jquery / sizzle

A sizzlin' hot selector engine.

Sizzle

A pure-JavaScript CSS selector engine designed to be easily dropped in to a host library.

• More information
• Documentation
• Browser support

Contribution Guides

In the spirit of open source software development, jQuery always encourages community code contribution. To help you get started and before you jump into writing code, be sure to read these important contribution guidelines thoroughly:

1. Getting Involved
2. JavaScript Style Guide
3. Writing Code for jQuery Organization Projects

What you need to build Sizzle

In order to build Sizzle, you should have Node.js/npm latest and git 1.7 or later (earlier versions might work OK, but are not tested).

For Windows you have to download and install git and Node.js.

Mac OS users should install Homebrew. Once Homebrew is installed, run brew install git to install git, and brew install node to install Node.js.

Linux/BSD users should use their appropriate package managers to install git and Node.js, or build from source if you swing that way. Easy-peasy.

How to build Sizzle

Clone a copy of the main Sizzle git repo by running:

git clone [email protected]:jquery/sizzle.git

In the sizzle/dist folder you will find build version of sizzle along with the minified copy and associated map file.

Testing

• Run npm install, it's also preferable (but not necessarily) to globally install grunt-cli package – npm install -g grunt-cli
• Open test/index.html in the browser. Or run npm test/grunt test on the command line, if environment variables BROWSER_STACK_USERNAME and BROWSER_STACK_ACCESS_KEY are set up, it will attempt to use Browserstack service (you will need to install java on your machine so browserstack could connect to your local server), otherwise PhantomJS will be used.
• The actual unit tests are in the test/unit directory.

Developing with grunt

• npm run build or grunt will lint, build, test, and compare the sizes of the built files.
• npm start or grunt start can be run to re-lint, re-build, and re-test files as you change them.
• grunt -help will show other available commands.

sizzle's Issues

Create 1.0 tag

Great work! Small request: could you please create a tag for the 1.0 version?

Test doesn't work from download

Following the instructions (i.e. open test/index.html) results in an error: "module is not defined". I'm guessing this is because qunit isn't in the zip file.

Filter VML on IE7+8

Brief Description:
VML elements within jQuery break Sizzel.selectors.filters.* internally when elem.type is accessed.

Story Description:
I know VML is a topic that tends to get ignored, but it seems to be breaking Sizzle.selectors.filters.* (particularly where ever elem.type is called). I'll try and liven it up a bit. If a VML object happens to wander into a jQuery element and is asked to be filtered, IE will exclaim, "Um, What? You have a .type? Why didn't anyone tell me?!". What's worse is IE will wander off your jQuery chain gang and into exception land where no more work can be done. And even that is fine as long as it's 5 o'clock and it's time to head home. But it turns out this isn't such a good thing at any other time of the day. Can we find a way for VML not to wander into jQuery, and if it does, not ask it about it's .type? Maybe we can find something else to ask it first.

CSS3 Selector compatibility

I believe the move to CSS3 selectors has become huge and, so far, Sizzle lacks support to several of the new selectors. The table on http://selectivizr.com/ shows how far behind it is compared to other Javascript frameworks. It shouldn't be too much of a hassle, and it would fix some of the issues in the issue queue.

Edit:

:lang()
:link
:visited
:target
:hover // this requires adding a special event to the DOM in unsupported browsers, we're not going to do this
:nth-last-child
:nth-of-type
:nth-last-of-type
:first-of-type
:last-of-type
:only-of-type
:root

Support for Namespaces

It would be useful to support Namespaces for selectors, preferably the CSS Namespaces Module if possible, or some similar method.

Consider the following elements:

<atom:entry xmlns:atom="http://www.w3.org/2005/Atom" />

and

<entry xmlns:"http://www.w3.org/2005/Atom" />

The selector atom:entry is required for the former, and entry for the latter. Without knowing the exact prefixes used, it can be difficult to parse namespaced documents.

