On Tue, Apr 18, 2023 at 7:34 PM Fergal Daly <fer...@google.com> wrote:
> [+iclelland]
>
> On Tue, 18 Apr 2023 at 23:19, Nic Jansma <nicjan...@gmail.com> wrote:
>
>> Hi Kenji,
>>
>> Will there be a way to detect whether or not unload will fire from JS?
>>
>
> I don't believe there is a way to query the policies that currently apply
> to a document.
>
> Ian, is there? Should there be?
>
You can query permissions policies currently, with
document.featurePolicy.allowedFeatures (although I'm working on replacing
that with a permissions-API-based interface, it will certainly work right
now). So if you're using permissions policy to control unload handlers,
then that will tell you whether it is enabled in the current frame.
If you're not using permissions policy, and just want to know whether
*this* proposal will cause unload handlers to be skipped, that's not going
to work. You'd need to do something like what you indicated below. (And, of
course, accept that unload handlers are fundamentally reliable, even if
Chrome *is* intending to execute them)
Ian
>
> You could attempt to add an unload handler then use
>
> getEventListeners(window).hasOwnProperty("unload")
>
> to see if it worked and then remove it.
>
> That would not detect when the current origin is disallowed but some other
> origin used in a subframe is allowed, however I don't think that case is
> relevant to you since you don't have any access to that origin,
>
>
>> We (Akamai) have some security products that may hook into unload, *only*
>> if they detect that *another* script has hooked into unload. When this
>> happens, we also shift some "cleanup" work to that unload handler that we
>> need to execute. Otherwise the cleanup work executes in
>> beforeunload/pagehide/etc.
>>
>> If we have knowledge that unload *won't* execute, we would skip hooking
>> unload and keep our cleanup behavior earlier. (we don't hook unload for
>> any other reason).
>>
>> For browsers that don't reliably execute unload today (e.g. Safari), we a
>> combination of UA sniffing and feature detection. But that could be
>> tricky if Chrome transitions from reliably firing it to not over time.
>>
>
> In the Safari case, what do you do? If an unload handler was installed,
> you cannot tell whether it will fire since it depends on whether the page
> is BFCacheable. Do you add it just in case but also do stuff in pagehide?
>
> F
>
>
>>
>> On Friday, April 7, 2023 at 5:03:45 AM UTC-4 Kenji Baheux wrote:
>>
>>> In light of this feedback, we are landing a metric to understand how
>>> often subframe unload handlers run without main frame navigation.
>>> We'll use the data to adjust our plans accordingly.
>>>
>>> Since it takes time to get & analyze the data, *we are pushing the
>>> proposed behavior change to M115*.
>>>
>>> *Adjustments based on feedback received thus far*
>>> We'll offer a "deprecation flag" (a flag to temporarily keep unload
>>> working as it currently does) which should be more convenient than the
>>> Extension suggestion (e.g. unmanaged devices where group policy can't be
>>> deployed).
>>>
>>> We are also removing the following aspect "This policy will also be
>>> enabled by default if Chrome detects that it is in an enterprise /
>>> education environment, as hinted by the presence of one or more existing
>>> group policies." from the proposal since we got feedback that it's better
>>> to keep group policy free of any automated logic.
>>>
>>> I've reached out to a few more communities and have only heard
>>> supportive comments so far (touching wood).
>>> Please continue to help us get the word out so that we can further
>>> minimize the risk of oversights.
>>> Thanks!
>>>
>>> On Thursday, March 30, 2023 at 3:23:16 PM UTC+9 Fergal Daly wrote:
>>>
>>>> [+sm...@mozilla.com]
>>>>
>>>> I'm relaying a piece of feedback from Mozilla in this github issue
>>>> <https://github.com/mozilla/standards-positions/issues/691#issuecomment-1484997320>
>>>> .
>>>>
>>>> It's possible that pages are depending on `unload` handlers in
>>>> subframes for functionality even without any main frame navigation. E.g a
>>>> page creates a subframe with an unload handler, when the subframe is
>>>> destroyed or navigates to somewhere else, that unload handler does
>>>> something interesting, e.g. notifies the outer frame that this has
>>>> happened.
