Hacker News new | past | comments | ask | show | jobs | submit login
PurifyCSS – Remove unused CSS (github.com)
231 points by k8t on June 12, 2015 | hide | past | web | favorite | 64 comments

I've been looking into this kind of tool for a while, but none of them are quite right for my purpose. I don't want a tool that spits out just-the-CSS-that-is-used, I want a tool that shows me blocks of CSS that it thinks are NOT used - ideally after I've loaded a number of different pages in a browser.

My use-case is for analyzing a large, existing website that has many years of accumulating CSS, and getting a feel for which blocks of code can be safely eliminated.

I'm pretty sure what I'm after needs a full-blown DOM to work effectively - either with something like PhantomJS or even running as a Chrome Extension. So far I haven't seen anything that fits the bill.

I honestly think this needs to be a part of Chrome Dev Tools's profiler. Already it's able to record sessions that persist across page loads, and tracks what CSS selectors are applied and which properties are overridden. I don't think it's much of a stretch to add a feature like this.

+1 for something along these lines to be available in all vendor's dev tools!

Agreed! We need something that collects data from multiple pages as you browse through and use your site. At the end it gives you a table of all your classes that didn't match anything (with line number references to your css file).

Though you need to manually visit all those pages. And it wouldn't be easy to traverse DOM with those server side logic/condition.

IMHO, do it offline and out of browser would be easier. But I also think doing it directly with the codes can be hard too since there are so many way that those css will be added to templates.

Unless the pages are dynamically generated

Firefox dev tools have something like that (experimental)


I have used a Firefox extension called Dust-Me Selectors[1] which does this. It breaks your css into used and unused selectors, it can spider the site for you, or you can leave it running as you browse the site and it will work automatically. Definitely slows things down though while you're using it.

1: https://addons.mozilla.org/en-US/firefox/addon/dust-me-selec...

100% safety is pretty difficult, but take a look at the tool described here starting at slide 60, http://talks.desp.in/fronteers2014

It's fairly easy to roll your own version of this and do analysis.

I might be wrong but I'm glad to see you try using this REGEX https://news.ycombinator.com/item?id=9708629

How about running a diff on pre-purify versus post-purify?

Chrome dev tool's audit tab has an "unused CSS" section. Have you tried that?

Thanks, didn't know this was there. It is somewhat buried, but does give you a list of unused selectors.

I built something like this at my old company. Scrape every tag on the current page, subtract this array of used tags against all the tags in the CSS. Didn't take too long to build.

I created and maintain Helium. https://github.com/geuis/helium-css

The problem with the PurifyCSS approach, like so many before it, is that you cannot really do this accurately from the server. It has to be done in the browser.

Helium is a dev tool. It takes a list of sample urls from the developer because its presumed the engineer will be able to make the best choice as to pages that represent all the aspects of their site.

It will work in any web page regardless of framework, because ultimately its all just web pages.

Helium will find the actual style sheets loaded into each page, then at the end of the scan give you an accurate and realistic report of the css selectors that were not found anywhere in your site.

There are some minor limitations, such as the inability to test for user-interactions with pseudo selectors like :active, :focus, :hover, etc.

As a designer that would love this tool for the speed and cleanliness factor it would give, i have a few other Qs:

1. Does it spit out a css file or multiple css file I can use immediately? Can i actually amend a local document with it, or do I have to copy paste?

2. Does it give any critique on poor selector methods? Definitely have a few foolish ways of doing things that could be improved with a little bit of "hey...stop that" feedback, in my work

3. ??? PROFIIT

4. Can it give me a few stats on the css to benchmark it? You're already showing me the unused ccs selectors, can you present other data that you could present to your boss and say hey...look at this, this is why it's worth doing this stuff

5. Known browser issues highlighted. "Hey! In IE8...that div ACTUALLY does this". Also, heres how to fix it. Insta–contextual documentation. As I say that I realise I'm asking for Clippy back in a way so I shall bow out. Responsive design? worth thinking about...anyway!

Really interesting project! I'm going it use it this weekend and see how it goes :)

It doesn't modify your css files. That's your job. It gives you a report of unused selectors. It can't test pseudo selectors requiring human interaction.