My suggestion, based on the CSS Namespace Module, would be something like this:

Sizzle.namespace("http://www.example.com/foo"); // Sets the default namespace
Sizzle.namespace("bar", "http://www.example.com/foo"); // Sets the namespace for the `bar` prefix

Sizzle("bar|foo") // Matches `foo` in the namespace denoted by `bar`
Sizzle("*|foo") // Matches `foo` in any namespace
Sizzle("|foo")  // Matches `foo` with no namespace
Sizzle("foo"); // Matches `foo` in the default namespace

selector doesn't work if id contains dot "."

line 1110
return makeArray( context.querySelectorAll( "#" + nid + " " + query ), extra );

needs to escape "." in nid

IE6 Test Framework Fail

In IE 6.0.2900 the code below will (wrongly?) alert "splat". Unfortunately the testing framework for Sizzle follows this exact same pattern by defining window.jQuery in sizzle/test/jquery.js (line 5568) and then conditionally redefining jQuery using var in /sizzle/test/data/testinit.js (line 1).

In the case of sizzle this makes it impossible to run any unit tests in IE 6. Removing the var from /sizzle/test/data/testinit.js (line 1) will avoid the issue.

<script type="text/javascript">
    window.jQuery = 'hello';
</script>

<script type="text/javascript">
    var jQuery = window.jQuery || 'splat';
    alert(jQuery);
</script>

[feature] Find + or ~ element using context in same generation

If I want to find all the paragraphs after a certain element, using that element as a context to search from, this is currently not possible, it seems to me.

Suppose you had this HTML:

<div>
  <p>1</p>
  <p>2</p>
  <p>3</p>
  <p>4</p>
</div>

If I wanted to get the last two paragraphs and had paragraph no. 3 already referenced as a DOM element (let's call it "para"), I'd like to be able to do this:

$('~ p', para);

This might sound like a silly/niche requirement, but it's useful when using events that have been applied to lots of similar elements in a container, and you want to access all the siblings after the one that was clicked (or whatever the event was).

EDIT:

This also applies to the child selector, >. If I wanted to access all SPAN elements within a paragraph (that I had as a DOM element), but only direct children (not grandchildren, etc).

<p>
  <span>A</span>
  <span>B</span>
  <a href="#"><span>C</span></a>
</p>

In the example above, only A and B would be selected, using a syntax like this:

$('> span', para);

Again, this is useful if the exact identity of the "para" element is not known.

[Feature] Using DocumentFragment as context

If you use a document fragment as context, Sizzle fails because it tries to use context.getElementsByTagName().
Making it possible would mean using a getElementByTagName(context, tagName) function either using context.getElementsByTagName(tagName) or checking in context.childNodes which ones have the good tagName depneding on if there the method is available or not.
Or checking if element is fragment and if it is, adding the method working as described above.

:focus selector for non-qSA case

See here:
http://bugs.jquery.com/ticket/8105

Sizzle.matches doesn't work with has attribute selector

Given a document containing the following element:

The following lines returns:
var input = Sizzle('[required]')[0]; //input#field
Sizzle.matches('[required]', [input]); // [ ]

Sibling selector bug

Sorry for the long absence,

I think there's still an open bug in sibling selector, when the search match is in the checkSet (this means that "div ~ div" doesn't work).

This is fixed removing the following codes from dirNode and dirNodeCheck:

1: (it becames obsolete)
var sibDir = dir == "previousSibling" && !isXML;

2:
if ( sibDir && elem.nodeType === 1 ) {
elem.sizcache = doneName;
elem.sizset = i;
}

It was a speed optimization I made, but it fails under that particular condition.
My apologies,

--Fabio

Attribute selectors fail in IE when attribute name looks like ns:attr

Attribute selectors fail in IE when attribute name looks like ns:attr and attribute doesn't exist.
For example: Sizzle("[ui\\:custom]") doesn't works when selector find some table without attribute ui:custom
And elem.getAttribute( name ) fails too when name like ns:attr for any type of DOM elements, so it's better to use elem.attributes.getNamedItem

I made changes to my copy of https://github.com/jquery/sizzle/blob/1.5.1rc1/sizzle.js in lines 803 — 805

was

803:                      elem[ name ] != null ?
804:                          elem[ name ] :
805:                          elem.getAttribute( name ),

now

803:                      "TABLE" !== elem.nodeName && elem[ name ] != null ?
804:                          elem[ name ] :
805:                          elem.attributes ?
806:                              elem.attributes.getNamedItem(name) :
807:                              null,

Works fine

:nth-of-type()

Sizzle comes with a an :nth-child() selector, but lacks an :nth-of-type() selector.

