on my Windows 7 VM. My impressions from 30 seconds with IE9:
- CSS3: Rounded corners, yes. Gradients, no.
- Better, if you can believe this, font rendering than Firefox 4.
This website is optimized to work in modern browsers like Safari 3+, Firefox 4+, Chrome 10+ and IE9+. If you are using a different browser, you may experience visual glitches or other problems.
on my Windows 7 VM. My impressions from 30 seconds with IE9:
Ran into this one today, in somewhat unique circumstances: I had a hover style where buttons inside a table row only appeared when the mouse was over the row, and if you clicked on one of the buttons but moved the mouse into another row before releasing the mouse button, the previous row would stay highlighted.
The authoritative writeup of the bug, and most others, refer to the problem happening with dynamically-displayed submenus, but make little mention of my particular problem. There was some hope that triggering
hasLayout would fix the problem, as it’s the closest thing we have to a magic bullet, but it (shocker!) didn’t work for me.
:hover pseudoclass. So much for IE7 implementing
:hover on non-anchor elements. Sigh.
1 See the first comment. And shame on you, Webmaster World, for not having permalinks to the comments. Get a load of their generated source, too: they’re still using
font tags. Seriously?! And this is a resource for webmasters? Come on.
Robert Nyman has posted an updated version of his ultimate getElementsByClassName. It takes advantage of the native getElementsByClassName support in Safari 3, Firefox 3 and Opera 9.5*, takes advantage of native XPath support if it’s available, and allows multiple class names in the request, among other things. ROCK.
* And ooh, big surprise, guess which modern browser is getting lapped by the others and choking on their dust? IE7.
So you may have seen me mention before how, every day since I wrote this post about a workaround for the IE7 z-index bug, it has accounted for anywhere between 25 and 50% of my traffic.
Which is cool, don’t get me wrong. But when I was checking out my site statistics this morning, I realized that one of my more recent visitors, going to that post, was coming from the microsoft.com domain, meaning that s/he is a Microsoft employee.
Oh, the irony is simply delicious.
Update 2009/09/12 09:19—The jQuery version of the dynamic z-index function has been optimized to use
hover instead of
mouseover/mouseout. Thanks to commenter Eric Conner for pointing out my mistake.
Update 2009/08/04 22:22—The jQuery version of the dynamic z-index function has been added.
Update 2009/07/09 10:52—The code has been updated to make use of the latest innovations in the Prototype library, and to take suggestions from commenters into account.
Now that we’ve officially transferred support of our big application from IE6 to IE7, I’ve been spending more and more time delving into the nitty gritty details of making that hugely complicated layout (recently made less complicated, but that’s another blog post) work in IE7.
I had thought, initially, that IE7′s superior CSS support would make my life a lot easier, and it did in some ways: my IE7-specific stylesheet is a fraction of the size of the IE6-specific one, and the ability to use attribute selectors is huge for me. But as I should have expected, I just traded one set of headaches for another. While IE7 is far superior to IE6 in terms of CSS support (and far inferior in other ways, user interface chief among them—but that’s another blog post, too), the bugs that still exist are more subtle and far more difficult to work around.
The IE z-index bug had been around since IE4 or 5, and was finally fixed in IE8. It’s still an issue in IE7, however, and chances are that you, the web developer, still have to support it. You should read Aleksandar Vacić’s in-depth characterization of the problem to really understand what’s happening, but what it boils down to is this: the CSS 2.1 spec says that a positioned element with any integer z-index value (i.e. not
auto) should create its own zero-based stacking context, and use the integer value specified to decide its place in its parent stacking context. In other words, if the positioned element has a z-index of
auto, its stacking context is inherited from its parent. Internet Explorer, however, creates a new stacking context for elements with any z-index value, including
auto, which wreaks all kinds of havoc and generally causes mayhem in your previously neat and orderly layouts.
Any halfway successful project is a significant learning experience, and the one we just completed at work was, bar none, the most challenging I’d ever faced. Because of the incredible complexity* of the UI we were trying to implement, I had to empty out my bag of tricks, and then find new ones to add to it. Quite aside from the innovations in design and implementation I perforce became familiar with, though, I fought a pitched battle with IE6.
It seemed that at every turn I was thwarted by some IE quirk, and I had to fix each problem as I found it; success was my only option. So I built up my little arsenal, and went to it. What follow are the biggest issues I faced in making the site IE6-compatible, and how I worked around them. They are not the only issues there were, but they were the most common, and are IMO the most important to keep in mind when developing any new site for IE6.
cf. The peekaboo bug.
list-style: none, the bullets still show up.
list-style: none noneinstead.
getElementsByClassNamefunction is doubling or tripling page load times in IE6.
