The instanceof check will work in most cases while it would not detect AggregateErrors created in a different context/realm (e.g., by creating errors in a context created by vm.createContext()). I believe it is fine to just always check for this property, in case it is not already an enumerable property.

This check would currently miss the case, if either side is an AggregateError and the other is not. That could be addressed in a similar way as I suggested above about the enumerable property.

This is definitely the right check as long the property is non-enumerable. If it is enumerable, we'd do the check twice and this could be costly and a lot of errors are created in the wild with a cause property set as an enumerable one.

In addition, we should also verify that the properties have the same enumerability. We don't yet do that for name and message, but that would be beneficial.

const cause1Enumerable = ObjectPrototypePropertyIsEnumerable(val1, 'cause');
if (cause1Enumerable !== ObjectPrototypePropertyIsEnumerable(val2, 'cause') ||
    !cause1Enumerable &&
    !innerDeepEqual(val1.cause, val2.cause, strict, memos)) {
    return false;
}

Nit (just a general suggestion without need to follow-up upon): the variable names could be changed in a way that it's clearer what they stand for. E.g., this one could be:

That way it would be easy to understand in the comparison that the entries are meant to be identical.

As far as I can tell, this means now we only compare the name and message fields when they are enumerable, which specificaly opposite from the documentation, and is a breaking change...

As far as I can tell, this means now we only compare the name and message fields when they are enumerable, which specificaly opposite from the documentation, and is a breaking change...

FWIK, As mentioned in ECMAScript 2025, name, message, cause and errors are all non-enumerable, so in the judgement above, I just compared the attributes whether they have same enumerable property, if it's same and they are non-enumerable, then compare the attributes contents.

So if the attributes are enumerable(for example in a custom Error), the next check keyCheck will check them, in the doc change, I have just added errors, and causes attributes, and didn't change name and message, I think its doesn't break the doc, have I missed something?

If it was a breaking change, I would expect we would see test failures.

assertDeepAndStrictEqual only accepts two arguments.

I believe the actual behavior unconditionally compares the errors property regardless of whether the error instance is an AggregateError or not. This could be reworded as

* [`Error`][] names, messages, causes, and errors are always compared, even if these are not enumerable properties.

Resolved. Thanks for the reminder