This doesn't seem to be very easy to implement as a custom selector in jQuery either. To illustrate the difference between :nth-child() and :nth-of-type(), as well as some of the problems in trying to write a custom jQuery filter for this, I'd like to refer to this Stack Overflow topic: :nth-of-type() in jQuery / Sizzle?

IE 6, IE 7: Matching against element.className only works on space-separated classes (and not with all whitespace characters)

If an element has a classname such as "foo\nbar", it is not possible to query for it by the selectors ".foo" or ".bar".

The bug is rather obvious: the matching is done by checking " ", and not by other whitespace characters such as newline, tab etc.

Proposed fix: match by regexp instead, example: /\s<class_escaped>\s/.test(" " + element.className + " ")

Work-around: wrap every class in spaces in the class-attribute on the element.

Inconsistency with results between invalid selectors

This is related to jQuery bug ticket 7205 ( http://bugs.jquery.com/ticket/7205 ).

In essence, there is a discrepancy between what is resultant from invalid selections ">child" and when passing a type selector to a filter's selection such as "type:has(>child)".

I have not yet researched broad enough into the issue to see if this exists outside of the :has() filter, or can be reproduced with other selectors.

Selecting elements by empty attribute in a context other than document fails

Dear Sizzle Devs,

While working with Sizzle I found a minor but annoying issue.

I have a dom tree that looks like this:

<div id="foo">
  <span data-foo></span>
</div>

Now when I execute this:
Sizzle("[data-foo]").length;

I get 1, which is obviously correct, but when I provide a context:
Sizzle("[data-foo]", document.getElementById("foo")).length;

It returns 0, which is wrong.

Some more information:

• It also happens when the attribute has an empty string as value ([data-foo=""])
• It also happens when I provide a tag name in front of the selector (span[data-foo])
• It also happens with other empty attributes like "readonly"

Here's a demo of the problem (including tests with native querySelectorAll):
http://tifftiff.de/sizzle_bug/

The problem exists in my Chrome 6, Safari 5, IE 8 and Firefox 3.6/4.0b.
In IE 6+7 both tests return 0 regardless of the given context.

"[href^=https:]" does not work while "[href^='https:']" does

Syntax error, unrecognized expression: ...

querySelectorAll fails as well: SYNTAX_ERR: DOM Exception 12

Reported from Chrome.

It seems to be failing on the : character.

Check for ":selected" needs check for null parentNode

http://github.com/jeresig/sizzle/blob/master/sizzle.js#L510

If called on a disconnected element it will try to access a property of the null parentNode.

Matching empty attributes doesn't match querySelectAll behavior

When Sizzle is using the browser implementation with querySelectorAll, this example:

<script src="https://raw.github.com/jquery/sizzle/master/sizzle.js"></script>
<a href="/foo" my-attr="">dead link</a>
<script type="text/javascript">
    console.log( Sizzle('a[my-attr]') );
</script>

returns me [a /foo]

However, when using the internal implementation, the result is different (I used this trick to undefine the browsers querySelectorAll implementation):

<script type="text/javascript">
    document.querySelectorAll = undefined;
</script>
<script src="https://raw.github.com/jquery/sizzle/master/sizzle.js"></script>
<a href="/foo" my-attr="">dead link</a>
<script type="text/javascript">
    console.log( Sizzle('a[my-attr]') );
</script>

Returns me: [ ]

This becomes a cross browser issue when e.g. using IE in a HTML document without DOCTYPE; at least in the versions I tested (IE9, IE8) and document without (recognized?) doctype does not expose querySelectorAll so Sizzle defaults to its internal implementation which doesn't match what browsers do.

Issue 24 [1] reads similar but I'm not sure if it's really related.

[1] #24

Selectors with unterminated tokens cause some browsers to hang

This selector input[id=01234567890123456789012345678901234567890123456789 (note that there is no closing ]) causes IE 9 to hang.

