This is hawt. I love the way you use the expression in the while statement though it could be done with a do-while for clarity. Very cool :{D

hawt +1

This would be even awesomer with some in-line notes explaining the magic. :) Nice work!

Pretty sweet, especially useful since I've seen some user agents where it has two different IE versions listed. http://www.zachleat.com/web/2008/10/19/jquery-bug-ie-reports-incorrect-browserversion/

A bit smaller and returns false instead of undefined for non IE browsers:

var ie = (function() {

    var v = 3, div = document.createElement('div'), a = div.all || [];

    while (div.innerHTML = '<!--[if gt IE '+(++v)+']><br><![endif]-->', a[0]); 

    return v > 4 ? v : !v;

}());

neato

@zach: Ben Alman, Resig and I revised the UA sniffing for jQuery 1.4 in January so the issue from your post has since been resolved.

I'll take a crack at documenting James' original:

var ie = (function(){

    var undef, v = 3, div = document.createElement('div');

    // the while loop is used without an associated block: {}
    // so, only the condition within the () is executed.

    // semicolons arent allowed within the condition,
    //   so a comma is used to stand in for one
    // basically allowing the two separate statements 
    //   to be evaluated sequentially.

    while (
        div.innerHTML = '<!--[if gt IE '+(++v)+']><i></i><![endif]-->',
        div.getElementsByTagName('i')[0]
    );

    // each time it's evaluated, v gets incremented and
    //   tossed into the DOM as a conditional comment
    // the i element is then a child of the div.

    // the return value of the getEBTN call is used as 
    //   the final condition expression
    // if there is an i element (the IE conditional
    //   succeeded), then getEBTN's return is truthy
    // and the loop continues until there is no 
    //   more i elements.

    // In other words:  ** MAGIC**

    return v > 4 ? v : undef;

}());

Yeah, the jQuery team was very quick to patch it up, just offering as evidence that UA strings can be corrupted.

+1 on the hawtness

Cool !
The "&&" operator could be used instead of the comma as another variation to continue the expression in the while.
But yes the comma is one byte less ;-)

Nice comments Paul!

this is hawt. My first inclination was to replace the d.isIE sniff for this, but have to consider CLA. bummer.

To compliment Paul's explanation I will get a bit more specy.
Paul is mixing up the terms Statement and Expression in his explanation of the while statement.

A while statement is made of the following parts
while ( Expression ) Statement

In James' example
while(a.prop=c, d.method()[0]);

spec-ifies to

while ( AssignmentExpression, Expression ) EmptyStatement

and could be broken down to

while ( LeftHandSideExpression = AssignmentExpression, CallExpression [ Expression ] ) EmptyStatement

and could be broken down even further.

There is no semi-colon in the expression because that would create an ExpressionStatement which is a Statement not an Expression.

Sorry about not adding comments to the original -- thanks Paul for adding some!

Thanks @jdalton for the improvement! I forgot about the whole live NodeList thing... very cool!

Awesome script!

Your comment If you're in IE (>5) is wrong. It should read If you're in IE (>=5)

@James you should totally sign a CLA so this can get added to dojo. It's superduper easy!

@Marcel, thanks, corrected.

@SlexAxton, Okay, where do I go to sign?

@jdalton, just noticed -- with yours, since it returns false, expressions like ie < 7 cannot be used because ie will be false and Number(false) is 0, and therefore the expression ie < 7 will return true in non-IE browsers.

I chose undefined as the non-IE return value because Number(undefined) is NaN and this means the following expressions can be used everywhere (without wrong results in non IE browsers):

 ie === 6 // IE6
 ie > 7 // IE8, IE9 ...
 ie < 9 // Anything less than IE9:

@dperini, plus, using && would require the first expression (div.innerHTML = ...) to be wrapped in parens, as && has a higher precedence than =.

@james good point on false vs undefined

@james http://www.dojofoundation.org/cla/ -- You can just scan it or take a picture of it or something. Then you can be a Dojo Committer!

YUICompressor hates and ClosureCompiler strips (!) the /*@cc_on conditional compilation action..

Not the end of the world here, but fair warning.

Why do you need that early exit anyway? What does it add, apart from a very slight optimization of a function that probably runs only once?

@Marcel I believe your suggestion is correct, comparing IE4 to a new browser is wrong anyway ;-)

return v > 4 || !v;

should be just enough (and shorter).

Just got rid of the early exit (with conditional compilation).

@dperini, that'd return false for non-IE browsers...

Let's get small

Really handy! Thanks

slightly improve to get smaller: http://gist.github.com/548648

How about simply returning NaN instead of an undefined variable?

Great stuff, what's the license? Or is it so small, that it would fall under public domain?...

@balupton, no worries, public domain.

Sweet thanks a lot! and thanks Paul for showcasing this in your latest video :-)

This bit has actually been incorporated into https://github.com/slexaxton/yepnope.js in the optional !ie prefix.

It's a fantastic bit of javascript.