Like most CSS tooling, this is a treatment for a symptom of an underlying disease: the inherently global nature of CSS[1].

Rather than "blob" all your CSS together, and then "de-blob" it, why not build your CSS dynamically based on what components your app is using?

[1] I strongly encourage you to watch this talk: https://vimeo.com/116209150

For many reasons, namely that we've already solved most of these "problems" years ago with simple concepts like OOCSS, SMACSS, and BEM. I suggest taking a look at this rebuttal. http://keithjgrant.com/posts/against-css-in-js.html

The reasons people are running into problems with global css scope is because they don't understand the basics of how to write effective and maintainable CSS. Seems like many front end devs nowadays grew up with css hand-holding libraries like bootstrap, and can't seem to wrap their heads around completely necessary things like taking account for CSS specificity and how the cascade works and how to use it to your advantage.

The way methodologies like BEM and SMACSS work is to acknowledge that the cascade never works to your advantage, and avoid it entirely.

The methodology gives you the tools to have control over the cascade. You cannot avoid it entirely.

Edit: Also the point I was getting at was that cascade is a useful and powerful tool, just as the methodologies for controlling it are. We will also soon have a property that will eliminate the cascade entirely... all:unset;

I just think the problems and solutions listed in the talk are either non-problems or are things that could be solved in superior ways. And that it introduces more problems and limitations than it solves.

We've been using inline styles in our main project (built with React), but we recently deciced to move to css modules:



Inline styles are great, but they don't support basic features such as pseudo classes/elements, so even implementing simple components like buttons was cumbersome.

With latest version of webpack's css loader you get css scoping (you no longer need to add lengthy prefixes to all selectors) and you can also use additional loaders for post-processing your styles (we use a lot of postcss plugins such as autoprefixer).

Have a look at this example:


And here's a great comparison of the main CSS in JS techniques:


Because it slows down your app by a relevant amount.

Edit: I see we're referring to inline styles instead of requirecss. In that case, it's because code reusability becomes a pita and you bloat your JS.

There's another way besides inlining and requirejs. In cases where you're doing server-side programming anyway, you can just have the server inject view-specific <link> elements pointing to individual CSS files as needed - augmenting the site's core CSS file. The downside is more requests to fetch those files, but after the first time you can take advantage of the browser's cache. As long as you keep the number of extra CSS files reasonable, this method costs almost nothing.

A variant of this is where you compile one CSS file from many view-specific ones beforehand. The advantage is it'll speed up things on the client side and the programming overhead is relatively small if you're using something to generate CSS anyway.

> In that case, it's because code reusability becomes a pita and you bloat your JS.

Does it? Last time I had a bigish web project, I was starting to inline everything (CSS and Javascript), and had my backend language deal with DRY. It even solved the problem CSS people use CSS generators for. As a rule, both CSS and javascrit suck for organizing content.

But then, I declared that project dead, and didn't wait to see the consequences of organizing things this way. So, somebody may have a much clear understanding about why that's wrong.

This looks awesome. If it did a second run to look up where the CSS comes from in my SASS and remove the SASS instead, that would be magical. :D

Until someone makes that, you should be able to just diff the two files, see what sections were removed and go based on that.

Couldn't you just add a Grunt task to purify the compiled SASS?

This would miss the point. Sure, it would still lighten the css file downloaded by the client, but it would do nothing to clean up the source file developers work with and thus lighten their cognitive burden, which imo is more important.

Nice. But from the description, it apparently doesn't work if you construct class names by string manipulation (like "item_" + (selected ? "selected" : "")). Maybe you shouldn't do that anyway, though.

Even worse, I've sometimes done things like

  $(...).addClass("bg" + getSomeClassIdentifier(obj))
Linking some string hanging off some runtime object to its CSS class usage quickly runs into halting problem issues.

There are certainly some arguments to keeping class names as opaque magic strings, but given certain levels of dynamic complexity and the lack of tools like CSS class inheritance, naming policy is sometimes far cleaner and more manageable.

Also, there is a minor concern about false positives. If you have a class called "name" or something generic like that, the odds of that string appearing in non-CSS usage your source code is fairly high.

