CSS performance revisited: selectors, bloat and expensive styles

What is fast CSS? Where are the bottlenecks? Are the rules of slow and fast selectors even valid anymore? Are the properties we use more important than the selectors? I felt it was time to revisit some of these questions.

In the broad scheme of things, CSS optimisation is certainly low down the priority order when trying to speed websites/web applications up. There are so many other optimisations that provide easier and greater gains. However, if tiny improvements are made to all areas of a website, including CSS, combined they will make a more substantial difference; the user will always benefit.

Whenever exchanging theories about the relative ‘speed’ of CSS, other developers often reference Steve Souders work on CSS selectors from 2009. It’s used to validate claims such as ‘attribute selectors are slow’ or ‘pseudo selectors are slow’.

For at least the last couple of years, I’ve felt these kinds of things just weren’t worth worrying about. The soundbite I have been wheeling out for years is:

With CSS, architecture is outside the braces; performance is inside

But besides referencing Nicole Sullivan’s later post on Performance Calendar to back up my assumptions that the selectors used don’t really matter, I had never actually tested the theory; a shortfall in my talent and a less than perfect analytical mind prevented me from even attempting it.

Nothings changed with my mind, but these days I feel happier to open myself to ridicule by attempting this – if only to get someone with more knowledge/evidence to provide further data. So I decided to create some primitive tests.

Steve Souders’ aforementioned tests use JavaScript’s new Date(). However, nowadays, modern browsers (iOS/Safari being notable exceptions) support the Navigation Timing API which gives us a more accurate measure we can use. I’ll be implementing it like this:

<script type="text/javascript">

    ;(function TimeThisMother() {

        window.onload = function(){

            setTimeout(function(){

            var t = performance.timing;

                alert("Speed of selection is: " + (t.loadEventEnd - t.responseEnd) + " milliseconds");

            }, 0);

        };

    })();

</script>

This lets us limit the timing between the point all assets have been received ( responseEnd) and the point the page is rendered ( loadEventEnd).

So, I setup a very simple test. 20 different pages, all with an identical, enormous DOM, made up of 1000 identical chunks of this markup:

        <div class="tagDiv wrap1">

          <div class="tagDiv layer1" data-div="layer1">

            <div class="tagDiv layer2">

              <ul class="tagUl">

                <li class="tagLi"><b class="tagB"><a href="/" class="tagA link" data-select="link">Select</a></b></li>

              </ul>

            </div>

          </div>

        </div>

Each page differed only in the rule applied to select the inner most node within the blocks. 20 different selection methods were tested to colour the inner most nodes red:

  1. Data attribute
  2. Data attribute (qualified)
  3. Data attribute (unqualified but with value)
  4. Data attribute (qualified with value)
  5. Multiple data attributes (qualified with values)
  6. Solo pseudo selector (e.g. :after)
  7. Combined classes (e.g. class1.class2)
  8. Multiple classes
  9. Multiple classes with child selector
  10. Partial attribute matching (e.g. [class^=“wrap”])
  11. nth-child selector
  12. nth-child selector followed by another nth-child selector
  13. Insanity selection (all selections qualified, every class used e.g. div.wrapper > div.tagDiv > div.tagDiv.layer2 > ul.tagUL > li.tagLi > b.tagB > a.TagA.link)
  14. Slight insanity selection (e.g. .tagLi .tagB a.TagA.link)
  15. Universal selector
  16. Element single
  17. Element double
  18. Element treble
  19. Element treble with pseudo
  20. Single class

The test was run 5 times on each browser and the result averaged across the 5 results. Modern browsers were tested:

A prior version of Internet Explorer (rather than the latest IE) was used to shed some light on how a popular browser behaved that doesn’t get the same rolling frequent updates of the other browsers.

Want to try the same tests out for yourself? Go and grab the files from this GitHub link:

https://github.com/benfrain/css-selector-test

. Just open each page in your browser of choice (remember the browser must support the Network Timing API to alert a response). Also be aware that when I performed the test I discarded the first couple of results as they tended to be unusually high in some browsers.