I think this is because the regex used suffers from catastrophic backtracking when a match can't be found.

I tested the above using jQuery 1.7. I'm not sure what version of sizzle that applies to.

Proprietary selectors should use a vendor prefix

Sizzle's proprietary selectors should use a vendor prefix to ensure that it won't find itself in a position where it interferes with evolution in the syntax or finds one of its extensions standardized in an incompatible way. http://www.w3.org/TR/CSS21/syndata.html#vendor-keywords

For compatibility, all current selector extensions should be supported in both prefixed and unprefixed variations, but authors should be advised to use the prefixed variation and the unprefixed variation marked deprecated.

:has(:nth-child(2)) throws exception in jQuery > 1.2.6

A :has( ... ) selector containing a :nth-child(an+b) selector fails in jQuery > 1.2.6 but works with jQuery 1.2.6.

test case

This is most likely some regex matching problem with Expr.leftMatch for CHILDREN

Check also the jQuery Ticket #8132 for more info

Sizzle should have standard CSS2.1 compliant regular expression matching "identifiers" /via @diegoperini

'when will jQuery have standard CSS2.1 compliant regular expression matching "identifiers" ? Range \u00C0-\uFFFF is wrong !' - @diegoperini (https://twitter.com/#!/diegoperini/status/68236401879154688)

In CSS, identifiers (including element names, classes, and IDs in selectors) can contain only the characters [a-zA-Z0-9] and ISO 10646 characters U+00A0 and higher, plus the hyphen (-) and the underscore (_); they cannot start with a digit, two hyphens, or a hyphen followed by a digit. Identifiers can also contain escaped characters and any ISO 10646 character as a numeric code (see next item). For instance, the identifier "B&W?" may be written as "B&W?" or "B\26 W\3F".
Note that Unicode is code-by-code equivalent to ISO 10646 (see [UNICODE] and [ISO10646]).

source: http://www.w3.org/TR/CSS21/syndata.html#value-def-identifier

Regex runs infinitely

EDIT: apparently github can't handle inline HTML. The correct post is also placed here:
http://pastebin.com/sbJ5h9R5
======================old, broken message===================
Try the following code in Google Chrome:

<script src="sizzle.js"></script><script> Sizzle("[onclick^*=\"window.open('http://www.firstload.de/\"]"); alert('Done'); // Never fired </script> The 'Done' alert will never be fired. This is because m = chunker.exec( soFar ); never finishes. The cause of this issue, and a fix is explained in http://code.google.com/p/v8/issues/detail?id=1888 by a Chrome developer.

Basically it's just a matter of getting rid of the (3) instances of "+)+" in the regex (which are all non-capturing, so it doesn't change the results of exec either):

chunker, currently
/((?:((?:([^()]+)|[^()]+)+)|[(?:[[^[]]]|['"][^'"]['"]|[^[]'"]+)+]|.|[^ >+,([]+)+|[>+])(\s_,\s_)?((?:.|\r|\n))/g
would become
/((?:((?:([^()]+)|[^()])+)|[(?:[[^[]]]|['"][^'"]['"]|[^[]'"])+]|.|[^ >+,([])+|[>+])(\s,\s_)?((?:.|\r|\n)_)/g

Expose wrapper regex for extension convenience.

Currently new selectors cannot be added via the extension API documented in the wiki.

The issue is the leftMatch object-literal, which is filled by the follow code at initialization time:

for ( var type in Expr.match ) {
  Expr.match[ type ] = new RegExp( Expr.match[ type ].source + (/(?![^\[]*\])(?![^\(]*\))/.source) );
  Expr.leftMatch[ type ] = new RegExp( /(^(?:.|\r|\n)*?)/.source + Expr.match[ type ].source.replace(/\\(\d+)/g, fescape) );
}

We need this code to be publicly exposed so that we can run it after adding new selectors via the extension API.