>>>>
>>>> This is definitely possible. It's also pretty easy to switch to
>>>> pagehide for this case but we should try to understand how common this is
>>>> before breaking it. It should be possible to measure how often subframe
>>>> unloads fire when the mainframe is not navigating. This will give us an
>>>> upper bound on the size of the problem,
>>>>
>>>> F
>>>>
>>>>
>>>> On Fri, 24 Mar 2023 at 10:16, Kenji Baheux <kenji...@chromium.org>
>>>> wrote:
>>>>
>>>>> Tl;dr: the presence of unload event listeners is a primary blocker
>>>>> for back/forward cache on Chromium based browsers and for Firefox on
>>>>> desktop platforms. On the other hand, for mobile platforms, almost all
>>>>> browsers prioritize the bfcache by not firing unload events in most
>>>>> cases. To improve the situation, we’ve been working with lots of partners
>>>>> and successfully reduced the use of unload event listeners over the
>>>>> last few years. To further accelerate this migration, we propose to have
>>>>> Chrome for desktop gradually skip unload events. If this call for
>>>>> feedback doesn’t unearth critical showstoppers and if the proposal makes
>>>>> it
>>>>> through the blink process, the behavior change could be starting from M114
>>>>> at the earliest (note: beforeunload will remain unchanged). We’d like
>>>>> feedback on this plan, in particular use cases that don’t yet have a
>>>>> viable
>>>>> alternative.
>>>>>
>>>>>
>>>>>
>>>>> Hi,
>>>>>
>>>>>
>>>>> This is a call for feedback about a tentative plan regarding unload
>>>>> events. Our goal is to identify use cases for which there isn’t any good
>>>>> alternative to unload events, and would therefore prevent this plan
>>>>> from moving forward.
>>>>>
>>>>>
>>>>> The unload event is extremely unreliable. It is ignored in most cases
>>>>> by all mobile browsers with the exception of Firefox on Android.
>>>>> Furthermore, in Safari, the unload event is ignored on both desktop &
>>>>> mobile platforms. In addition to being unreliable, the presence of
>>>>> unload event listeners on a page is a major back/forward cache
>>>>> blocker on desktop for Chromium browsers and Firefox. Based on Chrome
>>>>> stats, we believe that unload event listeners reduce bfcache’s
>>>>> ability to deliver instant back/forward navigation by ~18 percentage
>>>>> points
>>>>> (hit-rate).
>>>>>
>>>>>
>>>>> Over the course of 2021~2022, we ran a large collaborative effort to
>>>>> reduce the usage of unload event listeners, in particular across
>>>>> popular third parties. We’ve seen great progress with many sites and third
>>>>> parties having already completed their migration.
>>>>>
>>>>>
>>>>> Given how unreliable unload events are, the potential user experience
>>>>> upsides, and the great progress achieved by the ecosystem on switching
>>>>> away
>>>>> from unload, we’d like to help accelerate the migration by gradually
>>>>> skipping unload events on Chrome for desktop.
>>>>>
>>>>>
>>>>> 👉 Please note that beforeunload will remain unchanged as this event
>>>>> doesn’t have reliability issues and doesn’t block BFCache. 👈
>>>>>
>>>>>
>>>>> We are interested in hearing your feedback about this plan. In
>>>>> particular, please let us know if you are aware of unload event
>>>>> listener use cases that lack a viable alternative. Your feedback will
>>>>> inform the proposal (e.g. behavior and timeline).
>>>>>
>>>>>
>>>>> If this call for feedback doesn’t unearth any critical showstoppers,
>>>>> and if the proposal makes it through the blink process, we’d start the
>>>>> plan
>>>>> from M114 at the earliest by having a small likelihood of ignoring unload
>>>>> events while providing access to fine-tuning control (e.g.
>>>>> Permission-Policy:
>>>>> unload API <https://chromestatus.com/feature/5760325231050752>) and
>>>>> Enterprise/Edu carve outs. From there, we’ll continue to monitor the
>>>>> community’s feedback and gradually increase the likelihood over time. We
>>>>> are hoping to make significant progress by the end of this year, and hope
>>>>> to reach a satisfying state sometime in 2024.
>>>>>
>>>>>
>>>>> See the sections below for more context, our guidance for a
>>>>> post-unload web, an API to exert control over unload event listeners,
>>>>> and our approach to ease-in enterprise/edu products into this change.
>>>>>
>>>>>
>>>>>
>>>>> Background about bfcache
>>>>>
>>>>> Back/forward cache <https://web.dev/bfcache/> is a browser
>>>>> optimization that enables instant back and forward navigation. It’s an
>>>>> in-memory cache that stores a complete snapshot of a page (including the
>>>>> JavaScript heap) as the user is navigating away. With the entire page in
>>>>> memory, the browser can quickly and easily restore it
>>>>> <https://www.youtube.com/watch?v=cuPsdRckkF0> if the user decides to
>>>>> return.
>>>>>
>>>>>
>>>>> The multiple behaviors of bfcache with unload events
>>>>>
>>>>> Unfortunately, not all pages can be stored in bfcache. For instance,
>>>>> using certain APIs prevent pages from entering the bfcache. In particular,
>>>>> the presence of unload listeners on a page is the most common bfcache
>>>>> blocker.
