Bug: [unbound-method] should relax checks within `Reflect.apply`?

[andreww2012]

Bug: [unbound-method] should relax checks within `Reflect.apply`?

[Josh-Cena]

The rule is never aware that a method is able to bind the this value, because it never knows. Otherwise, you have to special-case a lot of array methods too.

[bradzacher]

Josh is correct - the rule is purposely shallow in how it reports on usage. It does not understand the inner workings of function calls - even those in the stdlib. Mostly because the complexity of building out support for all of the various forms that are safe is large and expensive and ultimately not worth the cost - both maintenance cost and runtime cost - given that such usecases are super rare in the general case.

Eg most people don't use the Reflect API because is relatively new and in particular Reflect.apply is really no better than the more direct method.apply.

I'm okay with the rule not handling these cases. If we want to document the edge case, we can?

[andreww2012]

Thanks for the response - I largely agree with you, I wasn't even sure if it's worth reporting.

Still want to leave here my implementation idea just in case there's any demand for covering such use case:

• Only check static method usages, for example Reflect.apply, Array.prototype.map.call(...) and not someVar.map(...) (but perhaps <array literal>.map(...)?
• Make this check opt-in if it runtime costly.

[Josh-Cena]

This rule already relies on type information, so checking someVar.map is not any harder than Reflect.apply, because we can use type information to know that someVar is an Array. The only problem is whether it's worthwhile to do so because we tend to not like carving special cases for built-in functions.

Maybe we need some generic way to exempt methods from being checked (or declare that they have a thisArg parameter), and add these built-ins as default options? Sounds really complicated to do right, though. Again, I don't really think this happens that often now.

[bradzacher]

The only problem is whether it's worthwhile to do so

My vote is that it's not.

The number of people that use that "this" arg is so, so, so low. "this" meta programming went out of style with TS cos it's a PITA to safely write.

The cost of writing the code plus the (negligible) cost at runtime just isn't worth it to catch an 0.0001% case, IMO.

[Josh-Cena]

I'm +0.8 on what Brad said, the rest 0.2 of reservation comes from me being a part time JavaScript Ninja that likes "obscure" patterns.