Repros in canary - It really is quite slow. Has it always been so slow?

Not really, but our app grew a lot during this time. I checked on Chromium 45.0.2454.104 (64-bit) and the results are similar: https://screenshot.googleplex.com/vDkjk4xN0LW

Able to reproduce the issue on windows-7, Mac-10.11.6 and Linux Ubuntu-14.04 using chrome stable version 54.0.2840.99 and latest canary 57.0.2935.3 with below steps.
1.Opened Chrome
2.Navigated to the url  https://adwords-next-nightly.qa.adz.google.com/aw/campaigns?ocid=24390&__u=4113224306&__c=8639657110
3.Scrolled to the bottom of the page.
4.Hovered over the row and clicked the pencil icon
Observed that the popup not open immediately.

This is non-regression issue observed from M30#30.0.1551.0.Hence marking it as Untriaged to get it addressed.

Thanks.

Do you have a test case available to non-googlers?

I just created very simple test:
https://x20web.corp.google.com/~rkj/slow_styles.html
Click start few times and notice how performance degrades drastically. Then add single style and see how slow it is.

Example: https://screenshot.googleplex.com/WmLtW8Z11Lo

Updated the file with 'Use class' checkbox. When you add a style that has only class sellector it is fine. When it is attribute then it is slow. From my other experiments using tagname is also slow (even if the tag is unique and doesn't appear anyway on the page yet).

The test is still inaccessible to me behind corp.google.com.

https://dl.dropboxusercontent.com/u/88085/slow_styles.html

This is because attribute selectors (without being qualified by any other selectors) are put into the bucket of rules which are matched against every element. So, if you have 3000 attribute selectors and 3000 nodes we will try to match selectors 3000*3000 times during a full style recalc.

FWIW, Gecko is slower than Blink in the dropbox case when adding 3000 nodes.

In order to fix this, we would need to group attribute selector rules in a hash map based on attribute like we do for class, id, and type selectors.

See  issue 541041 .

Is there a plan to do this, or we should move Angular to use shimming with classes instead of attributes?

@meade this might be something we should prioritize?

I've put it on our backlog to be discussed when we make 2017 plans.

This is super fast now since we removed style sharing and uncommon_attribute_selectors.