This refreshes all the css on a page by changing the query string

refresh_css.js

/*
Add a bookmark to this

javascript:(function(){var c=document.getElementsByTagName("link");for(var d=0;d<c.length;d++){var a=c[d];if(a.rel=='stylesheet'||a.type=="text/css"){var e="css_buster_"+Math.floor(Math.random()*1000000000);var g=a.href.split("?",2);var f;if(g.length>1){var b=g[1].indexOf("&")==-1;if(b){f=e}else{f=g[1]+"&"+e}}else{f=e}a.href=g[0]+"?"+f}}})();
*/
(function() {
  var links = document.getElementsByTagName('link');
  for (var i = 0; i < links.length; i++) {
    var l = links[i];
    if (l.rel == 'stylesheet' || l.type == 'text/css') {
      var rand_n = 'css_buster_' + Math.floor(Math.random() * 1000000000);

      var splitted = l.href.split('?', 2);

      var new_query_string;

      // Does it already have a query string?
      if (splitted.length > 1) {
        // does it have params or just a cache buster
        var has_amp = splitted[1].indexOf('&') == -1; // is it a query string? 
        // if it's just a cache buster, swap it out
        if (has_amp) {
          new_query_string = rand_n;
        } else {
          new_query_string = splitted[1] + '&' + rand_n;
        }
      } else {
        new_query_string = rand_n;
      }

      l.href = splitted[0] + '?' + new_query_string;
    }
  }
})()

type attribute is optional and not authoritative. You should be finding stylesheets by rel=stylesheet, not type.

@pornel good call. Made change

Check out ReCSS, basically the same but uses the current date (down to millisecond) instead of random numbers and doesn't get as complicated with the query string detection.

@pornel Type is optional, but it is also authoritative. If a type of foo/bar is encountered, it should be ignored. In the future, everyone might be using CSSX (fictional), which uses XPath selectors instead of weak CSS selectors, and it would go in a rel=stylesheet link but might have a type of application/cssx. The if statement should be changed to l.rel === 'stylesheet' && (l.type === 'text/css' || !l.type).

Edit: Just realized that this code is stylesheet language-agnostic. Keep it as-is ;)

@eligrey No, only rel matters. I can have:

<link rel=prefetch type=text/css href=x>

which is not a visible stylesheet, and there's no point forcing this to reload.

Type in HTML is only a hint. HTTP always overrides it. Besides, since HTML theoretically allows other stylesheet types than CSS, why take extra care to break that?

If check for rel=stylesheet seems unbelievably simple, then you might add check for rel="alternate stylesheet" (taking into account it's a space-separated token list) and link.disabled.

@pornel I never said type=text/css alone signifies a stylesheet. I said "Type is optional, but it is also authoritative. If a type of foo/bar is encountered, it should be ignored". That means that "type isn't pointless", as you may have <link rel=stylesheet type=foo/bar href=x>, which should be ignored by a browser unless it actually supports a foo/bar stylesheet language. Also, if you read my if condition replacement, you'd see that I didn't drop requiring a rel=stylesheet. Though the thing is, you can't tell what types a browser supports, so the best way to check if a link is an active stylesheet would be l.sheet && !l.disabled.

var splitted = l.href.split('?', 1); : splitted array limit is 1, so this condition can never happen : splitted.length > 1. or i am missing something ?

The second parameter to split specifies the limit of the number splits, not the limit of the returned array. (In other words, the returned array is one greater than the number of splits.)

I'm not sure i understand what you say. Actually i never used that parameter before. but here is what i got with Rhino :

js> "/style.css?foo".split('?', 0).length 
0
js> "/style.css?foo".split('?', 1).length
1
js> "/style.css?foo".split('?', 2).length
2
js> "/style.css?foo".split('?', 3).length
2

@jney Oh dang, you are indeed correct. I assumed the behavior was similar to python's .split. I guess JavaScript, Java, and Ruby is the number of elements max, whereas python is the number of splits max. Fixing code.