`Error.prototype.cause` has inaccurate value type

[crowlKats]

lib Update Request

Missing / Incorrect Definition

Error.prototype.cause's value is currently set to Error, which is inaccurate as it can accept anything, and as such should be unknown.

Sample Code

new Error("foo", { cause: "bar" });

Documentation Link

https://tc39.es/ecma262/#sec-installerrorcause

[DanielRosenwasser]

The cause property on Error instances might need to be typed as unknown or any, but the cause property on the options bag is possibly one of those places where we'd want to take a more prescriptive stance.

[fatcerberus]

I'm not so sure about that - if options.cause is typed as Error and the project has enabled useUnknownInCatchVariables then you can't write (and note that this is the intended use case of cause):

try {
    superFoo();
}
catch (err) {
    throw new Error("superFoo failed!", { cause: err });  // type error
}

...without casting to Error, which is questionable if things are being thrown that aren't Error instances (and since TS has no checked exceptions, you can't guarantee that some dependency, deep into a call chain, isn't doing exactly that). You also can't just write catch (err: Error) {} because, again, that requires checked exceptions.

[RyanCavanaugh]

I sort of wonder how much we pretend that non-Error exceptions exist vs how often end-users write code on the assumption that they don't

[Jamesernator]

As suggested to me cause could be used for giving extra info for things like typing or range errors.

i.e. You might want to do something like to attach the value that caused an error:

if (value === null) {
    throw new TypeError("value should not be null under the current state", { cause: { value } });
}

if (n < 0) {
    throw new RangeError("n should not be negative", { cause: { n } });
}

[neovov]

There is also some DOM errors that doesn't extends the Error interface.
For example MediaError doesn't have a name or stack properties, making it incompatible with cause.

[ljharb]

The proposal for causes was designed very intentionally to match the type of “what can be thrown” - which is only unknown. It’s also why an absent cause is different than a present undefined.

[ljharb]

I sort of wonder how much we pretend that non-Error exceptions exist vs how often end-users write code on the assumption that they don't

sounds like a great argument for TS to teach them that their assumption is incorrect with a helpful type error when they forget to explicitly narrow.

[conartist6]

My two cents: @ljharb says the spec has been written very carefully so that I can tell the difference between an error with no cause and a cause which has a useless value like null, which even so may really have been thrown.

I say this: whatever the spec says, anyone can still put anything in error.cause. They already do, and I don't have the luxury of assuming they were looking at the spec when they did it. If i get null as a cause, I'm just going to say that an error had no cause (no usable, meaningful cause). I don't see why I would spend many brain cells trying to figure out what someone was thinking when they wrote throw null, and if I did catch a thrown null, I would not put it in the cause of an error but would instead rethrow something like Caught unexpected primitive value `null`