I've hacked around this by creating a new method (using Sizzle via jQuery):
Expr = $.expr;
Expr.process = function(type) {
Expr.match[ type ] = new RegExp( Expr.match[ type ].source + /(?![^[]_])(?![^(]_))/.source );
Expr.leftMatch[ type ] = new RegExp( /(^(?:.|\r|\n)*?)/.source + Expr.match[ type ].source.replace(/(\d+)/g, function(all, num){
return "" + (num - 0 + 1);
}));
};

which I use like this:

$.expr.match['WIDGET'] = /@((?:[\w\u00c0-\uFFFF\-]|\\.)+)/;

$.expr.preFilter.WIDGET = function(match) {
  return match[1].replace(/\\/g, "");
};

$.expr.filter.WIDGET = function(elem, match) {
  return elem.getAttribute('data-widget') === match;
};

$.expr.process('WIDGET')

Sizzle needs to expose the code that fills the leftMatch object via a function that works something like the one I've shown here.

[style*="whatever"] does not work

The style attribute should be retrieved with getAttribute('style') I guess?
I understand that each browser render css text differently but it is still better than nothing.

No .hasOwnProperty() check in for() loops.

You should use the hasOwnProperty() method in your for(var attribute in object) loops.

Line 212:
for ( var type in Expr.filter ) {/* correction -> */ if (Expr.filter.hasOwnProperty(type)) {

Of course, you have to add a } on line 271 too.

If you don't think it's an issue, try using Sizzle with that code in the same HTML page :

Object.prototype.whatever = function(){};

Selector with '\' always fails

If a selector includes the backslash (''), it will always fails to find any elements. This is true for attribute, class, and id selector expressions:

Test HTML: the foo bar
Expression: Sizzle("#foo\bar") (or Sizzle(".foo\bar"), Sizzle("[id='foo\bar']") )
Expected Result: [...]
Actual Result: []

This is related to jQuery issue http://bugs.jquery.com/ticket/10270.

Sizzle.matchesSelector and Sizzle.matches return inconsistent results

Example can be found at http://jsfiddle.net/ZUs8f/3/ .

JQuery.fn.is function uses Sizzle.matches method when there are several elements and Sizzle.matchesSelector when there is only one. Both Sizzle.matchesSelector and Sizzle.matches use Sizzle function underneath except in one case - https://github.com/jquery/sizzle/blob/master/sizzle.js#L1241 if browser supports it or selector is appropriate then Sizzle uses native matches selector.

But results of Sizzle function are different from mozMatchesSelector and webkitMatchesSelector if we use [value="something"] selector. These native matchers use initial value of the attribute, not the one changed by user as Sizzle does.

Example provided at the beginning shows inconsistent results on chrome/safari/firefox. Simple solution would be to change:
!/!=/.test( expr )
to
!/!=|\[ *value[ =]/.test( expr )

VML on IE7+8

VML objects which make it into sizzle cause errors in IE7 and IE8. A test case can be found here: https://gist.github.com/2b7abeffc4573accc994 (Thank's to jfirebaugh for making the initial tests)

Also reported here: http://www.sitecrafting.com/blog/jquery-cufon-dont-mix/#update_2010-07-14

Tests for broken selectors fail in IE

The unit tests for broken selectors in sizzle test to see if the caught error message contains "Syntax error" but in IE (6 and 8, at least) the error message being thrown is "Object doesn't support this property or method."

:text selector throws an error in IE7 when there is a cross domain iframe on the page

When there is an iframe with a cross domain source, the :text selector fails in IE7 with the error "Permission denied". The problem lies in line 625:

        var attr = elem.getAttribute( "type" ), type = elem.type;

For some reason, getAttribute("type") is not allowed on cross domain iframes in IE7. A possible solution would be to only check the attribute when the element indeed is an input element:

    text: function( elem ) {
        var attr, type;
        // IE6 and 7 will map elem.type to 'text' for new HTML5 types (search, etc) 
        // use getAttribute instead to test this case
        return elem.nodeName.toLowerCase() === "input" && (type = elem.type, attr = elem.getAttribute("type"), "text" === type) && ( attr === type || attr === null );
    },

See http://jsfiddle.net/VvnGm/ for an example

filter() works differently from is()?

Somehow this doesn't work. I'd expect to see in the console 1, true, 1, but instead it's 0 for the one using filter.

