selector matching issues

[balazs.endresz]

The ​patch that prevents selectors to match inside square brackets causes some other problems:

• If the value contains a dot and an opening bracket then the class selector regex will match the string after the dot:

input[name=".types["]

• Also it won't match .foo in this expression:

.foo #bar\]

[balazs.endre]

Possible solution to the latter: /(?![^\\\[]*\])(?![^\\\(]*\))/

But this way the class regex (and the others) will wrongly match if a backslash is present after a dot inside an attribute selector: ´[foo=.ba\r]´ (besides the opening brackets, as in the first issue)

[balazs.endresz]

I've been thinking about the first issue and came up with a simple solution to detect if there's still an erroneous match inside an attribute selector. It doesn't impact speed - maybe only in these edge cases a bit - but the only problem is that it should be in the Sizzle.filter function, which isn't really the place where this should ideally happen. Here's a standalone code that demonstrates the logic:

var classregex = /\.((?:[\w\u0128-\uFFFF_-]|\\.)+)(?![^\\\[]*\])(?![^\\\(]*\))/;
var attrregex = /\[((?:[\w\u0128-\uFFFF_-]|\\.)+)\s*(?:(\S?=)\s*(['"]*)(.*?)\3|)\]/;
var expr = 'input[name=".types["]' //original example
var expr = "input[name='.types\\.asdf']";
 //this would match ".types" too beacause of the fix for `.foo #bar\]` in the previous comment
var match = classregex(expr);

//
// and in Sizzle.filter the following would go after 
// `if ( (match = Expr.match[ type ].exec( expr )) != null ) {`
//

var nextpos=match && match.index + match[0].length;
if( nextpos && expr.charAt( nextpos ).match(/[\[\(\\]/) ){
  var isattr=attrregex(expr);
  if( isattr && ( (isattr.index + isattr[0].length) > nextpos ) )
    console.log("false positive"); // continue;
}

[balazs.endresz]

It's getting even more interesting :) I tried to apply the above but the chunker splits this selector into two:

input[name=".types["]

becomes

input and name=".types["]

Now looking at the chunker this part: [^ >+~,(\[]+ is interesting -- why does it have to split the selector by the openening braces? Removing (\[ seems to solve this but I guess they have some purpose...

Btw, looks like people are already running into this bug: ​http://groups.google.com/group/jquery-en/browse_thread/thread/eefe67dca10c8163

[john]

I moved the discussion of input[name^{="types["] style selectors to bug #3928.}

[dmethvin]

Per John in #3928, fixed in SVN rev [6143].

[balazs.endresz]

I reopen this because #3928 was just a related problem. The original issue comes up when there's an opening bracket and a dot too in the attribute filter beacause that causes the class filter to match inside it (and the same goes for colons and pseudos).

[ekauffma]

I am still seeing this issue in 1.3.2.

It appears that as long as a left bracket occurs after a dot (even if it is not the first character in the string) the element will not be properly selected. For example:

div[name='a.b[c'] will fail to select

whereas:

div[name='a.b]c'] works fine...

[snover]

Confirmed. ​test case

[john]

Will check this out later, post-1.5.

[dmethvin]

Confirmed in bug triage, but low priority.

[dmethvin]

#5482 is a duplicate of this ticket.

[Timmy Willison]

valid if escaped: ​http://jsfiddle.net/yjCW7/1/