After spending a few years locked into an ancient tech stack for a project, I'm finally getting a chance to explore more current frameworks and dip a toe into React and Webpack. So far, it's for the most part been a refreshing and enjoyable experience, but I've run across some difficulty with Webpack that I'm hoping the hive-mind can help resolve.
I've been poring over the Webpack 2.0 docs and have searched SO pretty exhaustively, and come up short (I've only turned up this question, which is close, but I'm not sure it applies). This lack of information out there makes me think that I'm looking at one of two scenarios:
What I'm trying to do is insane.
What I'm trying to do is elementary to the point that it should be a no-brainer.
...and yet, here I am.
The short version of what I'm looking for is a way to exclude certain JSX components from being included in the Webpack build/bundle, based on environment variables. At present, regardless of what the dependency tree should look like based on what's being imported from the entry js forward, Webpack appears to be wanting to bundle everything in the project folder.
I kind of assume this is default behavior, because it makes some sense -- an application may need components in states other than the initial state. In this case, though, our 'application' is completely stateless.
To give some background, here are some cursory requirements for the project:
Project contains a bucket of components which can be assembled into a page and given a theme based on a config file.
Each of these components contains its own modular CSS, written in SASS.
Each exported page is static, and the bundle actually gets removed from the export directory. (Yes, React is a bit of overkill if the project ends at simply rendering a series of single, static, pages. A future state of the project will include a UI on top of this for which React should be a pretty good fit, however -- so best to have these components written in JSX now.)
CSS is pulled from the bundle using extract-text-webpack-plugin - no styles are inlined directly on elements in the final export, and this is not a requirement that can change.
As part of the page theme information, SASS variables are set which are then used by the SASS for each of the JSX components.
Only SASS variables for the JSX components referenced in a given page's config file are compiled and passed through sass-loader for use when compiling the CSS.
Here's where things break down:
Let's say I have 5 JSX components, conveniently titled component_1, component_2, component_3, and so on. Each of these has a matching .scss file with which it is associated, included in the following manner:
import React from 'react';
import styles from './styles.scss';
module.exports = React.createClass({
propTypes: {
foo: React.PropTypes.string.isRequired,
},
render: function () {
return (
<section className={`myClass`}>
<Stuff {...props} />
</section>
);
},
});
Now let's say that we have two pages:
page_1 contains component_1 and component_2
page_2 contains component_3, component_4, and component_5
These pages both are built using a common layout component that decides which blocks to use in a manner like this:
return (
<html lang='en'>
<body>
{this.props.components.map((object, i) => {
const Block = component_templates[object.component_name];
return <Block key={i}{...PAGE_DATA.components[i]} />;
})}
</body>
</html>
);
The above iterates through an object containing my required JSX components (for a given page), which is for now created in the following manner:
load_components() {
let component_templates = {};
let get_component = function(component_name) {
return require(`../path/to/components/${component_name}/template.jsx`);
}
for (let i = 0; i < this.included_components.length; i++) {
let component = this.included_components[i];
component_templates[component] = get_component(component);
}
return component_templates;
}
So the general flow is:
Collect a list of included components from the page config.
Create an object that performs a require for each such component, and stores the result using the component name as a key.
Step through this object and include each of these JSX components into the layout.
So, if I just follow the dependency tree, this should all work fine. However, Webpack is attempting to include all of our components, regardless of what the layout is actually loading. Due to the fact that I'm only loading SASS variables for the components that are actually used on a page, this leads to Webpack throwing undefined variable errors when the sass-loader attempts to process the SASS files associated with the unused modules.
Something to note here: The static page renders just fine, and even works in Webpack's dev server... I just get a load of errors and Webpack throws a fit.
My initial thought is that the solution for this is to be found in the configuration for the loader I'm using for the JSX files, and maybe I just needed to tell the loader what not to load, if Webpack was trying to load everything. I'm using `babel-loader for that:
{ test: /\.jsx?$/,
loader: 'babel-loader',
exclude: [
'./path/to/components/component_1/',
],
query: { presets: ['es2015', 'react'] }
},
The exclude entry there is new, and does not appear to work.
So that's kind of a novella, but I wanted to err on the side of too much information, as opposed to being scant. Am I missing something simple, or trying to do something crazy? Both?
