Nice article, but I don’t fully understand why the author argues for such an aversion to static functions.
> Moreover, it breaks the systematic approach of sending messages to an instance (often presented as one of the key arguments in favor of object-oriented programming).
Is this just a matter of “the code will become spaghetti once too many classes/methods/implementations exist”? And, conversely, is a non-static method with a constrained, i.e. somehow different type (or also the approach of moving a method outside of the class proper) not more confusing?
I'm building a new language and just a couple of days ago the concept of guard methods came up as I was trying to tighten up equality semantics to be more like Swift.
Things like Array.contains() only work if the element type implements the Equatable interface, so it would be a guard method. Maybe something like:
class Array<T> {
contains(value: T): boolean where T extends Equatable { ... }
}
Or possibly a constraint on the `this` type, TypeScript style:
> Although guarded methods seem necessary, unfortunately, I don’t know of any mainstream languages that allow their definition.
C++ has this feature. When you define a template class, you can use SFINAE (or, in modern C++, concepts) to make it so that certain methods only exist if the template parameter meets certain requirements.
(Though even this is often unnecessary - for something like your `flatten` example, you could just go ahead and define it unrestricted. Methods of template classes are typechecked lazily - in other words, they don't need to successfully typecheck unless they're called. For this specific use case, SFINAE/concepts would just make the error message nicer.)
Rust also has this feature, with conditional impls.
> Rust also has this feature, with conditional impls.
Rust's approach seems more similar to the approach described in the section on extension methods, which the author admits solves the problem but doesn't like the implications for encapsulation (which isn't a problem in Rust, as long as you're writing this method in the same crate as the type definition). But Rust also doesn't have classic Java-style classes, so the requirement to "keep the member definition within the class" is already meaningless in Rust terms.
Idiomatically Rust explicitly prefers to make a type more generic, for example it's fine to talk about a HashMap<Goose,String> for some arbitrary user defined type Goose with no traits, and Rust will even cheerfully make you one... but you can't put any key+value pairs into that hash map because Goose isn't Eq [or Hash] so all the APIs for actually inserting or modifying the data don't exist on this type.
In contrast C++ won't even let you have std::unordered_map<Goose,int> because it wants to know how to compare and hash the keys before making the type at all.
Nice article, but I don’t fully understand why the author argues for such an aversion to static functions.
> Moreover, it breaks the systematic approach of sending messages to an instance (often presented as one of the key arguments in favor of object-oriented programming).
Is this just a matter of “the code will become spaghetti once too many classes/methods/implementations exist”? And, conversely, is a non-static method with a constrained, i.e. somehow different type (or also the approach of moving a method outside of the class proper) not more confusing?
Cool to see this pop up today.
I'm building a new language and just a couple of days ago the concept of guard methods came up as I was trying to tighten up equality semantics to be more like Swift.
Things like Array.contains() only work if the element type implements the Equatable interface, so it would be a guard method. Maybe something like:
Or possibly a constraint on the `this` type, TypeScript style: https://github.com/elematic/zena/blob/8d77f2b36001078f4d5054...> Although guarded methods seem necessary, unfortunately, I don’t know of any mainstream languages that allow their definition.
C++ has this feature. When you define a template class, you can use SFINAE (or, in modern C++, concepts) to make it so that certain methods only exist if the template parameter meets certain requirements.
(Though even this is often unnecessary - for something like your `flatten` example, you could just go ahead and define it unrestricted. Methods of template classes are typechecked lazily - in other words, they don't need to successfully typecheck unless they're called. For this specific use case, SFINAE/concepts would just make the error message nicer.)
Rust also has this feature, with conditional impls.
> Rust also has this feature, with conditional impls.
Rust's approach seems more similar to the approach described in the section on extension methods, which the author admits solves the problem but doesn't like the implications for encapsulation (which isn't a problem in Rust, as long as you're writing this method in the same crate as the type definition). But Rust also doesn't have classic Java-style classes, so the requirement to "keep the member definition within the class" is already meaningless in Rust terms.
Idiomatically Rust explicitly prefers to make a type more generic, for example it's fine to talk about a HashMap<Goose,String> for some arbitrary user defined type Goose with no traits, and Rust will even cheerfully make you one... but you can't put any key+value pairs into that hash map because Goose isn't Eq [or Hash] so all the APIs for actually inserting or modifying the data don't exist on this type.
In contrast C++ won't even let you have std::unordered_map<Goose,int> because it wants to know how to compare and hash the keys before making the type at all.