Hacker News new | comments | show | ask | jobs | submit login
Show HN: Create a personalized and optimized webpack configuration (jakoblind.no)
92 points by znaky 89 days ago | hide | past | web | favorite | 45 comments

This is a cute tool, but the real solution to this problem is to avoid Webpack altogether.

Parcel accomplishes everything Webpack does with literally ~0~ configuration. If you make the switch, it will be the most productive 20 minutes of your life. (Yes, I am bitter over the MANY hours of my life I've wasted fiddling with Webpack.)


Parcel is great, but what you're saying is just not true.

Webpack's configuration gives it a great deal of flexibility for working with many different types of project.

I love parcel, I seem to personally have constant issues with my typescript react project related to breakpoints just completely not working, or when being triggered, mapping to completely other files.

I imagine it's a sourcemap issue I may post on the tracker, has anyone seen this themselves?

Wouldn't it make more sense to have a sort of "Spring Boot" [1] approach for Webpack, whereby a dependency like React (or a "webpack-react-starter") can list the features it needs to have enabled and Webpack can then automatically enable features based on that?

Tooling like this seems a bit backward, in the sense that you're now creating a tool to create configuration for a tool that runs other tools that~~stack overflow

[1] https://docs.spring.io/spring-boot/docs/current/reference/ht...

Nice! Although you might want to update this for Webpack 4.

Like for example, `ModuleConcatenationPlugin` no longer goes in `config.plugins`. It's now a boolean flag `plugins.concatenateModules`

This is everything you need to know about front end development in one comment

Why file loader and not URL loader?

To be fair this question is coming from a guy with a 4 day old stack overflow question that essentially amounts to "how the fuck do I use URL loader?"

From file-loader's README:

>Instructs webpack to emit the required object as file and to return its public URL

From url-loader's README:

>The url-loader works like the file-loader, but can return a DataURL if the file is smaller than a byte limit.

They are interchangeable, do the same thing, but url-loader can generate a dataUrl if the file is small, like an png/svg icon, so you can save one request where the latency could be higher than the download time.

I think this feature should be built into file-loader and toggled via loader option.

Oh, I read the docs start to finish, I just can't get the DataURLs to work lol. You seem like you know what you're talking about, so if you wanna take a crack at it... https://stackoverflow.com/questions/49934116/404s-on-some-im...

The fact that "lodash-webpack-plugin" and similar exists seems like a symptom of a larger problem. Why isn't tree shaking [1] good enough?

1. https://webpack.js.org/guides/tree-shaking/

Great stuff!. Specially at the beginning, webpack is intimidating and can turn complex quickly. Wish I had this when I began. Another add-on I would recommend is the addition of the: - target: 'node'

for those developer that are using webpack on the backend.

And encourage yarn instead of npm for both stability, sane defaults, and performance. :) Small detail though.

SASS (specifically SCSS) support would be good - AFAIK it's the most popular preprocessor and has been for a few years.

This is pretty nice. It's funny how all this time on Gulp/Grunt/Npm and a billion others, and finally we have something that seems to be staying still...

Now for the feature requests:

* Generally I have webpack.common.js for the common between all environments and webpack.config.js for dev (so the configuration file is named the default thing) and webpack.prod.js for production.

* Typescript checkbox :)

I was thinking about something similar. Making the webpack configuration accessible using a drag‘n‘drop editor, where you can define environments (prod, test, etc) and build configuration per environment using building blocks. JavaScript generation, JSON export (for imports), sharing, etc.

Never had the time to actually start it, though.

This is great! Would be great to see an option for Typescript, and splitting vendor and application bundles.

make = webpack, this tool = autotools

We are going around in a cycle!

I was just thinking about the 'full circle' today.

OSes of the past: Windows, Variations of Unix's,

OSes of 2018: Android, IOS, All Varations of Browsers, plus all the old ones


Cross platform (cross-os) development of the past: nasty (Widget libraries (wxWindows, QT, ... ), os-abstraction libs )

Cross platform development of 2018: nasty still (react-native, xamarin, etc .. )


Development pipeline tools of the past: Automake/cmake, make, linker

Development pipeline tools of 2018: npm, webpack


Compiler front ends of the past (things that produce assembler or low-level C or stack-machine codes): various C++ front ends, Eifel, etc

Compiler front ends of 2018 (things that spit out javascript): babel (more like macro-assembler), Clojurescript, ReasonML, Fay, Nim .. ,etc


In current times, we also added Wireframe/Styling tools into our development pipeline (I do not think we had it in the past). Eg Bootstrap and other styling (that are not integrated into 'compile-time-verification' stage of the modern Front-end languages.

I realize that above is somewhat front-end-dev oriented, but I think the desire to share 'same language' for front and backend development, causes the blur.

I also would add that for data crunching and data distribution (eg CORBA, DRPC) of the past -- we have gained more options, higher quality and free. So I would say this is a material advance for our productivity.

Your comparison is broken. You may as well compare apples to oranges, with the difference being potatoes.

Your phrase is broken. You can easily compare apples to oranges.