How can I get unused components to not be processed by Webpack?
TL;DR
Don't use an expression in require and use multiple entry points, one for each page.
Detailed answer
Webpack appears to be wanting to bundle everything in the project folder.
That is not the case, webpack only includes what you import, which is determined statically. In fact many people that are new to webpack are confused at first that webpack doesn't just include the entire project. But in your case you have hit kind of an edge case. The problem lies in the way you're using require, specifically this function:
let get_component = function(component_name) {
return require(`../path/to/components/${component_name}/template.jsx`);
}
You're trying to import a component based on the argument to the function. How is webpack supposed to know at compile time which components should be included? Well, webpack doesn't do any program flow analysis, and therefore the only possibility is to include all possible components that match the expression.
You're sort of lucky that webpack allows you to do that in the first place, because passing just a variable to require will fail. For example:
function requireExpression(component) {
return require(component);
}
requireExpression('./components/a/template.jsx');
Will give you this warning at compile time even though it is easy to see what component should be required (and later fails at run-time):
Critical dependency: the request of a dependency is an expression
But because webpack doesn't do program flow analysis it doesn't see that, and even if it did, you could potentially use that function anywhere even with user input and that's essentially a lost cause.
For more details see require with expression of the official docs.
Webpack is attempting to include all of our components, regardless of what the layout is actually loading.
Now that you know why webpack requires all the components, it's also important to understand why your idea is not going to work out, and frankly, overly complicated.
If I understood you correctly, you have a multi page application where each page should get a separate bundle, which only contains the necessary components. But what you're really doing is using only the needed components at run-time, so technically you have a bundle with all the pages in it but you're only using one of them, which would be perfectly fine for a single page application (SPA). And somehow you want webpack to know which one you're using, which it could only know by running it, so it definitely can't know that at compile time.
The root of the problem is that you're making a decision at run-time (in the execution of the program), instead this should be done at compile time. The solution is to use multiple entry points as shown in Entry Points - Multi Page Application. All you need to do is create each page individually and import what you actually need for it, no fancy dynamic imports. So you need to configure webpack that each entry point is a standalone bundle/page and it will generate them with their associated name (see also output.filename):
entry: {
pageOne: './src/pageOne.jsx',
pageTwo: './src/pageTwo.jsx',
// ...
},
output: {
filename: '[name].bundle.js'
}
It's that simple, and it even makes the build process easier, as it automatically generates pageOne.bundle.js, pageTwo.bundle.js and so on.
As a final remark: Try not to be too smart with dynamic imports (pretty much avoid using expressions in imports completely), but if you decide to make a single page application and you only want to load what's necessary for the current page you should read Code Splitting - Using import() and Code Splitting - Using require.ensure, and use something like react-router.
Related
TL;DR: How does svelte handle importing the same script/functions multiple times in several components?
Let's say I have a svelte-project using a script "magic.js" which exports a function "greatFunction" which is huge and has some dependencies. I want to use this function in several components like "Profile.svelte", "Project.svelte", ...
I want to write as little code as possible and I want to have the compiled built as lean as possible.
I see the possibility of importing the "greatFunction" right into main.js, passing it as prop to App.svelte, then passing it to "Profile.svelte" and to "Project.svelte" and so on:
<Profile greatFunction={greatFunction} />
But what if I import magic.js and its greatFunction several times on a component-level in "Profile.svelte" and "Project.svelte" again and again?
Will this bloat my final production built?
What's the common way to do this right?
Are there any significant difference between these two syntax?
Vue.component('base-table', () => import('./components/BaseTable.vue'))
Vue.component('base-table', require('./components/BaseTable.vue').default);
Is one or the other affects the loading performance of an application?
From the Webpack docs:
require
Synchronously retrieve the exports from another module. The compiler will ensure that the dependency is available in the output bundle
import()
Dynamically load modules. Calls to import() are treated as split points, meaning the requested module and its children are split out into a separate chunk.
Simply 1st puts your BaseTable.vue component into separate JS bundle (file) which will be requested by a browser at the moment component is rendered 1st time. Good for components/parts of app which are not needed immediately (per-route code splitting) or only for some users (admins for example). Your base bundle size is smaller and thus parsed faster by the browser -> better initial loading performance...
