Properties set on window.customElements can disappear due to GC

[Russell Bicknell]

Occasionally, properties that are set on window.customElements very early in the lifetime of the page are removed by garbage collection. To repro, put the attached files in the same directory and open 'index.html'. You may need to adjust the number of objects created in 'test.html' to balance the repro rate with the time per run. You can also trigger this (at a reduced rate) by putting an external script that spins for 10-20ms between the write / read instead of creating objects. This seems to affect 10, TP, and Nightly.

[Russell Bicknell]

Created attachment 311187 [details]
test.html

[Russell Bicknell]

Created attachment 311188 [details]
index.html

[Russell Bicknell]

Comment on attachment 311187 [details]
test.html

<script>
  // Uncomment this section to prevent the bug.
  /*
  Object.defineProperty(window, 'customElements', {
    value: window.customElements,
  });
  */

  window.customElements.prop = 1234;

  const a = [];
  for (let i = 0; i < 1000000; i++) {
    a.push({});
  }

  const pass = window.customElements.prop === 1234;
  window.parent.postMessage(pass, '*');
</script>

[Russell Bicknell]

Created attachment 311189 [details]
test.html

[Russell Bicknell]

This seems like it's probably related to 171567.

[Radar WebKit Bug Importer]

<rdar://problem/32440668>

[Ryosuke Niwa]

(In reply to Russell Bicknell from comment #5)
> This seems like it's probably related to 171567.

It's not.

[Ryosuke Niwa]

Created attachment 346581 [details]
Fixes the bug

[Ryosuke Niwa]

Committed r234578: <https://trac.webkit.org/changeset/234578>

[Darin Adler]

Comment on attachment 346581 [details]
Fixes the bug

View in context: https://bugs.webkit.org/attachment.cgi?id=346581&action=review

> Source/WebCore/dom/CustomElementRegistry.idl:29
>      EnabledAtRuntime=CustomElements,
> -    ImplementationLacksVTable,
>      JSGenerateToNativeObject,
> +    GenerateIsReachable=ImplScriptExecutionContext

We’ve been sorting these alphabetically and putting commas on every line. So this should be a line higher and have a comma.

[Ryosuke Niwa]

Sure. Fixed it in r234585.

[Truitt Savell]

Looks like the new test fast/custom-elements/custom-element-registry-wrapper-should-stay-alive.html

is a constant timeout on all debug platforms. 

Test History:
https://webkit-test-results.webkit.org/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=fast%2Fcustom-elements%2Fcustom-element-registry-wrapper-should-stay-alive.html

[Ryosuke Niwa]

Yeah, noticed. Will look into it today.

[Ryosuke Niwa]

The test failure is tracked by https://bugs.webkit.org/show_bug.cgi?id=188353.