Came back to use this, since it's awesome.

But I'm wondering why your counter starts at 3, when you return undefined for <=4. Shouldn't the counter start at 4?

Because James uses a pre-increment to increment v:

div.innerHTML = '<!--[if gt IE ' + (++v) + ']><i></i><![endif]-->'

so the first test actually runs with 4; when the while loop exits immediately, you're probably not using IE at all.

Would he have used a post-increment, then your assumption is correct, but it has the disadvantage that the version number should be decremented afterwards, otherwise the loop will end with v being incremented after the condition to exit the loop has been met.

Ah, sure. My bad. Since I've never encountered a false positive for $.browser.msie, I had used that to toggle the test, and only used this script as a more reliable way to find the version number. Thanks.

Anyone know how to make this Lint?

Why? This is valid JavaScript. JSLint is just very restrictive.

@Marcel: I didn't ask why... just wondered if anyone ported this to Lint-friendly code.

No, you didn't, but I did. ;)

Anyway, just work around the issues JSLint complains about; e.g., only use a comparison as the condition within parentheses and do the assignment inside a statement, like:

do {
    div.innerHTML = '<!--[if gt IE ' + (++v) + ']><i></i><![endif]-->';
} while (all[0]);

and make ie a local variable.

Unfortunately this won't work for IE10 as they've dropped support for conditional comments:
http://blogs.msdn.com/b/ie/archive/2011/07/06/html5-parsing-in-ie10.aspx

That's hopefully a good thing since IE10 shouldn't need singling out compared to it's predecessors.

Unfortunately IE10 still needs to be targeted.

Here's on of my latest problem with all versions of IE (including R10 as of today):
http://jsfiddle.net/wG8Rd/2/

Works perfectly in Chrome/Safari and Firefox.

"Not supporting transforms" is not an IE quirk. This is a case where feature testing would be much more appropriate than IE sniffing.

That's not the problem. The rotation transform works in IE 7-10 with the supplied example. The problem is that using a transform with opacity and border-radius will mess up the look of the border. (The background won't be transparent.)

border-radius works natively in IE9-10 but they both have the above problem. The easies solution is to skip the opacity, but the browser(s) needs to be targeted for this and thus it's a shame they've removed conditional comments...

You generally won't have a ton of luck using old proprietary DX Transforms and mixing it with anything. It seems like comparing the DX and the CSS3 ones is a little flawed. It's also guaranteed to be phased out eventually. Something like:

.no-css-transforms .flip {
  background-image:url(flipped-badge.jpg);
}

would be a safe way to not just back-fill IE 6-10, but also mobile browsers or any of the slew of browsers that have issues with transforms.

Cool technique, does it matter if IE10 is not detected?

Hmm, I wonder if this could be rolled into an IE CSS expression that adds classes to the document. That way HTML5 boilerplate could lose the comments in the HTML and instead shift it to the CSS.

IE10 still supports JavaScript conditional comments, so you could add something like this right below this code snippet:

// IE 10 no longer supports conditional comments in HTML, but does support JavaScript conditional comments
// This will have to be revisited once IE 11 comes out... :(
if (/*@cc_on!@*/false && !ie) {
    ie = 10;

You could potentially do user agent sniffing to ensure that it's indeed IE10.

For all how what to detect IE10:
It's almost like @Daniel15 suggestion.

(function() {
  "use strict";
  var tmp = (document["documentMode"] || document.attachEvent) && "ev"
       , msie = tmp 
                  && (tmp = window[tmp + "al"])
                  && tmp("/*@cc_on 1;@*/")
                  && +((/msie (\d+)/i.exec(navigator.userAgent) || [])[1] || 0)
  ;
  return msie || void 0;
})();

user-agent sniffing, yeah :(
But this would work with:

• any JS-preprocessors (GCC, for example, will delete all comments including conditional comments)
• with any IDE without errors for "eval" in "strict mode"

And still like @Daniel15 said:
"This will have to be revisited once IE 11 comes out... :("

Looks like I'm late to the party. Here's my party favor anyway.

var ie = (function(){
    // for-loop saves characters over while
    for( var v = 3,
             // b just as good as a div with 2 fewer characters
             el = document.createElement('b'),
             // el.all instead of el.getElementsByTagName('i')
             // empty array as loop breaker (and exception-avoider) for non-IE and IE10+
             all = el.all || [];
         // i tag not well-formed since we know that IE5-IE9 won't mind
         el.innerHTML = '<!--[if gt IE ' + (++v) + ']><i><![endif]-->',
         all[0];
       );
    // instead of undefined, returns the documentMode for IE10+ compatibility
    // non-IE will still get undefined as before
    return v > 4 ? v : document.documentMode;
}() );

Compatible with "use strict" (tested in Firefox 22 and Chrome 27). Uses no user agent string sniffing. Also compatible with all known JS minifiers as it uses no IE-specific JavaScript comments.

