A profile of the JS execution time in CSS/CSSPropertyUpdateValue.html benchmark reports:

Running Time	Self (ms)		Symbol Name
10986.0ms   11.1%	577.0	 	     v8::internal::LookupIterator::PropertyOrElement(v8::internal::Isolate*, v8::internal::Handle<v8::internal::Object>, v8::internal::Handle<v8::internal::Object>, bool*, v8::internal::LookupIterator::Configuration)

which means that DOM code that's setting inline styles for animations took a 10% perf hit by this change. We need to revert this change, or switch to following the spec and have real getter/setters or something. Right now Chrome matches neither Firefox, Safari or Edge.

This read through behavior in general is very strange, it seems pretty broken that setting a value on Object.prototype would break all of the instance interceptors.

ex. 

Object.prototype.test = "wtf";
localStorage.setItem("test", "should be test")
localStorage.test == "wtf";

but

Object.getOwnPropertyDescriptor(localStorage, "test").value == "should be test"

so JS is telling you the instance has this property, but setter/getting it always goes up the prototype chain. Technically Proxies are capable of this behavior, but it's super confusing, and results in a 10% performance slowdown on *everything* that has an interceptor.

Hmm so I can see why localStorage wants this:

localStorage.setItem("setItem", "10")
localStorage.setItem == Storage.prototype.setItem;

though the spec does result in some confusing behavior:

// Actually sets a data property, since hasOwnProperty is on the prototype chain.
localStorage.hasOwnProperty = "test";
localStorage.getItem("hasOwnProperty") == undefined

In either case CSSStyleDeclaration isn't supposed to use an interceptor and do named lookup like localStorage at all, so Chrome isn't following any spec with it's current behavior.

We should probably switch to having actual getter/setters so we at least follow the spec, and also so we don't take the 10% perf hit.

timloh@, could you triage this issue?  I think this is an issue of core/css/, at least it's CSS team who made a decision to implement CSS attributes as named properties.

These have been named properties forever*. What's the correct way to implement these?

* https://chromium.googlesource.com/chromium/src/+/b6cf53%5E%21/

Per the original report at #0, I suppose the following is the spec to be implemented.
https://drafts.csswg.org/cssom/#ref-for-cssstyledeclaration-5

Then, each _camel_cased_attribute must be an ordinary regular DOM attribute.  You should list all of them up in the .idl file.

Seeing the definition of CSSStyleDeclaration interface as below,
https://drafts.csswg.org/cssom/#cssstyledeclaration
CSSStyleDeclaration interface must not support named properties (though it supports indexed properties).

I haven't had time to look into this, unassigning myself.

It sounds like this isn't really P1. Dropping priority.

I'm not really sure why this was assigned to me... I'm not likely to work on it any time soon so removing myself.

Seems like this is now an interop issue because we're not conforming to a spec?

This issue has been Available for over a year. If it's no longer important or seems unlikely to be fixed, please consider closing it out. If it is important, please re-triage the issue.

Sorry for the inconvenience if the bug really should have been left as Available.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot