This patch caused a failure in the testharness.js test suite, and that failure demonstrated an edge case I hadn't considered. After looking in to it further (research summarized below), it looks like the edge case was limited to the infrastructure test itself. I don't think it demonstrates anything unsound about this patch.

It's possible that worker-based tests (which require done to be invoked) have been written to rely on the "done on uncaught exception" behavior which testharness.js currently exhibits. This change would cause those tests to begin reporting as TIMEOUT (as it did with the failing infrastructure test).

It doesn't look like this is a problem, though. There are six worker sources which enable allow_uncaught_exception:

$ git grep -lE 'importScript.*testharness' | xargs grep -l allow_unca
docs/writing-tests/testharness-api.md
html/webappapis/scripting/events/event-handler-processing-algorithm-error/synthetic-errorevent-click.worker.js
html/webappapis/scripting/events/event-handler-processing-algorithm-error/workerglobalscope-runtime-error.worker.js
html/webappapis/scripting/events/event-handler-processing-algorithm-error/workerglobalscope-synthetic-errorevent.worker.js
html/webappapis/scripting/events/event-handler-processing-algorithm-error/workerglobalscope-synthetic-event.worker.js
html/webappapis/scripting/processing-model-2/unhandled-promise-rejections/support/promise-rejection-events.js
resources/test/tests/functional/worker-uncaught-allow.js

With the exception of the testharness.js test mentioned above, all of them invoke done() at the top-level (and the uncaught exception does not interfere with this).

There are three multi-global tests that use allow_uncaught_exception:

$ find . -name '*any.js' | xargs grep -l allow_unca
./IndexedDB/idb-explicit-commit-throw.any.js
./web-locks/held.tentative.https.any.js
./html/webappapis/microtask-queuing/queue-microtask-exceptions.any.js

For these, the done() call is inserted by the server, but again, the uncaught exception does not occur at the top level, so each invocation is always evaluated.

In principle this looks OK, I think. But please push to at least one of the trigger branches for Firefox or Chrome nightly in order to check for unexpected regressions.

So I see a Fx run; the wpt.fyi diff is https://wpt.fyi/results/?diff&filter=ADC&run_id=321040007&run_id=344290004 Are any of those changes worrying?

@jgraham I think we're good

More interestingly, there were 4 tests that had bugs:

• fetch/corb/script-resource-with-nonsniffable-types.tentative.sub.html (fix under review [fetch] Correct tentative tests #19480)
• html/semantics/scripting-1/the-script-element/module/charset-02.html (fix under review [html] Correct test resources #19481)
• pointerevents/pointerevent_setpointercapture_inactive_button_mouse.html (fix under review [pointerevents] Stabilize test #19708)
• IndexedDB/keypath-exceptions.htm (fix under review [IndexedDB] Avoid race condition #19713)

...and 5 tests whose modified results represent a correction (including one that I didn't spot above):

• custom-elements/parser/parser-fallsback-to-unknown-element.html
• html/semantics/scripting-1/the-script-element/module/error-and-slow-dependency.html
• html/webappapis/scripting/window-onerror-parse-error.html
• html/webappapis/scripting/window-onerror-runtime-error-throw.html
• html/webappapis/scripting/window-onerror-runtime-error.html

@jgraham Do you think this needs further validation?