W3C home > Mailing lists > Public > whatwg@whatwg.org > December 2017

Re: [whatwg] DOM Feature Mod: Add metering / parallelism & throttling options to AddEventListenerOptions

From: David Bruant <bruant.d@gmail.com>
Date: Fri, 1 Dec 2017 10:36:05 +0100
To: Jonathan Zuckerman <j.zuckerman@gmail.com>, Sylon Zero <sylonzero@gmail.com>
Message-ID: <657a3105-66c5-974c-0e84-9dc7d6015f53@gmail.com>
Cc: whatwg@whatwg.org
Le 28/11/2017 à 00:48, Jonathan Zuckerman a écrit :
> You’re probably aware there are libraries that offer functionality of this
> sort (debounce and throttle in underscore/lodash is the one I’m most
> familiar with) and the web community seems content to add a small
> dependency when such functionality is required.
With all due respect, i think it's dangerous to start a sentence by "the 
web community" given how broad and ever-changing this group of human 
beings is.

For one data point, it's been 6 years that i regularly give advanced 
JavaScript trainings to people who are already JS practitionners and 
they don't know about the concept of event queue and barely about basic 
perf advices (reduce round trips to a minimum, minimize the size of 
what's sent, etc.). Debounce/throttling is further down the road.

By the logic of "the web community seems content to add a small 
dependency", the 'once' option [1], probably shouldn't have seen the 
light given how easy it is to rewrite. Same for 'passive'. Same for 
Element.prototype.remove ("DOM4"). etc.
I'm glad they exist though.

David

[1] 
https://developer.mozilla.org/en-US/docs/Web/API/EventTarget/addEventListener

> How would you convince
> browser vendors to implement this?
>
>
> On Mon, Nov 27, 2017 at 18:05 Sylon Zero <sylonzero@gmail.com> wrote:
>
>> *Core Problem Statement* Processor functions that subscribe to events via a
>> topic string may need to be prioritized for processing based on the topic
>> itself. Conversely, certain events may be more numerous but should not
>> limit the ability of the JS environment to respond and process other
>> events, that may be more critical to either the User Experience (UX) or
>> integrity of the system (e.g. events that trigger saving data to a
>> back-end).
>>
>> *Background Information* As Browser/CommonJS environments bring paradigms
>> like UI event handling and back-end event handling into the same problem
>> space, it is useful to apply common patterns used in Message-based
>> Publish-Subscribe environments with message brokers, which is what the JS
>> execution context often behaves as. The use case here is to ensure that one
>> kind of event (e.g. event listeners for a ‘mouseMove’ event) don’t saturate
>> or delay execution of other events (e.g. ‘dataAvailableForAutosave’) due to
>> massive differences in event volume or conversely, expensive operations
>> that block the execution thread in question.
>>
>> *Proposed Solution* Add metering options to the addEventListener
>> *Options* configuration
>> object. These options control how the JS execution environment controls the
>> throttling/firing of event handler instances in response to events that
>> match the topic string of the subscription created by addEventListener.
>>
>> Proposed options:
>>
>>     - maxInstances [Number / Function] used to decide how many event
>>     listeners can be invoked before throttling occurs. Throttling does not
>> lose
>>     events but simply queues them.
>>     - throttlingQueueLength [Number] used to maintain an in-memory queue of
>>     un-processed events per Topic string, after throttling kicks in.
>>     - throttlingQueuePolicy [String] Values could be exception - throws an
>>     exception when the queue length is exceeded, rolling - drops the oldest
>>     events and pushes newer ones into the queue, expand- allow the queue to
>>     expand to cover all events.
>>
>> *Additional Options* It might be even more useful if the options allowed
>> targeting or creation of Web Workers (or Node child processes, depending on
>> the execution context) based on the event handler configuration. This could
>> potentially target CPU cores and/or O/S child processes / threads
>> (depending on the O/S terminology for parallel execution).
>>
>> *Related Thread* The proposal identified in the link below (by Šime
>> Vidas) could
>> be part of this solution as it defines other metering options around
>> debounce (which improves scale around event handling, which is in the same
>> problem space) and handling throttling through frequency, which should be
>> one of the alternatives in addition to my proposal above (as I believe they
>> are orthogonal): https://discourse.wicg.io/t/add-event-throttlin
>> g-and-debouncing-to-addeventlisteneroptions/2436/19
>>
>> Sai Prakash
>> @SylonZero
>>
Received on Friday, 1 December 2017 09:36:40 UTC

This archive was generated by hypermail 2.3.1 : Friday, 1 December 2017 09:36:40 UTC