When considering the results, I don’t think that one browser compared with another really tells us much. That is not the purpose of the tests. The purpose is purely to try and evaluate the comparative difference in selection speed between the different selectors employed. Therefore, when looking at the table, it makes more sense to look down the columns than across the rows.

Here are the results. All times in milliseconds:

Test Chrome 34 Firefox 29 Opera 19 IE9 Android 4
1 56.8 125.4 63.6 152.6 1455.2
2 55.4 128.4 61.4 141 1404.6
3 55 125.6 61.8 152.4 1363.4
4 54.8 129 63.2 147.4 1421.2
5 55.4 124.4 63.2 147.4 1411.2
6 60.6 138 58.4 162 1500.4
7 51.2 126.6 56.8 147.8 1453.8
8 48.8 127.4 56.2 150.2 1398.8
9 48.8 127.4 55.8 154.6 1348.4
10 52.2 129.4 58 172 1420.2
11 49 127.4 56.6 148.4 1352
12 50.6 127.2 58.4 146.2 1377.6
13 64.6 129.2 72.4 152.8 1461.2
14 50.2 129.8 54.8 154.6 1381.2
15 50 126.2 56.8 154.8 1351.6
16 49.2 127.6 56 149.2 1379.2
17 50.4 132.4 55 157.6 1386
18 49.2 128.8 58.6 154.2 1380.6
19 48.6 132.4 54.8 148.4 1349.6
20 50.4 128 55 149.8 1393.8
Biggest Diff. 16 13.6 17.6 31 152
Slowest 13 6 13 10 6

The Biggest Diff. column shows the difference in milliseconds between the fastest and slowest selector. Of the desktop browsers, IE9 stands out as having the biggest difference between fastest and slowest selectors at 31ms. The others are all around half of that figure. However, interestingly there was no consensus on what the slowest selector was.

I was interested to note that the slowest selector type differed from browser to browser. Both Opera and Chrome found the ‘insanity’ selector (test 13) the hardest to match (the similarity Opera and Chrome here perhaps not surprising given their shared blink engine), while Firefox struggled with a single pseudo selector (test 6), as did the Android 4.2 device (a Tesco hudl 7" tablet). Internet Explorer 9’s achilles heel was the partial attribute selector (test 10).

One thing we can be clear on is that using a flat hierarchy of class based selectors not only produces more modular and less specific code making it more modular and re-usable, it also provides selectors that are as fast as any others (yes, ID selectors would probably be faster but I for one don’t fancy building a large code base up relying on ID selectors).

What does this mean?

For me, it has confirmed my believe that it is absolute folly to worry about the type of selector used. Second guessing a selector engine is pointless as the manner selector engines work through selectors clearly differs. Further more, the difference between fastest and slowest selectors isn’t massive, even on a ludicrous DOM size like this. As we say in the North of England, ‘There are bigger fish to fry’.

If the number of DOM elements on the page is halved, as you would expect, the speed to complete any of the test drops commensurately. But getting rid of the DOM isn’t always a possibility. This made me wonder what difference the amount of unused styles in the CSS would have on the results.

Another test: I grabbed a big fat style sheet from fiat.co.uk. It was about 3000 lines of CSS. All these irrelevant styles were inserted before a final rule that would select our inner a.link node and make it red. I did the same averaging of the results across 5 runs on each browser.

I then cut half those rules out and repeated the test to give a comparison. Here are the results:

Test Chrome 34 Firefox 29 Opera 19 IE9 Android 4
Full bloat 64.4 237.6 74.2 436.8 1714.6
Half bloat 51.6 142.8 65.4 358.6 1412.4

This provides some interesting figures. For example, Firefox was 1.7X slower to complete this test than it was with its slowest selector test (test 6). Android 4.3 was 1.2X slower than its slowest selector test (test 6). Internet Explorer was a whopping 2.5X slower than it’s slowest selector!

You can see that things dropped down considerably for Firefox when half of the styles were removed (approx 1500 lines). The Android device came down to around the speed of its slowest selector at that point too.

