PIE 2’s new loading mechanism

One of the guiding principles in the development of CSS3 PIE has always been to make it as simple to use as possible. That’s why it was designed as only a single standalone .htc file, without any external dependencies. Copy a single file and reference it — can’t get much simpler than that!

However, having a single monolithic file has posed some other problems. Especially with the introduction of IE9 compatibility which required entirely new renderers that use SVG instead of VML, PIE.htc grew to contain a lot of code that isn’t used by whichever browser is currently downloading it. IE9 has to download all the code for parsing and rendering border-radius and box-shadow using VML, for example, even though it has a native implementation. Similarly, IE6-8 have to download the IE9 SVG rendering code which they cannot use.

To make this even worse, IE has a strange bug that sometimes causes the .htc file to be fully or partially downloaded more than once, so the hit can be as much as doubled in some cases.

It was clear that the simplicity of a single file wasn’t worth this performance hit, so while I was designing the next generation of PIE one of my main goals was to implement a conditional loading system which didn’t sacrifice simplicity of use. I’m going to describe what’s been implemented so you know what to expect in the upcoming PIE 2.0 (first beta will be released soon!), and to solicit your feedback on it.

What’s different?

Firstly, PIE.htc is now very small. Like 2KB small. All it contains is enough logic to figure out what secondary JS file to load and where to load it from. So even if the double-htc-download bug happens, it’s only a tiny thing being downloaded twice.

The secondary JS files contain all the rest of the logic for parsing and rendering the CSS3 styles. There are currently two logic files: PIE_IE678.js and PIE_IE9.js, containing only the code needed by IE6-8 and IE9 respectively. The .htc file loads only the one appropriate for the current browser. (It’s possible that in the future these can become even more granular, for instance minimal builds that don’t include code for unused CSS3 properties.)

These secondary logic JS files will also double as the files for PIE JS Edition — just include them directly using conditional comments so only the appropriate one is loaded.

Can we get some numbers?

Let’s compare the difference in total download size. These are actual numbers from loading the css3pie.com homepage in IE:

Version	Request	Size
PIE 1.0: All IEs	PIE.htc first request	8.27 KB
	PIE.htc second request	40.35 KB
	Total	48.62 KB
PIE 2.0: IE6-8	PIE.htc first request	2.00 KB
	PIE.htc second request	0.17 KB
	PIE_IE678.js	37.29 KB
	Total	39.46 KB
PIE 2.0: IE9	PIE.htc first request	2.00 KB
	PIE.htc second request	0.17 KB
	PIE_IE9.js	25.80 KB
	Total	27.97 KB

So we’re seeing about a 9 KB smaller total download in IE6-8, and 20 KB smaller in IE9. Of course that’s not the entire story — there’s the extra HTTP request for the JS file to take into account. So let’s take a look at the actual network timeline, on a simulated slow connection. Here’s PIE 1.0 in IE9, loading css3pie.com/blog:

And here’s PIE 2.0 with the new loading system:

As you can see, even with the extra HTTP request the total download time has still improved by more than a second. But what’s even more interesting is that while IE waits for the .htc file before firing document ready, it does not wait for the secondary .js file, so it fires a whole 6 seconds earlier! This means your other scripts that wait for document ready can start running much earlier, making your page feel snappier.

Customizing the loading path

When PIE.htc loads the secondary files, by default it loads them from the same server path as PIE.htc itself. This keeps things as simple as possible for users; as long as you keep the distribution files together in the same directory, you’re set.

However, for extra flexibility I’ve also added the ability for you to override that location if you wish, by specifying a new -pie-load-path property on the html element:

html {
    -pie-load-path: "http://any.server/path/to/pie-js-files/";
}

The path you specify can be anywhere on any server, as the JS files are not subject to same-domain limitations. You can even choose to load them from a CDN to take advantage of caching and geolocated delivery.

Help us test!

If you have any feedback on this new loading system, good or bad, I’d love to hear it. It’s very likely that unforeseen issues could pop up once it starts being used in the wild, so getting your help testing the new system is crucial! There will be an official 2.0 beta build released in the very near future; or if you can’t wait you can build PIE yourself from the master branch on GitHub and try it out right now. If you find any issues please post them in the forums.