Re: User Timing Mark Name for "Critical Content Loaded"?

I don't want to derail this thread into other topics, but we should
probably address a few things here.

Performance markers, while they aren't specific from the spec that they
can't be used as platform indicators, are one of the closest things we have
to being able to use as an indicator. If they shouldn't be used as such,
that should probably more explicit, but having an API for indication would
be helpful. A use case I always come back to deals with painting: Why can't
a UA delay first paint until the application designates itself as visually
loaded? This would help avoid flashes of uninitialized content among other
things.

But I digress. The reason Firefox OS diverged from the recommended mark
names was that they weren't sufficient in describing the more complex
interactions that are involved in the web-as-OS paradigm. At the time it
seemed that modification of the recommended mark names wasn't viable at the
time, coupled with the inconsistencies and differently-focused definitions,
adopting different metric names is reasonable. They are custom markers
after all. They don't technically need to have special meanings to anyone
outside of Firefox OS.

That said, the markers have been useful for our analytics and opened up the
ability for richer performance testing of Firefox OS. I just wrote a blog
post today [1] about these challenges. We aggregate the metrics in
dashboards [2] with data gathered from User Timing numbers.

Overall though, I'm open to any discuss any direction that anyone feels is
beneficial for tooling, platforms, or moving the web forward.

[1] http://eliperelman.com/performance-testing-firefox-os/
[2] http://raptor.mozilla.org

Thanks,

Eli Perelman
Mozilla


On Wed, Jun 24, 2015 at 2:32 PM, Philippe Le Hegaret <plh@w3.org> wrote:

> On 06/24/2015 05:06 PM, Jonas Sicking wrote:
>
>> I still think that the bigger question isn't what string value to use,
>> but whether it is appropriate to say that certain "mark" names have
>> semantic meaning or not.
>>
>> The way that the spec currently works, from my understanding, is that
>> it says "you can use whatever name of a mark you want. The names have
>> whatever meaning you assign to them. Except if you use names X, Y or
>> Z, those have a very specific meaning and will cause A, B and C to
>> happen.".
>>
>
> I actually don't believe the spec says that it will cause anything besides
> recording the time of the mark. They were more intended to help web
> performance analytics [1] [2]. As such, the developers have been encouraged
> to use them but we made it clear that the user agent does not validate that
> the usage of those marks is correct.
>
>  Another way to put this is that it feels weird that we have an API
>> which allows a page to add page-defined marker to a timeline. Except
>> that if you give those marker a specific name, it affects when the UA
>> render the page, which is a functionality completely unrelated to the
>> timeline (other than that both functionalities involve "time").
>>
>
> It shouldn't affect the UA. If it did, this went beyond the original
> intent.
>
> Imho, we should ask ourselves if those marks have been useful for
> analytics. It seems that the answer from Firefox OS is "no thanks, we made
> our own".
>
> It would be useful to have some measures of usage out there of those marks
> and also get a picture of which analytic tool, if any, is actually using
> them (or if they recommend a different set as well).
>
> Based on that, we could either keep what we have, drop it, or recommend a
> new set of marks for analytics. If we want to go beyond that, ie actually
> provide hints for user agents, that would be a new direction imho.
>
> Philippe
>
> [1] https://lists.w3.org/Archives/Public/public-web-perf/2011Jan/0038.html
> [2] https://lists.w3.org/Archives/Public/public-web-perf/2011Mar/0003.html
>