<html>
<head>
    <title>jQuery "filter" test</title>
    <script src="https://ajax.googleapis.com/ajax/libs/jquery/1.6.1/jquery.min.js" type="text/javascript" charset="utf-8"></script>
    <script type="text/javascript" charset="utf-8">
        $(document).ready(function() {
            console.log($("[data-test]").length);
            console.log($(document).children().eq(0).is("[data-test]"));
            console.log($(document).children().filter("[data-test]").length);
        });
    </script>
</head>
<body>
    <div data-test>Hi, I'm a div.</div>
</body>
</html>

Attribute selectors fail when they contain single quotes

Attribute selectors fail whenever the selector value contains a single quote. I've added a test case that shows this in markstory@854de47124424b531b3dc136a64a9a39eb2ee7d2

I'll try and make a patch for the solution as well.

chunker.exec(match[3]) is null

Error appears in FF 3.5

on query: "input:not(input[type=hidden]):not(input[type=image]):not(input[type=button]):not(input[type=reset]):not(input[type=submit]), textarea, select"

I can send document html on which I get that error (it's reproducible always on my side)

I've fixed it temporarily on my side by changing:

if (chunker.exec(match[3]).length > 1 || /^\w/.test(match[3]) ) {

into

var res = chunker.exec(match[3])
if ((res && res.length > 1) || /^\w/.test(match[3]) ) {

:input with dot and brackets

When an input element's name contains a dot and brackets, it fails to find this element with :input selector.

Tests for this bug:

t( "Attribute containing dot and []", 'input[name="foo.name[bar]"]', ["hiddenBar"] ); //it passes
t( "Attribute containing dot and []", ':input[name="foo.name[bar]"]', ["hiddenBar"] ); //it should pass
t( "Attribute containing dot and []", 'input[name="foo.name[0]"]', ["hidden0"] ); //it passes
t( "Attribute containing dot and []", ':input[name="foo.name[0]"]', ["hidden0"] ); //it should pass

The elements in index.html:

        <input type="text" style="display:none;" name="foo.name[bar]" id="hiddenBar"/>
        <input type="text" style="display:none;" name="foo.name[0]" id="hidden0"/>

I've noticed this bug when I wrote a testcase in AngularJS: wanted to enter some text into my input which name was "invoice.Customer.AddressLine[0]".

Needs jslint

A couple times now I’ve committed patches that pass the Sizzle selector tests but bomb out with jslint when they get pulled into jQuery; mea culpa for not checking there too, but Sizzle should probably run its own jslint as part of the build.

Support periods in data attribute selectors

Sizzle supports data attribute selectors when they include dashes and underscores, but not when they include periods. This would be useful. Periods are valid in the naming rules.

data-example1.0="This is a valid in HTML5 "
data-example1-0="This is a valid in HTML5 "

$( '[data-example1-0]' ) // Targetable in jQuery.
$( '[data-example1.0]' ) // Throws "unrecognized expression" Sizzle error in jQuery.

IE7 Selector Not Working

This selector:

$(xml).find("routeStop:gt(0) load integer, routeStop:gt(0) unload integer").each(function(index){
});

Throws the following in IE7 (however it doesn't throw an error in IE8 running as IE7):

"Number expected"

...
if (sortOrder) {
hasDuplicate = false;
result.sort(sortOrder); <== error on this line

if (hasDuplicate){
....

If you need more detail I can try to scrounge it up.

Support `:root`

/via @dperini: http://twitter.com/diegoperini/status/158523985703870464

:root should map to document.documentElement.

Where is the minified sizzle?

In the source I only found the original sizzle.js file (26 KB) not the minified one, any idea where to find the minified?

:first selector regression since jQuery 1.4.3

Some adjacent sibling selectors prefixed by :first do not seem to work since jQuery 1.4.3. This regession does not affect selectors prefixed by :first-child.

