Intent to Ship: Back/forward cache NotRestoredReasons API

[Yuzu Saijo]

Contact emails

yu...@google.com, yu...@chromium.org, fer...@chromium.org

Explainer

https://github.com/WICG/bfcache-not-restored-reason/blob/main/NotRestoredReason.md

Specification

https://github.com/whatwg/html/pull/9360

Design docs

https://github.com/WICG/bfcache-not-restored-reason/blob/main/NotRestoredReason.md

Summary

NotRestoredReason API will report the list of reasons why a page is not served from BFcache in a frame tree structure, via PerformanceNavigationTiming API.

Blink component

UI>Browser>Navigation>BFCache

TAG review

https://github.com/w3ctag/design-reviews/issues/739

TAG review status

Issues addressed

Risks

Interoperability and Compatibility

Gecko: Defer (https://github.com/mozilla/standards-positions/issues/766) Once issues (standardized reasons & unsalvageable documents), they would switch to positive.

WebKit: No signal (https://github.com/WebKit/standards-positions/issues/154)

Web developers: Positive (https://github.com/w3c/navigation-timing/issues/171#issuecomment-1062672989)

Other signals: Positive from Origin Trial users:

How likely are you to keep using this feature?

92% answered likely, 8% (1 vote) is unsure

Security

We do not report detailed information about cross-origin iframes. See Security and Privacy section in the explainer.

WebView application risks

Does this intent deprecate or change behavior of existing APIs, such that it has potentially high risk for Android WebView-based applications?

No.

Debuggability

In DevTools console, try:

performance.getEntriesByType('navigation')[0].notRestoredReasons;

Will this feature be supported on all six Blink platforms (Windows, Mac, Linux, Chrome OS, Android, and Android WebView)?

Yes.

NotRestoredReasons API is available on all platforms including WebView, but back/forward cache is not enabled on WebView. So on WebView, NotRestoredReasons API should always say that the page is blocked from being restored from bfcache with the reason being something like “not supported”.

(Currently it reports null due to a bug)

Is this feature fully tested by web-platform-tests?

Yes. https://source.chromium.org/chromium/chromium/src/+/main:third_party/blink/web_tests/external/wpt/performance-timeline/not-restored-reasons/

DevTrial instructions

https://github.com/rubberyuzu/bfcache-not-retored-reason/blob/main/HowToTest.md

Flag name

blink RunTimeEnabledFeature: BackForwardCacheSendNotRestoredReasons

Requires code in //chrome?

False

Tracking bug

https://bugs.chromium.org/p/chromium/issues/detail?id=1326344

Launch bug

https://launch.corp.google.com/launch/4200848

Estimated milestones

Shipping on desktop	116
OriginTrial desktop last	114
OriginTrial desktop first	109
DevTrial on desktop	108

Shipping on Android	116
OriginTrial Android last	114
OriginTrial Android first	109
DevTrial on Android	108

Shipping on WebView	116
OriginTrial WebView last	114
OriginTrial WebView first	109
DevTrial on WebView	108

Anticipated spec changes

Open questions about a feature may be a source of future web compat or interop issues. Please list open issues.

https://github.com/whatwg/html/pull/9360

Link to entry on the Chrome Platform Status

https://chromestatus.com/feature/5684908759449600

Links to previous Intent discussions

Intent to prototype: https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAP-nMoGAzjUjzv3WmxcRpUSBgnA-AHQ05kh9gXc%2BQB8pRM6%2BfA%40mail.gmail.com Intent to Experiment: https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAP-nMoHe391sAB2PdbEVw9uiSPFxTB_EYsRizcPpZ7-pg16O0A%40mail.gmail.com

Intent to Extend Experiment: https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAA5e698QcKZSthm%3Dz_4pi8cOzi4kfbx-AXveC%2BAKimUh-tMycA%40mail.gmail.com

This intent message was generated by Chrome Platform Status.