Removing unused styles

Does this kind of horror scenario sound familiar to you? Enormous CSS files with all manner of selectors (often with selectors in that don’t even work), lumps of ever more specific selectors 7 or more levels deep, non-applicable prefix’s, IDs all over the shop and file sizes of 50–80KB (sometimes more).

If you are working on a code base that has a big fat CSS file like this, that no-one is quite sure what all the styles are actually for – look there for your CSS optimisations before the selectors being employed.

Tackling this first seems to make more sense than being picky over the selectors used. It will have double the impact; less code for the user to download but also less for the UA to parse – a speed bump all around.

Then again, that won’t help with the actual performance of your CSS.

The final test I ran was to hit the page with a bunch of ‘expensive’ properties and values.

.link {

        background-color: red;

        border-radius: 5px;

        padding: 3px;

        box-shadow: 0 5px 5px #000;

        -webkit-transform: rotate(10deg);

        -moz-transform: rotate(10deg);

        -ms-transform: rotate(10deg);

        transform: rotate(10deg);

        display: block;

    }

With that rule applied, here are the results:

Test Chrome 34 Firefox 29 Opera 19 IE9 Android 4
Expensive Styles 65.2 151.4 65.2 259.2 1923

Here all browsers are at least up with their slowest selector speed (IE was 1.5X slower than its slowest selector test (10) and the Android device was 1.3X slower than the slowest selector test (test 6)) but that’s not even the full picture. Try and scroll! Repaint on these kind of styles will make your computer cry.

The properties we stick inside the braces are what really taxes a system. It stands to reason that scrolling a page that requires endless expensive re-paints and layout changes is going to put a strain on the device. Nice HiDPI screen? It will be even worse as the CPU/GPU strains to get everything re-painted to screen in under 16ms.

With the expensive styles test, on the 15" Retina MacBook Pro I tested on, the paint time shown in continuous paint mode in Chrome never dropped below 280ms (and remember, we are aiming for sub–16ms). To put that in perspective for you, the first selector test page, never went above 2.5ms. That wasn’t a typo. Those properties created a 112X increase in paint time. Holy ’effing expensive properties Batman! Indeed Robin. Indeed.

An ‘expensive’ property/value pairing is one we can be pretty confident will make the browser struggle with when it has to repaint the screen (e.g. on scroll).

How can we know what will be an ‘expensive’ style? Thankfully, we can apply common sense to this and get a pretty good idea what is going to tax the browser. Anything that requires a browser to manipulate/calculate before painting to the page will be more costly. For example, box-shadows, border-radius, transparency (as the browser has to calculate what is shown below), transforms and performance killers like CSS filters – if performance is your priority, anything like that is your worst enemy.

Juriy “kangax” Zaytsev did

a fantastic blog post also covering CSS performance

back in 2012. He was using the various developer tools to measure performance. He did a particularly good job of showing the difference that various properties had on performance. If this kind of thing interests you then that post is well worth your time.

These are my takeaways from this little episode:
- sweating over the selectors used in modern browsers is futile; most selection methods are now so fast it’s really not worth spending much time over. Furthermore, there is disparity across browsers of what the slowest selectors are anyway. Look here last to speed up your CSS.
- excessive unused styles are likely to cost more, performance wise, than any selectors you chose so look to tidy up there second. 3000 lines that are unused or surplus on a page are not even that uncommon. While it’s common to bunch all the styles up into a great big single styles.css, if different areas of your site/web application can have different (additional) stylesheets added (dependency graph style), that may be the better option. If your CSS has been added to by a number of different authors over time, look to tools like UnCSS to automate the removal of styles – doing that process by hand is no fun!
- the battle for high performing CSS will not be won in the selectors used, it will be won with the judicious use of property and values
- getting something painted to screen fast is obviously important but so is how a page feels when the user interacts with it. Look for expensive property and value pairs first (Chrome continuous repaint mode is your friend here), they are likely to provide the biggest gains.