Minifies to 171 bytes.

var ie=function(){for(var a=3,b=document.createElement("b"),c=b.all||[];b.innerHTML="<!--[if gt IE "+ ++a+"]><i><![endif]-->",c[0];);return 4<a?a:document.documentMode}();

Unfortunately, it's now so small that gzip -9 yields 175 bytes.

I needed IE10 support, as I was using the imagesloaded plugin in combination with imgrefresh (https://code.google.com/p/image-refresh/) (so to refresh unloaded images in ie), as IE still has issues with cached images! So the check had to be extended to IE10. Here's the code. I'm not a JS guru, so I've probably brutalized it, but it does work now for IE10 -- based on Daniel15's Idea.

If someone wants to make it all sparkly again, by all means.

var ie = (function(){
  var undef,
      v = 3,
      div = document.createElement('div'),
      all = div.getElementsByTagName('i');
  while (
    div.innerHTML = '',
    all[0]
  ) { 
    ver = v;
  }
  //ie10 support
  if (/*@cc_on!@*/false && typeof ver === 'undefined') {
    ver = 10;
  } else {
    ver = 3;
  }

  return ver > 4 ? ver : undef;
}());

Regarding IE 10 detection and following @Daniel15's idea, we can use @jscriptversion instead of the UA (see http://msdn.microsoft.com/en-us/library/ie/s59bkzce%28v=vs.94%29.aspx):

var ie_version = (function() {
    var undef,
        v = 3,
        div = document.createElement('div'),
        all = div.getElementsByTagName('i');

    while (
        div.innerHTML = '<!--[if gt IE ' + (++v) + ']><i></i><![endif]-->',
        all[0]
    );

    return v > 4 ? v : undef;
}());
/*@cc_on
if (typeof(ie_version) == 'undefined') {
    ie_version = parseInt(@_jscript_version);
}
@*/

Note that we could only use that last if statement, but then that would return incorrect version numbers when changing the browser mode.

As mentioned before, this may break in IE 11.

Beware I wasted an awful of time on a IE issue related to requirejs trying to eval the last version of this gist submitted by @ecstaticpeon. So if you use this through requirejs and have a http://requirejs.org/docs/errors.html#fromtexteval error msg, it's definitely because of /*@cc_on I just added a space like this /* @cc_on and now the eval() pass. But I didn't tested on IE10 yet... anyway I don't like this conditional compilation stuff, I'll try to find something else.

@nrako: thanks for the heads up. But adding a space like you did does break the code. Try it on http://jsfiddle.net/P5z4N/4/

@nrako: here is a different version that might work for you :)

Try it on http://jsfiddle.net/e8G3S/1/

// Returns the version of Internet Explorer or a -1 (indicating the use of another browser)
function getInternetExplorerVersion(){
  var rv = -1; // Return value assumes failure.
  if (navigator.appName == 'Microsoft Internet Explorer'){
    var ua = navigator.userAgent;
    var re  = new RegExp("MSIE ([0-9]{1,}[\.0-9]{0,})");
    if (re.exec(ua) != null)
      rv = parseFloat( RegExp.$1 );
    }
  return rv;
}
function displayAlert(){
  var msg = "You're not using Internet Explorer.";
  var ver = getInternetExplorerVersion();
  if ( ver > -1 ){
    if ( ver <= 8.0 ){
      msg = "You're using Internet Explorer 8 or below" ;
    }
    else if ( ver >= 9.0 && ver < 10.0 ){
      msg = "You're using IE 9 or above";
    }
    else{
      msg = "You're using IE 10"; 
    }
  }
  alert( msg );
}

@adrien-be indeed it fail on IE10, the last code you sent is quite different it's solely useragent sniffing, and I still love the conditional comment better, I (and requirejs) just don't like the condition compilation stuff for IE10 detection.

Smaller yet? https://gist.github.com/devxdev/6506658

I Can't test I'm on Ubuntu. maybe smaller if undefined isn't required in the last ()

The expression in the while statement was done proper though for clarity I'd use a do-while.
Interesting :{D, fail on IE10, the last code you sent looks odd to me mainly because agent sniffing, I find the conditional comment interesting, I and requires just don't like the condition compilation stuff for IE10 detection. Because James uses a pre-increment to increment v:

I wanted this script to include IE10+11 as well, and my tests works fine by just returning document.documentMode instead of undefined, since it will be undefined for non-IE browsers anyway (I also prefer do...while...):

var ie = (function() {
  var v = 3
    , div = document.createElement( 'div' )
    , all = div.getElementsByTagName( 'i' )
  do
    div.innerHTML = '<!--[if gt IE ' + (++v) + ']><i></i><![endif]-->'
  while 
    (all[0])
  return v > 4 ? v : document.documentMode
}())

So I tried the code above, also davidhellsing's version, and I tested this on ie8, and it always says it's version 7, when it's 8. So I don't think this works on ie8.

Also, just tried it in ie10 and ie gets declared as undefined.