Hacker News Reader

Comment by esperent | Hacker News Reader

I agree that Typescript improves with each release as I and the people I work closely with use it, because we use the KISS approach to typing so we rarely do use new features, but when we do need them they're useful.

But then I find some third party code which seems to use the kitchen sink approach to typing, AKA the "look at how smart we are" approach, and it's a nightmare to understand. A type system is supposed to make code easier to work with, by the time you're spending more time understanding the types than the actual code there's something wrong.

In these cases I wish Typescript was more limited and I would gladly give up a lot of the new stuff to enforce simple types across the ecosystem.

That said, I don't really have an opinion on who is at fault here - coders for writing overly elaborate types, or Typescript for allowing them.

View on HN

> That said, I don't really have an opinion on who is at fault here - coders for writing overly elaborate types, or Typescript for allowing them.

Backwards compatibility is to blame. Typescript was designed to support a number of patterns which was already used in JavaScript.

For example events are registered in the DOM like `element.addEventListener("click", (ev) => ...)`. The type of the callback will depend on the string passed as the first argument, so the type system need to support literal strings as type discriminators which is a weird feature not necessary in most other languages. If the API had been designed with static typing in mind from the beginning, it would have been designed differently and wouldn't need such a complex type system.

Weird - using constant string discriminators feel very natural and normal to me in Typescript. What do other languages do without them?

Most model it as an actual enum, if backed by strings. Typescript, by comparison, requires being able to indicate a type just from a string argument

Allowing literal strings and type aliases of unions of strings to represent enums doesn't seem unreasonable. It's the same concept as many other languages with some extra quotes, ex. https://ocaml.org/docs/data-types#a-simple-custom-type

This doesn’t change anything:

  foo(get_some_myenum_value(), …)
  {type:get_another(), …}

These are as unguessable as if myenum was a string.

Hmm. That doesn't feel like much of a difference to me?

It's huge? I'm really confused lol. I write in Dart/Java/C++/ObjC/TS regularly to maintain a cross-platform library and this is a massive difference from every typechecker in any other language, much less the narrower thing of if the idea of reifying random strings into an enum is a novel feature. I get a very strong sense from the thread that people are talking past eachother, somehow.

The event type could just be exposed as a regular property so you could write `element.onclick.add((ev => ...)` so you would get static typing without any need for magic strings.

Magic string are use all over JavaScript for different purposes. For example createElement('P') could just be new PHtmlElement() ...

In other cases magic strings could be replaced with enums.

Changing Javascript is explicitly out of scope for Typescript.

I was curious how other languages did discriminators in a way that makes Typescript's feel alien. Just having different functions doesn't really tackle that.

The pattern I'm most familiar with is using type constructors:

   type Shape = Square Int | Circle Int

and then pattern matching:

   print (Square size) = ..
   print (Circle radius) = ...

In Typescript this would be something like:

   type Shape = { type: 'square', size: number } | { type: 'circle', radius: number }

And string matching for type narrowing:

   function print(shape: Shape) {
     if (shape.type === 'square) {
       ...
     } else if (shape.type === 'square) {
      ...
     }
   }

The TypeScript approach is comparatively weird and verbose but also pretty clever since it doesn't require any change to the core language.

It also have some pitfalls, e.g. you can write:

  print({type: 'square', size: 10});

But you cant write:

  const x = {type: 'square', size: 10};
  print(x); <-- type error

You have to write:

  const x = {type: 'square' as const, size: 10};
  print(x);

Things like this just complicates the type system and type inference.

This example looks like a simple sum type with method overload. Don't think that's complex, that's bread and butter in some functional languages.

Now take mapped types, template literal types, key remapping... Those features aren't necessary to express legacy DOM APIs, this is just for the sake of having a complex typing system. And that's not to add a new feature that unlocks some type magic that wasn't possible before. No, those are just shorthands for things that you could already express in the existing typing system.

It is not just to support the DOM it is also frameworks like JQuery, React, Vue etc.

For example the 'options' parameter is a common pattern where an object with a set of properties are provided to override a some default behavior. So you need to define a Foo type with a set of properties, and then you need to define a type which is like Foo except all properties are optional.

Some frameworks implement their own flavor of inheritance e.g. through mixins. The type of an object with mixins is not just the sum of all properties, you might also have one property overriding a property of a different type in base mixin, so you are adding and removing properties. You need a really powerful type system to be able to represent this.

> A type system is supposed to make code easier to work with, by the time you're spending more time understanding the types than the actual code there's something wrong

This may be true in a lot of cases, but certainly not all.

I’ve spent a ton of time on the types that validate that the input/output schema of my request handler function matches the defined schemas, and that was hell, but now any other of our developers will see delightfully red squiggles when they do something wrong. That’s a force multiplier and absolutely worth the effort.

Also agree, with each release I get an "Eureka effect" that now I can solve the type issue I strugled couple months ago trying to create to just make some highly used function safer/easier to use for the developers.

Example the new satisfies and some upcomign "as const" features to generics I'm looking forward to

What's bad about using clever types? My (limited) experience is that if they're done well they just work. I don't know why React-Query's types work the way they do, but I quite like that they do. Developing an intuition for it was a bit hard, but I believe this might rather be a failure of the documentation.

And if they aren’t, good luck debugging it. I remember recently some typing issue in a webpack config where a correct key was marked as a type error in a presence of another construct, despite being documented and having an effect.

Also try exploring npm/telegraf classes (auto-documented) by looking at their type definitions. The whole bot api could have been written as simple repeated definitions, but someone chose the clever way of multi-level type mutations and derivatives.

Typescript will soon need secondary community-maintained type contracts that are human-readable and cross-checked with those in libraries. Something like @humane-types/*.

Reply on news.ycombinator.com