>>>>>
>>>>>
>>>>> The use of unload listeners is highly discouraged because it’s a
>>>>> fundamentally unreliable event:
>>>>>
>>>>> -
>>>>>
>>>>> On desktop, Chrome and Firefox are currently firing unload events
>>>>> at the cost of the user experience, while Safari will attempt to cache
>>>>> some
>>>>> pages with an unload event listener (skipping the event in doing
>>>>> so).
>>>>> -
>>>>>
>>>>> On mobile, Chrome and Safari will attempt to cache pages with an
>>>>> unload event listener. On the other hand, Firefox treats pages
>>>>> that use unload event listeners as ineligible for the bfcache,
>>>>> except on iOS, which requires all browsers to use the WebKit rendering
>>>>> engine (i.e. all browsers inherently behave like Safari on this
>>>>> platform).
>>>>>
>>>>>
>>>>> Alternatives to unload event listener
>>>>>
>>>>> The recommended alternatives to unload event listeners are to:
>>>>>
>>>>> -
>>>>>
>>>>> Use the pagehide event listener
>>>>>
>>>>> <https://web.dev/bfcache/#only-add-beforeunload-listeners-conditionally:~:text=Instead%20of%20using%20the%20unload%20event%2C%20use%20the%20pagehide%20event.%20The%20pagehide%20event%20fires%20in%20all%20cases%20where%20the%20unload%20event%20currently%20fires%2C%20and%20it%20also%20fires%20when%20a%20page%20is%20put%20in%20the%20bfcache.>
>>>>> (note: despite the name, this serves a different purpose than the page
>>>>> visibility API).
>>>>> -
>>>>>
>>>>> For the cases where user interaction would be useful, conditionally
>>>>> use the beforeunload event listener
>>>>>
>>>>> <https://web.dev/bfcache/#only-add-beforeunload-listeners-conditionally>
>>>>> .
>>>>> -
>>>>>
>>>>> Use sendBeacon
>>>>> <https://developer.mozilla.org/en-US/docs/Web/API/Navigator/sendBeacon>
>>>>> or fetch keepalive
>>>>>
>>>>> <https://developer.mozilla.org/en-US/docs/Web/API/fetch#:~:text=keepalive,Navigator.sendBeacon()%20API.>
>>>>> to send analytics data.
>>>>>
>>>>>
>>>>> In addition, you may be interested in the origin trial
>>>>> <https://developer.chrome.com/origintrials/#/view_trial/1581889369113886721>
>>>>> for the Pending Beacon API
>>>>> <https://chromestatus.com/feature/5690553554436096>. This
>>>>> bfcache-friendly API allows sending a bundle of data to a backend server,
>>>>> ideally at the ‘end’ of a user’s visit to a page. From our observations,
>>>>> we
>>>>> believe this is the most common use case for unload event listeners.
>>>>> Compared to the methods highlighted above, this API has better ergonomics.
>>>>>
>>>>>
>>>>> Test driving a web free of unload event listeners!
>>>>>
>>>>> To understand how the plan might play out, please consider joining the
>>>>> origin
>>>>> trial
>>>>> <https://developer.chrome.com/origintrials/#/view_trial/1012184016251518977>
>>>>> for the Permissions-Policy: unload API
>>>>> <https://chromestatus.com/feature/5760325231050752>. This API allows
>>>>> any site to:
>>>>>
>>>>> -
>>>>>
>>>>> Exert control over unload event listeners (e.g. completely
>>>>> disallow them, or selectively allow them for specific origins).
>>>>> -
>>>>>
>>>>> Report the use of unload event listeners to an endpoint for
>>>>> assessment purposes.
>>>>>
>>>>>
>>>>> Chrome for Enterprise & Education
>>>>>
>>>>> We also acknowledge that providers of enterprise & education solutions
>>>>> may not always have the flexibility to quickly update existing
>>>>> deployments.
>>>>> To minimize concerns, we’ll offer a group policy to keep the current
>>>>> behavior for unload events. This policy will also be enabled by
>>>>> default if Chrome detects that it is in an enterprise / education
>>>>> environment, as hinted by the presence of one or more existing group
>>>>> policies. For unmanaged Enterprise/Edu environments, a simple Chrome
>>>>> extension could inject the relevant calls to the Permission-Policy:unload
>>>>> API for temporarily opting-out the relevant origin(s).
>>>>>
>>>>
--
You received this message because you are subscribed to the Google Groups
"blink-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to blink-dev+unsubscr...@chromium.org.
To view this discussion on the web visit
https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAK_TSXLi3Fm_hSkPMyEsJWNhOoS3g0v1YuBROFCSLAgYsXx0zw%40mail.gmail.com.