Declaimer: Commenting from Vue/Webpack POV. I know nothing about Laravel except it's PHP thing...
I am trying to optimize a RequireJS project first for HTTP/2 and SPDY, and secondly for HTTP/1.x. Rather than concatenating all of my RequireJS modules into one big file, I am hacking r.js to resave the modules that would have been concatenated as separate files. This will be better for the new protocols, because each file will be cached individually, and a change to one file won't invalidate the caches of all the other files. For the old protocols, I will pay a penalty in additional HTTP requests, but I can at least avoid subsequent rounds of requests by appending revision hashes to my files (i.e. "module.js" becomes "module.89abcdef.js") and setting a long max-age on them, so I want to do that.
r.js adds names to previously-anonymous modules. That means that the file "module.js" with this definition:
define(function () { ... })
Becomes this:
define("module.js", function () { ... })
I want this, because I will be loading some of my modules via <script> elements (to prevent RequireJS HTTP requests acting as a bottleneck to dependency resolution; I can instead utilize r.js's topological sorting to generate <script> elements which the browser can start loading as soon as it parses them). However, I also plan to load some modules dynamically (i.e., via RequireJS's programmatic insertion of <script> elements). This means RequireJS may at some point request a module via a URL.
Here is the problem I have run into: I am generating a RequireJS map configuration object to map my modules' original module IDs to their revision-hashed module IDs, so that when RequireJS loads a module dynamically, it requests the revision-hashed file name. But the modules it loads, which are now named modules thanks to r.js, have the un-revisioned name of the module, because I don't revision the files until after I minify them, which I only do after r.js generates them (thus r.js doesn't know about the revisions).
So the file "module.89abcdef.js" contains the module:
define("module.js", function () { ... })
When these named modules execute, define will register a module with the wrong name. When other modules require "module.js", this name will be mapped to "module.89abcdef.js" (thanks to the map configuration), but since a module with that name was not actually defined after the onload event of a <script> with the src attribute "module.89abcdef.js" fired, the module they receive is undefined.
To be faithful to the new file name, the file "module.89abcdef.js" should probably instead contain the definition:
define("module.89abcdef.js", function () { ... })
So, I believe the solution to my problem will be to replace the unrevisioned name of the module in the r.js-generated file with the revisioned name. However, this means the hash of the file is now invalidated. My question: In this situation, does it matter that the hash of the file is invalid?
I think it's impossible that the module name in the source code could be faithful to the file name, because if the module name had a hash, well, then the hash of the file would be different, so the module name would become the new hash, and so on recursively.
I fear there might be a case where the unfaithful hash would cause the file to still be cached even when it'd actually changed, but I cannot think of one. It seems that appending the same hash for the file's original contents to the file's name and the module's name should always still result in a unique file.
(My lengthy explanation is an attempt to eliminate certain "workaround"-type answers. If I weren't loading modules dynamically, then I wouldn't need to use map and I wouldn't have a file-name-vs-module-name mismatch; but I do load modules dynamically, so that workaround won't work. I also believe I would not need to revision files if I were only serving clients on newer protocols [I could just use ETags as there would be no additional overhead to make the comparison], but I do serve clients on the older protocols, so that workaround doesn't work either. And finally, I don't want to run r.js after revisioning files, because I only want to revision after minifying the files, in case minification maps two sources to the same output, in which case I could avoid an unnecessary cache invalidation.)
Problem
This is how the app behaves:
every folder corresponds to a dynamic "feature" in the app.
features are independent of each other. (Inside each folder, none of the files have dependencies on other folders. However, each folder may use the same common libraries like jquery.)
not all features are needed all the time
within each folder, the files are not necessarily dependent on each other (but it makes sense to bundle them together because that's how a "feature" works)
the number of features is not fixed ... so a manual process where you have to manually update various configurations/files each time you add/remove a feature is NOT ideal
Desired Solution
Given what I described, what seems to make sense is to create chunks for each feature/folder and then load each chunk asynchronously (on demand) as needed by the app.
The problem is that this doesn't seem easy with webpack. I'd like to be able to do this
require.ensure([], function() {
var implementation = require('features/*/'+feature);
//...
where each feature (folder) goes in one chunk.
Unfortunately, if I'm understanding require.context correctly, that's not possible.
An alternative?
