Re: Firefox and User Timing API in workers

On Tue, Jun 2, 2015 at 1:20 PM, Ilya Grigorik <igrigorik@google.com> wrote:

> On Tue, Jun 2, 2015 at 11:14 AM, Boris Zbarsky <bzbarsky@mit.edu> wrote:
>
>> On 6/2/15 2:07 PM, Ilya Grigorik wrote:
>>
>>> anything in Firefox, or are you aware of any
>>> particular deployments that might be affected?
>>>
>>
>> No on both counts.  I just know we've been shipping this for a while and
>> so has Chrome and right now we have compatible behavior and this behavior
>> is not what you want to spec...
>>
>> I guess it's worth checking at least how often performance.now() is used
>> in a worker at all.
>
>
> I'll take a look if we can get some telemetry on that.


Unfortunately we don't have any existing Chrome telemetry on this one
either.

That said, as a proxy data point, I ran [1] some queries against the 08/15
HTTP Archive run [1] (top 300K sites):
- 733 distinct instances of calling "new Worker(...)"
- 315 distinct instances of calling "new SharedWorker(...)"

For SharedWorker, almost all calls were made by the same library (hosted on
different origins), which appears to belong to whiteops.com (ad fraud). For
dedicated workers, there is also a significant portion of calls made by the
same library (foresee.com), plus a tail of JQuery plugins and other ad-hoc
scripts... Most importantly, doing manual spot checks against above
matches, I didn't find any perf now() calls.

As such, I'm inclined to say that the risk here is minimal.. If we do
affect anyone, it'll be a very small and manageable set.

ig

[1] https://gist.github.com/igrigorik/e7676cda3a07f8365360