performance regression with comma-separated selectors

[[email protected]…]

There appears to be a performance regression in v1.8.2 with regard to complex comma separated selectors. I've prepared a demo in which such a selector takes 2 orders of magnitude more in v1.8.2 than in v1.7.2 (running into several seconds). See - ​http://jsfiddle.net/yuvalshirav/y97n9/1/

[timmywil]

This is a known regression. On a page with fewer elements, the current implementation is actually faster. The original idea was to remove the need for document sorting. However, given positional selectors and the fact that jQuery.unique is a public method, this wasn't possible so we could switch it back to recursive handling of the selector split by commas and sorted afterwards. We'll see.

Ported to Sizzle issue: ​https://github.com/jquery/sizzle/issues/164