A possible alternative is the following:
use a single config file that lists each feature (we actually already use this),
use the config file to auto-generate a module that requires each folder asynchronously. (Basically, the "module" is a factory that the app can use to load each feature.)
add an entry point to the auto-generated module (force the chunks to be built).
So to summarize, add a build step that generates a factory that asynchronously load features. The problem is that this seems unnecessarily complex and I'm wondering if there's an better way to implement the solution. Or am I on the wrong track? is there a completely different solution?
Code
Folder structure
features
sl.graded.equation
graded.equations.js
graded.equation.edit.js
graded.equation.support.js
sl.griddable
griddable.js
griddable.edit.js
//...
Config file
{
"sl.graded.equation": [ //basically lists the feature "entry points" (not every file)
"sl.graded.equation",
"sl.graded.equation.edit"
],
"sl.griddable": [
"sl.griddable",
"sl.griddable.edit"
]
//...
}
Auto-generated factory
module.exports = function(feature, callback) {
switch(feature) {
//we can get fancy and return a promise/stream
case 'sl.graded.equation':
require.ensure([], function() {
var a = require('sl.graded.equation');
var b = require('sl.graded.equation.edit');
callback(a,b);
}
//...
I have a main page that uses ajax to load subpages and one of these subpages contains a google map, and so it loads the google maps api using the <script> tag:
<script type="text/javascript" src="http://maps.googleapis.com/maps/api/js?key=YOUR_API_KEY&sensor=SET_TO_TRUE_OR_FALSE">
I noticed that this loads a bunch of css and js files into both my main page and subpage. When I click on a different link in my main page, I want to be able to unload all of these files and remove any js objects that were created, i.e., clean up everything and return to the original state. Is there any way to do this?
The answer to your question is actually a bit more complicated than you might think. A good question and set of answers that deal with many of the related details are at: What is the Proper Way to Destroy a Map Instance?.
I'm not sure from your question, but it sounds like you may have created a page that loads the Google Maps API more than one time (or could, depending on user choices) and you should avoid that entirely. Google admits there are memory leak bugs associated with reloading the map and strongly recommends against multiple map reloads. Google essentially does not support multiple map load use cases.
Check out some of the information available at the question link included above; it contains some good discussion and information.
EDIT in Response to #Engineer's Comment:
Check out the Google Maps API Office Hours May 9 2012 Video where Chris Broadfoot and Luke Mahe from Google discuss: that they don't support use cases that involve reloading the map, that the API is intended to be loaded only once, and their acknowledgement that there is a memory leak bug. Set the playback to ~12:50 to view the section about destroying the map, problems with reloading the map, and suggestions they offer to avoid problems. Primarily, if you must hide and then show a map, they recommend reusing a single map instance.
Old question.. But here is my solution:
// First getting rid of the google.maps object (to avoid memory leaks)
// Then, we are also removing google-maps related script tags we can identify.
// After unloaded, if maps is reloaded more than once on the same page;
// we'll also get a warning in the console saying: "Warning: you have included the
// Google Maps API multiple times on this page. This may cause unexpected errors."
// This script will also avoid that warning.
if (window.google !== undefined && google.maps !== undefined) {
delete google.maps;
$('script').each(function () {
if (this.src.indexOf('googleapis.com/maps') >= 0
|| this.src.indexOf('maps.gstatic.com') >= 0
|| this.src.indexOf('earthbuilder.googleapis.com') >= 0) {
// console.log('removed', this.src);
$(this).remove();
}
});
}
Update: Note that this is not a full-proof solution. There might be copied/cloned/referenced objects. A better way would be to sandbox the map in an iframe and remove the iframe from DOM.
Not the way you're thinking. The easiest way to accomplish this would be to use an iframe to load the "heavy" parts of your application. Then when you get rid of the iframe you get rid of the CSS and JS associated with the map.
In version 2 the Google Maps API had a GUnload() call but the version 3 API doesn't have this call.
with vanilla JavaScript:
if (window.google?.maps) {
delete google.maps
document.querySelectorAll("script").forEach((script) => {
if (
script.src.includes("googleapis.com/maps") ||
script.src.includes("maps.gstatic.com") ||
script.src.includes("earthbuilder.googleapis.com")
) {
script.remove()
}
})
}