Structured cloning of errors makes invalid assumptions about [[GetPrototypeOf]]

[bzbarsky]

https://html.spec.whatwg.org/multipage/structured-data.html#structuredserializeinternal step 17 (the [[ErrorData]] case) substep 2 does:

Let valueProto be ! value.[GetPrototypeOf].

But [[GetPrototypeOf]] can in fact return an abrupt completion, so this step can't use !. Does it mean to use ? or do something else? Are there tests for the throwing case here?

@yutakahirano @TimothyGu @domenic

[yutakahirano]

value has an [[ErrorData]] internal slot, so it is an ordinary object and [[GetPrototypeOf]] never throws. Should we use OrdinaryGetPrototypeOf?

[bzbarsky]

Ah, I see. That might be clearer, yes.

[domenic]

I prefer not bypassing the [[ methods, so instead I guess we could add a note explaining the assert.

[bzbarsky]

That would be fine too, thank you.

[annevk]

The fix here would be adding something like "value is an ordinary object due to its [[ErrorData]] internal slot." as a note below the text quoted in OP.

[TimothyGu]

I don't think an implementation is necessarily disallowed from creating an exotic object with [[ErrorData]] internal slot. I'd suggest the following fix:

-if value has an [[ErrorData]] internal slot and value is not a platform object, then:
+if value is an ordinary object, has an [[ErrorData]] internal slot, and is not a platform object, then:

[domenic]

This is obsolete now that #5150 is merged. We could open a new issue/PR for the issue @TimothyGu raised, but that would be a separate normative change (of not structured-cloning exotic objects with an [[ErrorData]] internal slot), and a rather theoretical one.