Ah, thanks for noticing. Actually, with the current implementation, it would still work I believe

I _think_ it shouldn't be too much code. For my rails project, I wrote a simple shell script[1] to get all unused css classes and remove it my hand in final step, to make sure i will not remove css classes that actually used (e.g pre-define and use in future, overriding of third party css class)

The script basically does:

1. Use REGEX match all css classes, `cat` into one place

2. Read the class line by line and search in html,js files to see if this css is used (even support #addClass from js)

2.1 it also supports several class adding styles e.g. class="abc", class: "abc", :class=> 'abc' or even "xxx" class in this: class="abc xxx ijk"

2.2 People even do this in js: $modal.append($('<div class="modal-close-icon"></div>')); and the REGEX can also detect this lol.

[1] https://gist.github.com/50kudos/3028fac585eda85aea9a

You can adapt my REGEX, and custom your directory where those css files are in.

Simply copy the code and save into your filename.sh, and can run safely because my script don't write any file of your project.

This setup is ideal IMO. Keep your existing files, like bootstrap, intact in case you ever DO use them but integrate this into your deploy process to do compression and minification prior to getting out to your server.

Workflow wise, this is a huge win.

Apparently it also has a gulp plugin: https://github.com/purifycss/gulp-purifycss

(just in case you were also wondering) =)

This could be really useful when I do a major refactor later this year. Thanks.

Kind of funny. Although maybe true.

This is nice. Just in case, you might also like to look at the likes of https://github.com/giakki/uncss

Yes! that is a great library. Unfortunately, that library does not automatically detect the classes used in JS. So we thought we could give it a try to make one that would detect classes used in JS by default

How do you do this for JavaScript? In Java I used a mixture of static analysis[1] and code coverage[2] with great success. For JavaScript there seem to be many (dead) coverage tools but nothing comparable to UCDetector.

[1] http://www.ucdetector.org/

[2] http://www.eclemma.org/jacoco/trunk/doc/

I use a similar tool on my personal site (www.peterbe.com) It also inlines all CSS but only does so with the selectors that are actually ever used.

Go to http://www.peterbe.com/plog/mincss and click to view source. 120Kb of bootstrap excess baby!

What is the actual performance advantage gained in contrast to a simple minifier?

Given today's CPU speeds, I think the only thing you can save with this tool is bandwidth on mobile and even with this, the effort is not always worth the resulting speed gains as mobile networks get faster and faster.

The dynamic class allocation here looks really robust and amazing. Huge concern in similar libraries.

>Able to also detect dynamically-loaded CSS classes in your javascript.

Awesome! I worked with a tool like this that did not have dynamically loaded classes in mind.

For people not sure on how to use this, I personally use tools like these when working with a CSS framework.

An alternative to this is symdiff http://symdiff.github.io/ (Disclaimer: I'm the author).

The auto-remove part you would have to code yourself though.

Wrote a quick tutorial http://grunt-tasks.com/grunt-purifycss/ on how to use it with grunt and grunt-usemin

Cool. I will try later. I'm using UnCSS[0] with Gulp. What is the different with UnCSS?

[0] https://github.com/giakki/uncss

Mobile developers dream tool!

This is awesome and makes so much sense, great work team.

This is a very handy tool. I have a question, would it be able to take wildcard and recursive argument for filepath of HTML, JS? Thanks.

AWESOME!! this is just what i was looking for!!

When it says it works on single-page apps, does that mean ONLY single page apps?

Thanks for giving a heads up. It works with all apps. We just put the emphasis on SPAs. Updated readme for this

This is amazing.

Cool. Nice work!

super awesome.

So basically it's a tool to remove the mess that we've created? Maybe we shouldn't have done so in the first place.

Do you also scorn at the existence of debugging tools because we should've written the code correctly the first time?

Ha! that's one way of putting it. Another use case would be if you were to use a CSS framework (Bootstrap, Foundation) and you don't utilize all of it (they are both ~6k line files)

Half the time, I use tools the way a race car driver uses a five point harness. The things let me move at a pace that would otherwise be dangerous.

The rest of the time I use it to illustrate when we aren't living up to agreements that we've already made.

But we're just so good at creating messes!

I am looking for best practices for CSS. Can someone point me to one they found really useful?

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