[Yoav Weiss]

On Thu, Jul 6, 2023 at 7:28 AM 'Yuzu Saijo' via blink-dev <blin...@chromium.org> wrote:

Contact emails

yu...@google.com, yu...@chromium.org, fer...@chromium.org

Explainer

https://github.com/WICG/bfcache-not-restored-reason/blob/main/NotRestoredReason.md

Specification

https://github.com/whatwg/html/pull/9360

Design docs

https://github.com/WICG/bfcache-not-restored-reason/blob/main/NotRestoredReason.md

Summary

NotRestoredReason API will report the list of reasons why a page is not served from BFcache in a frame tree structure, via PerformanceNavigationTiming API.

Blink component

UI>Browser>Navigation>BFCache

TAG review

https://github.com/w3ctag/design-reviews/issues/739

TAG review status

Issues addressed

Risks

Interoperability and Compatibility

Gecko: Defer (https://github.com/mozilla/standards-positions/issues/766) Once issues (standardized reasons & unsalvageable documents), they would switch to positive.

It seems like the "standardized reasons" part is addressed in your PR. Is the same true for the second point?

 

--
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+...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAP-nMoHYpT3sxWV%2BEipL5NcNSWy8fOdDdAroucmNb%3DZvxJWRBA%40mail.gmail.com.

[Domenic Denicola]

I have some questions about how well the implementation here matches up with the spec.

First, there doesn't appear to be any NotRestoredReasons interface defined in Chromium? The relevant attribute on PerformanceNavigationTiming returns object?. That seems like a problematic mismatch...

Second, I can't find exactly where the list of script-exposed not restored reasons are. But, I'll note that Chromium seems to have ~50 such reasons, whereas you've only specified 4 (fetch, navigation-failure, parser-aborted, websocket). Can you confirm that you're only shipping the specified four?

Thanks!

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAL5BFfXtkH6O82W%2BWm9ckCyYasSJt2cbs9VA4VZAmYhtivgj4g%40mail.gmail.com.

[Domenic Denicola]

Also, checking the tests, it seems like the currently-implemented reasons don't match the spec. E.g. this test requires the reason to be "WebSocket", but the specification says "websocket" (lowercase). I couldn't find tests for the other three reasons...

[Yuzu Saijo]

Sorry for the delayed response.

> there doesn't appear to be any NotRestoredReasons interface defined in Chromium?

Let me address this implementation and delay the shipping until the chromium implementation matches the proposed spec. Thanks for pointing it out!

Same for WPT. I will add tests for all the standardized reasons.

> Can you confirm that you're only shipping the specified four?

We do have ~50 not restored reasons, and in theory we will be able to remove most of them except for the standardized four reasons. 

However, in reality it will take time for us to support all the reasons and we need to keep blocking on them for a while.

In the meantime, our plan was to expose the non-standardized reasons too, but in a way that's distinguishable from standardized reasons as you suggested here.

I realized that we need to add browser specific reasons to the spec as well. Let me add that and send a review request again.

Thanks!

Yuzu

[Yuzu Saijo]

+bfcache-dev

I was talking to Fergal today and discussed this, and I am not sure about adding browser-specific reasons to the spec.

For example, some reasons like "speech synthesis API is used" / "unload handler" are completely specific to Chrome, and it doesn't really make sense to add them to the spec, even with the namespace (x-speechsysthesis / x-unloadhandler).

Maybe we can document the reasons somewhere in a shared list but not in the spec?

I think the API would be more useful if it can give as much information as possible, not limited to the specced reasons.

Please let me know what you think!

Yuzu

[Domenic Denicola]

I think specifying these reasons is important. As noted in the linked issue, I think the end goal should be:

• Every reason that a browser ever emits, is found in a specification somewhere. (It doesn't have to be the HTML spec, e.g. the speech synthesis reason could live in the speech synthesis spec.)
• If browsers prevent bfcache restoration for a reason not found in a spec, it is always translated to a standardized reason such as "unknown".

This avoids the usual interop problems with vendor-specific extensions to the web platform, such as: no clear specification for what strings to use; no clear point at which the reason is added to the document's reasons list; etc. Although you claim these reasons are idiosyncratic to Chrome, that won't necessarily be the case; e.g. Firefox has unload handler as a reason, and I suspect most user agents have memory limitations or similar.

We could have a discussion about allowing vendor-specific information in the API in addition to the standardized reasons. For example, we could have one of the standardized reasons be "user-agent-specific", and then add an additional field userAgentSpecificInfo. But I would like to see significantly more discussion with other vendors before going that route.