* No, really. For this one project, we used what seemed like every modern UI widget in existence: multi-column layouts, tabbed interfaces, flyout menus, drag and drop lists, you name it.
I mean it. Terrible. On one of my most recent projects at work, a Java-based, heavily JS & AJAX-reliant web-application, we discovered that our page load times in IE were anywhere between two to ten (!) times as long as those in Firefox. Which was, obviously, unacceptable, especially given the fact that the app’s user-base is comprised almost entirely of IE6/Win users. So what was I to do? Optimize, of course.
On my quest to find how to optimize my JS to run faster in IE, I came across these great articles, straight from the horse’s mouth, on how to do it:
Unfortunately, while they were certainly illuminating, these articles didn’t do much to solve my problem. Their recommendations, such as caching DOM lookups, weren’t much help to me when I only needed to do those lookups once.
After that, I thought briefly that part of the reason the page controls were taking so long to become available was that the onload event was being delayed due to images loading on my page, thereby causing an overall delay in total page-load time. And I found this nifty method to kick off
window.onload after the DOM has loaded, but before other things like images finish loading.
As it happened, though, that didn’t make a bit of difference with my particular issue. What was causing the problem—as I found out after using the brute-force method to comment out all the code in the offending JS function, then adding it back in line by line to see what broke the performance camel’s back—was that using Prototype‘s built-in
getElementsByClassName function was doubling, or tripling, or worse, IE page-load times. Especially since I called it in 2 or 3 different functions, all of which ran at
So I needed to get (or, god forbid, write) an optimized
getElementsByClassName function. Luckily, though, I was spared that horrible fate, because I found the Ultimate
getElementsByClassName, courtesy of a guy by the name of Robert Nyman. In a fit of desperation, I added this to my JS library and changed all my code to call it instead of Prototype’s version. Et voilà: just from that one change, I had quadrupled (quadrupled!) my JS performance in IE. Pretty amazing.
Update 2007/10/24 1453: I recently ran into a similar problem again, where page load times, particularly for large pages, were up to 10 times slower in Safari than Firefox, and 20 or more times in IE. The problem was eventually traced back to the use of the Prototype double-$ lookup for CSS-style selectors. The solution? Just don’t use it. It might take more code to find the item you’re looking for, but the performance gain you get from not using the $$ lookup is worth it.
Making your site work in IE
So there you are, with a good understanding of the standards, armed with some great CSS UI conveniences, and a snazzy prototype. But then you fire it up in Internet Explorer 6 for Windows, and it all goes to hell. You’ve got unnecessary horizontal scroll bars, magically disappearing and reappearing content, and list bullets where you don’t want them. Mysterious extra right and left padding, odd positioning, and static content that jumps when you hover on a nearby link. Problems galore, and you’re ready to tear your hair out. (My colleagues say that they know when I’m working with IE because I start swearing at my computer.) It’s inevitable, really.
But cling to hope, gentle reader, because you’re not the first, and, I am confident in saying, not the last to have these problems. In fact, most of the problems you will encounter have already been found—and solved. But before I get to that, let’s discuss the problem of how to hack your CSS to work in IE such that it’ll degrade gracefully when your users inevitably upgrade to IE7.
Up till now, there have been a number of ways to create blocks of your CSS that are only visible to IE, but they’re, well, hacks, and can’t be trusted to work forever. The best way to add IE-specific styling, I’ve found, is by using conditional comments and an entirely separate stylesheet.
OK. Moving on. Now you know how to add IE-specific styling to your app, let’s find out what that styling should be. You’ve got a passel of problems, but no idea how to solve them. Where do you go for help? You go to the Bible: PositionIsEverything’s list of Internet Explorer bugs and their workarounds, if there are any. Another great article on the subject can be found at CommunityMX (a mostly subscription-only site, unfortunately): How to attack an Internet Explorer (Win) display bug.
For further reading on IE’s display bugs, don’t miss these articles:
Other articles you may find useful:
With all these tools in your arsenal, you can go forth and battle most all the weird browser bugs out there (more on this in another post) and come out the other side alive and even relatively unscathed.
Browsers and browser-plugins
First, let me explain a bit of my strategy. As I said, I code to the standards and then hack for browser compatibility. My first line of defense is Safari/Firefox, and once I have my styles/layout working in both of those, I move on to IE6. I am not supporting IE7 at this time, because our clients run mostly IE6 and have no immediate plans to upgrade.
The browsers alone are not enough, though. Each needs some kind of enhancement or companion tool to help me debug my CSS code.
XyleScope does have some shortcomings, though. One is that it only uses the Safari rendering engine, which gives it limited usefulness when testing cross-browser compatibility. The second—and crucial—one is that it doesn’t deal with generated elements very well. So when I’m testing with my webapp running, and much of the content is generated by my server-side code, XyleScope gives up.