Bug 239134 – Web Inspector: `inspector/debugger/breakpoints/resolved-dump-all-pause-locations.html` is a flakey failure

• Status: RESOLVED
• Resolution: FIXED
• Component: Web Inspector

[Patrick Angle]

To reproduce: `run-webkit-tests --iterations=100 inspector/debugger/breakpoints/resolved-dump-all-pause-locations.html`

This began being flakey after r291746, which added a test case. It appears the we finally hit the edge of our luck with racing scripts being sent to the frontend, and this became flakey about 5% of the time (varies by platform).

https://results.webkit.org/?suite=layout-tests&test=inspector%2Fdebugger%2Fbreakpoints%2Fresolved-dump-all-pause-locations.html

[Radar WebKit Bug Importer]

<rdar://problem/91639437>

[Patrick Angle]

Created attachment 457349 [details]
Patch v1.0

[Joseph Pecoraro]

Comment on attachment 457349 [details]
Patch v1.0

Oo, neat!

[Devin Rousso]

Comment on attachment 457349 [details]
Patch v1.0

View in context: https://bugs.webkit.org/attachment.cgi?id=457349&action=review

r=me

> Source/WebInspectorUI/UserInterface/Protocol/InspectorBackend.js:247
> +    async awaitPendingDispatches()

Rather than have a new method, could we make `runAfterPendingDispatches` just return a `Promise` if a `callback` is not provided (i.e. just like what we do with protocol commands in the frontend)?

> LayoutTests/inspector/debugger/breakpoints/resources/dump.js:10
> +    window.addDumpAllPauseLocationsTestCase = async function(suite, {name, scriptRegex, resourceRegex}) {

Why does this need to be `async`?

> LayoutTests/inspector/debugger/breakpoints/resources/dump.js:27
> +                await InspectorBackend.awaitPendingDispatches();

Another option to waiting for pending dispatches could be to do something like `WI.networkManager.mainFrame.resourceCollection.addEventListener(WI.Collection.Event.ItemAdded, (event) => { ... })`.  I feel like that's probably safer than just waiting for pending dispatches as I'm not sure it's guaranteed that the script message is pending (i.e. it could be coming *right* after the current messages).

> LayoutTests/inspector/debugger/breakpoints/resources/dump.js:51
> +                await InspectorBackend.awaitPendingDispatches();

possibly a ditto of :27 by waiting for `Debugger.breakpointResolved` instead?

[Patrick Angle]

Comment on attachment 457349 [details]
Patch v1.0

View in context: https://bugs.webkit.org/attachment.cgi?id=457349&action=review

>> LayoutTests/inspector/debugger/breakpoints/resources/dump.js:10
>> +    window.addDumpAllPauseLocationsTestCase = async function(suite, {name, scriptRegex, resourceRegex}) {
> 
> Why does this need to be `async`?

Oops. Remnant of solutions past.

>> LayoutTests/inspector/debugger/breakpoints/resources/dump.js:27
>> +                await InspectorBackend.awaitPendingDispatches();
> 
> Another option to waiting for pending dispatches could be to do something like `WI.networkManager.mainFrame.resourceCollection.addEventListener(WI.Collection.Event.ItemAdded, (event) => { ... })`.  I feel like that's probably safer than just waiting for pending dispatches as I'm not sure it's guaranteed that the script message is pending (i.e. it could be coming *right* after the current messages).

Yeah, let me take another look at this... You are probably right.

[Patrick Angle]

Comment on attachment 457349 [details]
Patch v1.0

View in context: https://bugs.webkit.org/attachment.cgi?id=457349&action=review

Not sure why, but the bug is now reproducing after I pulled this patch back to my machine, so I'm not convinced I've actually/fully fixed this.

>>> LayoutTests/inspector/debugger/breakpoints/resources/dump.js:10
>>> +    window.addDumpAllPauseLocationsTestCase = async function(suite, {name, scriptRegex, resourceRegex}) {
>> 
>> Why does this need to be `async`?
> 
> Oops. Remnant of solutions past.

Oops. Remnant of solutions past.

>>> LayoutTests/inspector/debugger/breakpoints/resources/dump.js:27
>>> +                await InspectorBackend.awaitPendingDispatches();
>> 
>> Another option to waiting for pending dispatches could be to do something like `WI.networkManager.mainFrame.resourceCollection.addEventListener(WI.Collection.Event.ItemAdded, (event) => { ... })`.  I feel like that's probably safer than just waiting for pending dispatches as I'm not sure it's guaranteed that the script message is pending (i.e. it could be coming *right* after the current messages).
> 
> Yeah, let me take another look at this... You are probably right.

Yeah, let me take another look at this... You are probably right.

[Patrick Angle]

Pull request: https://github.com/WebKit/WebKit/pull/1066

[EWS]

Committed r294907 (251030@main): <https://commits.webkit.org/251030@main>

Reviewed commits have been landed. Closing PR #1066 and removing active labels.