A test case is available at http://jsfiddle.net/mxcurioni/hryGh/

Failing Unit Tests

Trying to run the unit tests for Sizzle CSS Selector Engine - v1.0 at sizzle/test/index.html results in this failing error stack:

too much recursion

test (line 53)

too much recursion
return (function (a, b) {
qunit.js (line 820)

too much recursion
})(args[0], args[1]) && argument...(this, args.splice(1, args.length -1));
qunit.js (line 830)

too much recursion
if ( ! innerEquiv(a[i], b[i])) {
qunit.js (line 797)

too much recursion
return callbacks[prop].apply(callbacks, args);
qunit.js (line 706)

too much recursion
return bindCallbacks(a, callbacks, [b, a]);
qunit.js (line 826)

This happens in Safari 4.0.3, Mac Opera 9.64 and Mac FireFox 3.0.10, at least.

input:focus test fails when running tests in an unfocused tab

In Chrome 12.0.742.30 dev-m the input:focus test fails when running tests in an unfocused tab.

To reproduce open the qunit test in a background tab and let it run.

1. selector: pseudo - misc (2, 17, 19)
Element focused (input:focus)
Expected:   
[
  <input id="focus-input"></input>
]
Result: 
[]
Diff:   
[
  <input id="focus-input"></input>
] [] 
Source: 
    at t (http://cloud9ide.com/davidmurdoch/sizzle/workspace/test/data/testinit.js:33:2)
:focus Matches

Expr.relative truncates prematurely

Copied from http://bugs.jquery.com/ticket/11109

Test case: http://jsfiddle.net/JBBM8/

Because Sizzle.selectors.relative functions "return" only a single match, Sizzle can be trapped into a false negative with respect to preceding parts of the full selector when the relative selector tests against multiple elements (e.g., "~" and "").

I'm pretty sure that fixing this will require changing the Sizzle.selectors.relative Extension API to allow "returning" arrays, so I will hold off working on this until the core team weighs in with their comments/approval.

Technical explanation: Given a query like "A + B ~ C", Sizzle:

1. Gets a set of all C, then
2. Uses Sizzle.selectors.relative["~"] to replace each element with the first preceding sibling element matching B (or false if nonexistent), then
3. Uses Sizzle.selectors.relative["+"] to replace each element with the preceding sibling element if it matches A (or false if it doesn't), then
4. Returns all elements from the set at step 1 that stayed non-false through all subsequent steps

The problem here is that step 2 fails to report back all preceding siblings matching B, erroneously restricting the input set for subsequent step 3.

img[src=relurl] selectors don't work (Firefox 3)

Assuming the sizzle test suite has an , jQuery("img[src=../speed/logo.png]") returns [], requiring a selector like jQuery("img[src=file:///Users/jhs/sizzle/speed/logo.png]") to return a non-empty array. This is because img.src != img.getAttribute("src").

Above statements at least true in Firefox 3.

:heading selector

I wish Sizzle supported a ":heading" selector, as an alias of the ":header" selector, since (strictly speaking) I think that <h1>, <h2>, etc. are "headings" not "headers"? http://www.w3.org/TR/html401/struct/global.html#h-7.5.5

Incomplete or typo'd bit of code?

While attempting to fix the global variables with JSLint, I found this one:
var ret = aRange.compareBoundaryPoints(Range.START_TO_END, bRange);
where Range doesnt appear to have any reference. It's possibly a typo of aRange?

Memory leaks in IE versions <= 7

Sizzle code contains memory leak patterns as documented in http://msdn.microsoft.com/en-us/library/bb250448(VS.85).aspx - In particular the DOM insertion order pattern.
Eg. - Current IE leaky code:
// We're going to inject a fake input element with a specified name
var form = document.createElement("div"),
id = "script" + (new Date()).getTime();
form.innerHTML = "";

// Inject it into the root element, check its status, and remove it quickly
var root = document.documentElement;
root.insertBefore( form, root.firstChild );

Modified IE no-more-leaky code:
// We're going to inject a fake input element with a specified name
var form = document.createElement("div"),
id = "script" + (new Date()).getTime();

// Inject it into the root element, check its status, and remove it quickly
var root = document.documentElement;
root.insertBefore( form, root.firstChild );
// Add element to DOM before performing innerHTML to prevent memory leaks in IE
form.innerHTML = "<a name='" + id + "'/>";