[Fergal Daly]

On Wed, 9 Aug 2023 at 12:01, Domenic Denicola <dom...@chromium.org> wrote:

I think specifying these reasons is important. As noted in the linked issue, I think the end goal should be:

• Every reason that a browser ever emits, is found in a specification somewhere. (It doesn't have to be the HTML spec, e.g. the speech synthesis reason could live in the speech synthesis spec.)

There's no intrinsic reason for speech synthesis to block BFCache. It just happens that Chrome blocked it. There's no spec reason for unload to block BFCache, in fact the spec says that it doesn't.

I think it's good for us to have agreed names, e.g. "unload-event-handler". Should we put into various specs "if an implementer chooses to block BFCache because X has been used, they should use the reason `Y`"?

• If browsers prevent bfcache restoration for a reason not found in a spec, it is always translated to a standardized reason such as "unknown".

This avoids the usual interop problems with vendor-specific extensions to the web platform, such as: no clear specification for what strings to use; no clear point at which the reason is added to the document's reasons list; etc. Although you claim these reasons are idiosyncratic to Chrome, that won't necessarily be the case; e.g. Firefox has unload handler as a reason, and I suspect most user agents have memory limitations or similar.

Chrome has over 100 reasons. I'd say at least 50 of them are actionable such that you wouldn't want to lump them into an opaque "unknown" category.

I do not relish the idea of updating 50 places in spec to insert a name to be used if you decide to block.

How about maintaining a central list of reasons with low friction to add new reasons even if they are browser-specific? The cases where you must block should still be inline in spec (and also on the list),

F

 

You received this message because you are subscribed to the Google Groups "bfcache-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bfcache-dev...@chromium.org.
To view this discussion on the web, visit https://groups.google.com/a/chromium.org/d/msgid/bfcache-dev/CAM0wra-P3NxELP28%3Dgh%3D3ROC35m8ijS_5RRcStyjFew1AXNyEg%40mail.gmail.com.

[Domenic Denicola]

On Wed, Aug 9, 2023 at 6:44 PM Fergal Daly <fer...@google.com> wrote:

On Wed, 9 Aug 2023 at 12:01, Domenic Denicola <dom...@chromium.org> wrote:

I think specifying these reasons is important. As noted in the linked issue, I think the end goal should be:

• Every reason that a browser ever emits, is found in a specification somewhere. (It doesn't have to be the HTML spec, e.g. the speech synthesis reason could live in the speech synthesis spec.)

There's no intrinsic reason for speech synthesis to block BFCache. It just happens that Chrome blocked it. There's no spec reason for unload to block BFCache, in fact the spec says that it doesn't.

I think it's good for us to have agreed names, e.g. "unload-event-handler". Should we put into various specs "if an implementer chooses to block BFCache because X has been used, they should use the reason `Y`"?

• If browsers prevent bfcache restoration for a reason not found in a spec, it is always translated to a standardized reason such as "unknown".

This avoids the usual interop problems with vendor-specific extensions to the web platform, such as: no clear specification for what strings to use; no clear point at which the reason is added to the document's reasons list; etc. Although you claim these reasons are idiosyncratic to Chrome, that won't necessarily be the case; e.g. Firefox has unload handler as a reason, and I suspect most user agents have memory limitations or similar.

Chrome has over 100 reasons. I'd say at least 50 of them are actionable such that you wouldn't want to lump them into an opaque "unknown" category.

I do not relish the idea of updating 50 places in spec to insert a name to be used if you decide to block.

How about maintaining a central list of reasons with low friction to add new reasons even if they are browser-specific? The cases where you must block should still be inline in spec (and also on the list),

That sounds great to me. We should probably make this separation clear in the spec, e.g. the "must" list will have cross-references you can follow, whereas the "may" list ends up only being cross-referenced from some generic location like https://html.spec.whatwg.org/multipage/browsing-the-web.html#note-bfcache:~:text=User%20agents%20may,keeping%20it%20cached. .

[Yuzu Saijo]

Sounds good, I will create a list on the explainer for the "may block" reasons then.

Re: exposing NotRestoredReasons interface instead of object in idl:

I'm working on the implementation in this CL.

This might be a basic question, but is there any difference on how to call the API from users' perspective, when the exposed attribute is an interface vs object?

[Chris Harrelson]

Please also make sure to complete all of the other shipping gate reviews.

Thanks!

Chris

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/43e32f0e-454e-4525-b317-cbe492e2f23bn%40chromium.org.

[Yuzu Saijo]

This is now ready to ship, now that we have all the approvals on the ChromeStatus and the spec draft is close to agreement.

Can you please take a look at this again?

Thanks!

[Domenic Denicola]

I am happy with the spec progress here and don't think it's a significant blocker for the Intent at this point.

On the tests and implementation:

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.
• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?
• I found a nonstandard toJSON() in NotRestoredReasonDetails in Chromium. Was the intent to specify that?
• Can you confirm that Chromium does not plan to ship any nonstandard not restored reason strings, beyond the specified "fetch", "navigation-failure", "parser-aborted", "websocket", "lock", and "masked"?

[Fergal Daly]

On Tue, 6 Feb 2024 at 15:13, Domenic Denicola <dom...@chromium.org> wrote:

I am happy with the spec progress here and don't think it's a significant blocker for the Intent at this point.

On the tests and implementation:

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.
• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?
• I found a nonstandard toJSON() in NotRestoredReasonDetails in Chromium. Was the intent to specify that?
• Can you confirm that Chromium does not plan to ship any nonstandard not restored reason strings, beyond the specified "fetch", "navigation-failure", "parser-aborted", "websocket", "lock", and "masked"?

I don't know specifically what is there right now but I would expect that we will ship others. E.g. BroadcastChannel blocks BFCache on Chrome and Mozilla but not WebKit and there is currently disagreement. Why would it be better to show "masked" for that case?

F

[Domenic Denicola]

On Tue, Feb 6, 2024 at 6:40 PM Fergal Daly <fer...@google.com> wrote:

On Tue, 6 Feb 2024 at 15:13, Domenic Denicola <dom...@chromium.org> wrote:

I am happy with the spec progress here and don't think it's a significant blocker for the Intent at this point.

On the tests and implementation:

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.
• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?
• I found a nonstandard toJSON() in NotRestoredReasonDetails in Chromium. Was the intent to specify that?
• Can you confirm that Chromium does not plan to ship any nonstandard not restored reason strings, beyond the specified "fetch", "navigation-failure", "parser-aborted", "websocket", "lock", and "masked"?

I don't know specifically what is there right now but I would expect that we will ship others. E.g. BroadcastChannel blocks BFCache on Chrome and Mozilla but not WebKit and there is currently disagreement. Why would it be better to show "masked" for that case?

The idea is to follow the standards and not ship nonstandard behavior. The current spec PR actually only allows sending "masked" in the cross-origin case, and doesn't allow sending it for BroadcastChannel. If the intention is to send some value in the BroadcastChannel case (which is this part of the spec) then that needs to be specified in the spec PR before shipping such a value in Chromium.

[Fergal Daly]

On Wed, 7 Feb 2024 at 12:26, Domenic Denicola <dom...@chromium.org> wrote:

On Tue, Feb 6, 2024 at 6:40 PM Fergal Daly <fer...@google.com> wrote:

On Tue, 6 Feb 2024 at 15:13, Domenic Denicola <dom...@chromium.org> wrote:

I am happy with the spec progress here and don't think it's a significant blocker for the Intent at this point.

On the tests and implementation:

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.
• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?
• I found a nonstandard toJSON() in NotRestoredReasonDetails in Chromium. Was the intent to specify that?
• Can you confirm that Chromium does not plan to ship any nonstandard not restored reason strings, beyond the specified "fetch", "navigation-failure", "parser-aborted", "websocket", "lock", and "masked"?

I don't know specifically what is there right now but I would expect that we will ship others. E.g. BroadcastChannel blocks BFCache on Chrome and Mozilla but not WebKit and there is currently disagreement. Why would it be better to show "masked" for that case?

The idea is to follow the standards and not ship nonstandard behavior. The current spec PR actually only allows sending "masked" in the cross-origin case, and doesn't allow sending it for BroadcastChannel. If the intention is to send some value in the BroadcastChannel case (which is this part of the spec) then that needs to be specified in the spec PR before shipping such a value in Chromium.

BFCaching is never required by spec. That means any browser can block BFCache at any time, for any reason and still be in spec.

I think what's missing is that said we would maintain a registry of reasons that were not in the spec so that when we block for unspecced reasons, we don't proliferate a bunch of undocumented names.

I'm not sure how to express that in the spec,

F

[Domenic Denicola]

On Wed, Feb 7, 2024 at 12:51 PM Fergal Daly <fer...@google.com> wrote:

On Wed, 7 Feb 2024 at 12:26, Domenic Denicola <dom...@chromium.org> wrote:

On Tue, Feb 6, 2024 at 6:40 PM Fergal Daly <fer...@google.com> wrote:

On Tue, 6 Feb 2024 at 15:13, Domenic Denicola <dom...@chromium.org> wrote:

I am happy with the spec progress here and don't think it's a significant blocker for the Intent at this point.

On the tests and implementation:

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.
• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?
• I found a nonstandard toJSON() in NotRestoredReasonDetails in Chromium. Was the intent to specify that?
• Can you confirm that Chromium does not plan to ship any nonstandard not restored reason strings, beyond the specified "fetch", "navigation-failure", "parser-aborted", "websocket", "lock", and "masked"?

I don't know specifically what is there right now but I would expect that we will ship others. E.g. BroadcastChannel blocks BFCache on Chrome and Mozilla but not WebKit and there is currently disagreement. Why would it be better to show "masked" for that case?

The idea is to follow the standards and not ship nonstandard behavior. The current spec PR actually only allows sending "masked" in the cross-origin case, and doesn't allow sending it for BroadcastChannel. If the intention is to send some value in the BroadcastChannel case (which is this part of the spec) then that needs to be specified in the spec PR before shipping such a value in Chromium.

BFCaching is never required by spec. That means any browser can block BFCache at any time, for any reason and still be in spec.

Yes. But a browser cannot create values for the NotRestoredReasonDetails's reason property which are not in the spec, while staying spec-compliant. This is similar to how we cannot have, e.g., DOMException's name property returning arbitrary values; we instead document them all in the spec, and then document that some of them may be thrown in implementation-specific circumstances (example).

 

I think what's missing is that said we would maintain a registry of reasons that were not in the spec so that when we block for unspecced reasons, we don't proliferate a bunch of undocumented names.

I'm not sure how to express that in the spec,

We discussed how to do so upthread: https://groups.google.com/a/chromium.org/g/bfcache-dev/c/ufQx6r6su6U/m/vyQM9nGHAwAJ 

[Yuzu Saijo]

Thanks Domenic for bringing up the concerns!

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.

> Updated all the strings to match the spec-defined strings.

• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?

> Now the failing tests and the expected files are down to three.

1) parser-aborted

We currently block with different reason("loading"), as we haven't worked on differentiating loading vs parser getting aborted.

2) navigation-failure

We do report "navigation-failure" when the document errors(implementation), but somehow the test only reports "http-status-not-ok" which is the chrome internal reason.

I will look into this more.

3) weblock

Chrome currently blocks with another reason here (masked), so this failure will not go away. Maybe I should make WPTs to test if the expected reason exists in the list, instead of checking the complete list.

• I found a nonstandard toJSON() in NotRestoredReasonDetails in Chromium. Was the intent to specify that?

> Added this to the spec, thanks!

• Can you confirm that Chromium does not plan to ship any nonstandard not restored reason strings, beyond the specified "fetch", "navigation-failure", "parser-aborted", "websocket", "lock", and "masked"?

> We plan to add user-agent specific reasons to the spec in the may-block section.

This is the draft PR (have't added the explanation for each reason yet).

Is it okay to ship while we work on the follow-up PR?

[Domenic Denicola]

On Thu, Feb 22, 2024 at 4:20 PM Yuzu Saijo <yu...@google.com> wrote:

Thanks Domenic for bringing up the concerns!

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.

> Updated all the strings to match the spec-defined strings.

• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?

> Now the failing tests and the expected files are down to three.

1) parser-aborted

We currently block with different reason("loading"), as we haven't worked on differentiating loading vs parser getting aborted.

Note that "loading" is a nonstandard reason, so it would be bad to ship in that state. It should either be the correct reason ("parser-aborted") or the generi "masked" reason.

 

2) navigation-failure

We do report "navigation-failure" when the document errors(implementation), but somehow the test only reports "http-status-not-ok" which is the chrome internal reason.

Similar to the above.

 

I will look into this more.

3) weblock

Chrome currently blocks with another reason here (masked), so this failure will not go away. Maybe I should make WPTs to test if the expected reason exists in the list, instead of checking the complete list.

Allowing an implementation to always do "masked" and pass the tests seems reasonable to me.

 

• I found a nonstandard toJSON() in NotRestoredReasonDetails in Chromium. Was the intent to specify that?

> Added this to the spec, thanks!

• Can you confirm that Chromium does not plan to ship any nonstandard not restored reason strings, beyond the specified "fetch", "navigation-failure", "parser-aborted", "websocket", "lock", and "masked"?

> We plan to add user-agent specific reasons to the spec in the may-block section.

This is the draft PR (have't added the explanation for each reason yet).

Is it okay to ship while we work on the follow-up PR?

You could ship the portion that is fully specified, but the portions in the draft PR would not be approved for shipment until they reach the usual bars (e.g., a fully reviewed spec, web platform tests, other-vendor positions, etc.).

[Fergal Daly]

On Thu, 22 Feb 2024 at 16:32, Domenic Denicola <dom...@chromium.org> wrote:

On Thu, Feb 22, 2024 at 4:20 PM Yuzu Saijo <yu...@google.com> wrote:

Thanks Domenic for bringing up the concerns!

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.

> Updated all the strings to match the spec-defined strings.

• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?

> Now the failing tests and the expected files are down to three.

1) parser-aborted

We currently block with different reason("loading"), as we haven't worked on differentiating loading vs parser getting aborted.

Note that "loading" is a nonstandard reason, so it would be bad to ship in that state. It should either be the correct reason ("parser-aborted") or the generi "masked" reason.

"parser-aborted" is a reason that Chrome doesn't currently emit (it doesn't exist in the code). I'm not sure how we ended up speccing a reason that doesn't exist but I don't think we punt the entire NRR feature for another milestone for that.

 

 

2) navigation-failure

We do report "navigation-failure" when the document errors(implementation), but somehow the test only reports "http-status-not-ok" which is the chrome internal reason.

Similar to the above.

I think this one we can just make change http-status-no-ok to navigation-failure.

 

 

I will look into this more.

3) weblock

Chrome currently blocks with another reason here (masked), so this failure will not go away. Maybe I should make WPTs to test if the expected reason exists in the list, instead of checking the complete list.

Allowing an implementation to always do "masked" and pass the tests seems reasonable to me.

I think this is a general issue with the testing. It should always be OK for a UA to have extra reasons, e.g. when we do add parser-aborted, loading will continue to show up. I guess we could add a hack to suppress loading if parser-aborted is present but really what we care about in these tests is that the specced reason is present in the specced case,

F

[Domenic Denicola]

On Thu, Feb 22, 2024 at 6:01 PM Fergal Daly <fer...@google.com> wrote:

On Thu, 22 Feb 2024 at 16:32, Domenic Denicola <dom...@chromium.org> wrote:

On Thu, Feb 22, 2024 at 4:20 PM Yuzu Saijo <yu...@google.com> wrote:

Thanks Domenic for bringing up the concerns!

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.

> Updated all the strings to match the spec-defined strings.

• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?

> Now the failing tests and the expected files are down to three.

1) parser-aborted

We currently block with different reason("loading"), as we haven't worked on differentiating loading vs parser getting aborted.

Note that "loading" is a nonstandard reason, so it would be bad to ship in that state. It should either be the correct reason ("parser-aborted") or the generi "masked" reason.

"parser-aborted" is a reason that Chrome doesn't currently emit (it doesn't exist in the code). I'm not sure how we ended up speccing a reason that doesn't exist but I don't think we punt the entire NRR feature for another milestone for that.

 

 

2) navigation-failure

We do report "navigation-failure" when the document errors(implementation), but somehow the test only reports "http-status-not-ok" which is the chrome internal reason.

Similar to the above.

I think this one we can just make change http-status-no-ok to navigation-failure.

 

 

I will look into this more.

3) weblock

Chrome currently blocks with another reason here (masked), so this failure will not go away. Maybe I should make WPTs to test if the expected reason exists in the list, instead of checking the complete list.

Allowing an implementation to always do "masked" and pass the tests seems reasonable to me.

I think this is a general issue with the testing. It should always be OK for a UA to have extra reasons, e.g. when we do add parser-aborted, loading will continue to show up. I guess we could add a hack to suppress loading if parser-aborted is present but really what we care about in these tests is that the specced reason is present in the specced case,

As we've discussed elsewhere on this thread, extra reasons are not OK; the APIs we ship need to be specified. "masked" is always OK though, per the spec.

[Fergal Daly]

On Thu, 22 Feb 2024 at 18:03, Domenic Denicola <dom...@chromium.org> wrote:

On Thu, Feb 22, 2024 at 6:01 PM Fergal Daly <fer...@google.com> wrote:

On Thu, 22 Feb 2024 at 16:32, Domenic Denicola <dom...@chromium.org> wrote:

On Thu, Feb 22, 2024 at 4:20 PM Yuzu Saijo <yu...@google.com> wrote:

Thanks Domenic for bringing up the concerns!

• I found performance-navigation-timing-navigation-failure.tentative.window.js which seems like it needs to be updated from "error-document" to "navigation-failure". That's worth looking into in case it means the implementation is also not yet updated.

> Updated all the strings to match the spec-defined strings.

• I also found that the Chromium test directory is full of -expected.txt files, which seem to match up with the failures on wpt.fyi. Will those be addressed before shipping?

> Now the failing tests and the expected files are down to three.

1) parser-aborted

We currently block with different reason("loading"), as we haven't worked on differentiating loading vs parser getting aborted.

Note that "loading" is a nonstandard reason, so it would be bad to ship in that state. It should either be the correct reason ("parser-aborted") or the generi "masked" reason.

"parser-aborted" is a reason that Chrome doesn't currently emit (it doesn't exist in the code). I'm not sure how we ended up speccing a reason that doesn't exist but I don't think we punt the entire NRR feature for another milestone for that.

 

 

2) navigation-failure

We do report "navigation-failure" when the document errors(implementation), but somehow the test only reports "http-status-not-ok" which is the chrome internal reason.

Similar to the above.

I think this one we can just make change http-status-no-ok to navigation-failure.

 

 

I will look into this more.

3) weblock

Chrome currently blocks with another reason here (masked), so this failure will not go away. Maybe I should make WPTs to test if the expected reason exists in the list, instead of checking the complete list.

Allowing an implementation to always do "masked" and pass the tests seems reasonable to me.

I think this is a general issue with the testing. It should always be OK for a UA to have extra reasons, e.g. when we do add parser-aborted, loading will continue to show up. I guess we could add a hack to suppress loading if parser-aborted is present but really what we care about in these tests is that the specced reason is present in the specced case,

As we've discussed elsewhere on this thread, extra reasons are not OK; the APIs we ship need to be specified. "masked" is always OK though, per the spec.

Any restore can be blocked for multiple reasons. That can be for a mix of common and UA-specific reasons (all of which appear in the spec). As long as the common reason appears in the test for the common reason, the test should pass. Otherwise are writing tests that assert that UA-specific reasons are absent in some cases. I don't think we should be doing that. UAs can block on UA-specific reasons whenever they want and be within spec,

F 

[Yuzu Saijo]

Now that M123 implementation is aligned with the spec, can we ship the feature to M123?

We're currently working on adding browser-specific blocking reasons (i.e. reasons that may block) to the spec in this PR, and while this is still under review and the details of the names can change, the overall direction will stay the same. Firefox folks are also engaged with speccing the names on the PR.

Please let me know what you think, thanks!

Yuzu

[Daniel Bratell]

LGTM1

/Daniel

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/2e913c96-1ebc-415c-9def-81695f5e74d4n%40chromium.org.

[Mike Taylor]

LGTM2

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/67ea5933-8efb-43d8-abe1-6638415411da%40gmail.com.