web app, and for that reason it's best to keep it as short as necessary.
  • main.js - This is the entry point for all of the frontend code, where our modern web app starts. This file is responsible for setting up the Vue.js vue-router, the client-side machinery that replicates the effect of having "different pages with different URLs" (modern web dev trickery that unbelievably isn't as ridiculous as it sounds). It is also responsible for loading all of the other "pages" (located in views/) and code (located in js/) that the website has.
    • Note: If you want to add a "new page" to the site (i.e. a new link in the top <nav> bar), then you will need to modify this file to update the routes.
  • views/ - This contains all of the interface (pages and components) of the website. These are stored as .vue files.
  • utils/ - Folder for placing "handy js that's used in lots of places" that you wrote (not third-party, that goes in assets/vendor/).
  • assets/ - Grunt simply copies the folders within here to the output directory: dist/. This is where you drop in CSS files, images, and any other static assets.
    • Note: Currently we do not translate, optimize, minify, or fingerprint any of these assets, but we will (using grunt). We won't, however, be "bundling" these assets via Webpack or rollup. "Bundling" refers to the practice of concatenating js, css (sometimes even images!) and putting into a single .js file. This might have made sense in the HTTP/1.0 days, however HTTP/2.0 makes this bizarro practice totally unnecessary.

How do I add a new page to the website?

  1. Create either a new .vue or .ejs file within simple/views
  2. Link to it using the router-link directive, and if necessary update the index.html file (which currently has a global header/footer with page links)
  3. Update the vue-router mapping configuration in simple/main.js

If you're adding an .ejs file, you might add the following to the router config:

path: '/example-page',
component: { template: wrap(require('./views/test.ejs')) },
meta: {
  title: 'Example Page'
}

The wrap function wraps the content of the file in a tag (div by default) in order to prevent fragment instances (which aren't allowed in Vue.js 2.0).

If you're adding a .vue file, you could add:

path: '/example-page',
component: ExamplePage,
meta: {
  title: 'Example Page'
}

How does the Logical layer (Model) of the website works?

Read Information-Flow.md.


Where should I put non-JavaScript assets like CSS, images, etc.?

If they're your assets:

Just place them into the appropriate folder(s) within frontend/assets. You might need to re-run grunt dev and refresh the page (our gruntfile currently doesn't watch for asset changes). Grunt will copy them into dist/.

Then load them like normal (so frontend/assets/images/bitcoin.png is loaded as <img src="/assets/images/bitcoin.png">).

If they're someone else's (e.g. a CSS framework):

Place them in assets/vendor.

Please add only uncompressed/unminified assets to this folder (if possible) as that makes debugging simpler. We'll add minification later on (via grunt).


Where should I put JavaScript someone else wrote (e.g. jQuery)?

Recall that single-page-apps (SPAs) often have a single global JavaScript bundle that gets loaded. In our setup, this file is stored in dist/simple/app.js.

There are two ways to include third-party code:

  1. "Asynchronously" / "Lazily" / "On Demand": This is the preferred approach as it prevents app.js from getting bloated.
  2. "Globally": This is through the standard use of require. Anything that's require'd gets placed into the app.js bundle (unless you're using webpack's fancy-but-unnecessary code-splitting features). Unless a third-party library is used so frequently that it makes sense to have it in app.js, you should prefer the lazy-load approach.

We cover both approaches below.

Method 1.1: Lazy-loading a "vendor" lib using Script2

There are a variety of ways to load code asynchronously. We're going to ignore all of the fancy new techniques that have come along for doing this (that includes RequireJS-style AMD files, code-splitting, "System.import", and whatever else they think of next).

Instead, my strong recommendation after evaluating all of those … interesting alternatives, is to simply use the good old-fashioned <script> tag that everyone is already familiar with, unless there's a very compelling reason not to. This is possible in SPAs (even in frameworks that don't officially support it!) thanks to simple libraries like VueScript2. You'll waste less time, your website will load faster, and your designer will thank you.

Example: Loading jQuery only on specific "pages"/routes

First, grab the latest version of a library you want to include from from a trusted source on npm. For example, here's how we did this with jQuery:

npm install jquery --save

You can check the version that we're using by looking inside package.json. Updating to the latest version is done by explicitly requesting it:

npm install jquery@latest --save

Next we symlink it into our vendor/ folder:

$ cd frontend/simple/assets/vendor
$ ln -s ../../../../node_modules/jquery/dist/jquery.js

The symlink will make it simple for us to stay up-to-date with the latest version of the library. grunt dev will copy the actual file (not the symlink) into the dist/ folder. We've symlink'd the un-minified version of the library to make debugging easier, and because we can use grunt to minify it when creating production builds.

Next, just load jQuery in your template like normal:

<script src="/simple/vendor/jquery.js"></script>
<script> /* do something with $ */ </script>

This will work in .vue files thanks to the script2ify browserify transform.

Behind the scenes, this <script> tag is transformed into a VueScript2 Vue.js component that injects and loads the scripts. Using VueScript2 prevents jQuery from being bundled, and will load it only when the user visits a route that requests it (and only if it's not already loaded). By default, scripts are loaded one at a time in the order they appear on the page. You can add an async attribute to the <script> tag to have it be injected immediately without waiting for others.

Method 1.2: Lazy-load an entire route

You can lazy-load an entire route (i.e. an entire "page") by leveraging the code-splitting capabilities of browserify or webpack.

Method 2: Globally including a commonly used third-party library

Any use of require (without code splitting) will result in that module's direct inclusion within the app.js bundle.

  • In .vue files, require can be used within the <script> section (not to be confused with any VueScript2 <script> tags in the <template> section, see note below).

NOTE: require cannot be used within inlined <script> VueScript2 tags since that code is not parsed when the bundle is created, but at "runtime" when it's injected into a page. However, code within VueScript2 tags can access global variables that were require'd elsewhere.

Please ask before adding any new project dependencies! (Either in our chat, on the forums, or via a GitHub issue.)


Where's the best place to put JavaScript that I create?

  • If the JavaScript is specific to a page you're working on: put it directly into the .vue files.
  • Otherwise, for JS that's used across multiple files, place it into a .js file within frontend/utils and then import it within a .vue file

My web app's app.js bundle is huge, how do I break it up?

See Methods 1.1. and 1.2 in Where should I put JavaScript someone else wrote (e.g. jQuery)?.


My question isn't listed here!

That's OK! Ask us in the chat or the forums or open a GitHub issue! :)

Appendix

What Vue.js is good for (and not)

  • Reducing HTML markup and making collaboration easier. To open a facebook-style chatbox at the bottom of the window, you just insert <chatbox param1="foo" param2="bar"></chatbox>, etc.
  • Synchronizing data between the view (the widget) and the model (the JS object). I.e. two-way data bindings. Rule of thumb: if data-binding is unnecessary then there's no need for Vue.js to be involved.
  • Creating logic-heavy widgets for widget-heavy "apps" like Slack, etc.
  • Serving "web 2.0" (2.5?) static websites. I.e. you put your entire modern website in an .html file, have it call an API, and serve it to people over a CDN. Thanks to client-side routers the URL will magically change on "page visit". 👈 This might be the most relevant part for us as it works well with the future move to Ethereum.

It's not convenient for much else. It is especially not good for designers who are used to plain-old markup. Apparently, being friendly toward designers is one of its strong points, but that's compared to other frameworks like React.js / Angular / Ember. Nor is it useful for creating relatively "small" widgets like custom input fields or buttons.

Useful Links