I know about the phrase I just felt like giving an equally vague answer to your comment.

To stay on topic what makes you think that the comparison is flawed?

What makes webpack uncomparable to make?

How can android / browsers not be compared to windows / linux?

Why can't make be used to pack web applications?


The difference being that Make is actually usable.

For the silent down-voters: Tell me why the above statement is not true.

It's a punctuation heavy format with a bunch of shell code mixed in, which means projects go from using one language (Python, Ruby, JavaScript etc) to three (whatever they were using previously plus Make plus shell).

Because you imply that Webpack isn't. It gives you much more that with make you still have to invent.

The newer versions of WP is surprisingly simple to set up.

It's 100% true.

I've been using Make for 15 years across multiple programming languages. It's a small, dependable, and simple utility.

I've been painfully slogging through Webpack configs for 4 years. I dread dealing with it every single time I open the file.

If only Makefiles were usable. tabs vs space...

Production mode already defines NODE_ENV. [1] Your usage of DefinePlugin is redundant in that case. Any reason why you are defining it twice?

[1]: https://webpack.js.org/concepts/mode/

I think for beginners it's great to see what production mode actually does under the hood, but it shouldn't end up in the final config.

Also, DefinePlugin is superseded by EnvironmentPlugin: https://webpack.js.org/plugins/environment-plugin/

Interesting. I was attempting to compile some bundles over the weekend and was noticing drastically larger chunk sizes. I realized this was because I had neglected to specify NODE_ENV=production, and my babel-loader was using my dev config (which leaves prop-types in the code). Based on your comment, and the above docs, I shouldn't be seeing what I'm seeing.

Are you using webpack 3? That change happened in February with webpack 4:


Cool tool. Would be great if it was also able to load and manage sections of existing Webpack configs.

Why do people import PNGs into their JavaScript? That seems pretty weird... Not a criticism of this tool, which seems quite nice.

It's a convenience thing. Once you get used to it, it works well. You define a loader in your webpack config, which then decides what to do with the PNG.

Usually, what happens if you require a PNG is that it passes via url-loader or file-loader, and optionally image-loader. This'd mean that you'd get a string (either a full URL or a base64-encoded inline image if it's small enough) that points to the image, which is optionally minified by image-loader.

If you require CSS or SASS, your webpack loaders will ensure the CSS is loaded.

No, I get what is trying to be achieved, I am just questioning why it is being done this way and not something much simpler... I would much prefer that my JavaScript bundler only bundles JavaScript. Converting a PNG into a JavaScript module through base64 should be pretty trivial, so I'd prefer to do that with another tool. I do something similar for inlining CSS in my own projects.

You can use one tool to optimize and base64 inline images and another to compile SCSS and inline it of course. But then you discover that you also want to optimize and inline images placed within css url() properties so you implement that feature in your build pipeline, then you discover... you get the point. Webpack gives you a flexible configurable and extensible way to handle any such case in any environment which is why everyone uses it for modern frontend development. I think questioning webpack is like questioning why people are using dependency injection frameworks in modern java.

And then you wind up spending multiple days trying to figure out how to do something that doesn't quite fit in with what other people have already done.

How many times do we have to learn the same lessons over and over? Small tools that compose together is infinitely more flexible than any plugin based tool can possible be.

Importing PNGs to JS is about dependency-wrangling, not task-running.

That is, the point of using webpack isn't that it's the best way to optimize a bunch of PNGs. The point of webpack is that it can parse your dependency tree and figure out which PNGs are depended on by a given entry point. If you also want your PNGs optimized, you can do that with a webpack plugin or a separate tool, as you like - that's orthogonal to the matter of why it's useful to import PNGs into JS.

Yeah, sorry but I don't buy that. webpack outputs JavaScript bundles, not a dependency graph. Importing PNGs into JavaScript is only useful if you use a JavaScript AST to then do something with those imports.

I'm not selling it, and I don't follow what you're disagreeing with. Webpack can output JS bundles, but it understands dependencies between non-JS assets as well, and will process any file in the dependency graph any way you like.

As such the point of requiring a PNG into JS (or CSS, or a Vue/React component, etc) is to tell webpack there's a dependency on it. Then that image will get pulled in to the build process for entry points that need it, and not otherwise.

I'm sorry, but what is wrong with the <img> tag?

Scope Hoisting is already part of "production" mode now which enables this and other optimizations automatically.

Another poster mentioned this, but I'll reiterate: "avoid Webpack altogether".

ES6 or UMD/AMD please. Please do not subject me to toolchain choices just to use a lib.

If you're a TypeScript or Babel user, you can share code between backend, web tier, and native code without Webpack or other complex build tools.

Webpack is unnecessary in most cases.

If I had to give one Tipp about front-end development: transition the buildsystem to gradle and call webpack from there. Also npm can be managed with it.

Then you can slowly introduce kotlin code, which is way more productive.

In the end I found that writing everything in kotlin, using tachyons for CSS, the kotlin-minifier for code minification and closure-compiler for optimization is the most stable way to go.

JavaScript isn't a productive language. Having build-tools in JS just complicates everything.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact