From Flow to Typescript. Why?

Nikita Sobolev - Apr 2 '19 - - Dev Community

Originally published in my blog: https://sobolevn.me/2019/03/from-flow-to-typescript

It all started almost two years ago. I was sick with constant silly javascript mistakes like undefined is not a function in my apps. So, I have decided to add optional static typing.

Two years ago the javascript land was completely different. Both Flow and TypeScript had a lot of disadvantages: poor libraries support, from none to almost none IDE support, type issues, and limitations. I have chosen Flow just because it was easier to start with. Plug it into your .babelrc and start working.

Almost 6 months ago I have made a decision to move all our frontend projects from Flow to TypeScript. It was a tough battle inside my head. So, I have decided to write it down to help other people to choose the right tool. And not to make my mistakes.

These tools are quite similar, both providing type-safety to plain javascript. And this article is not about types or differences between structural and nominal subtyping. It more about the current state of these tools.

You rarely change typing tools because of types.

Hype

I will start with the most important aspect of all technical decisions. Yes, it is about hype-driven-development.

Hype ORLY

Please, do not make your face like this. I will explain how it works, and possibly you will change your mind.

I quite often speak about our stack, write articles about it, etc. And every time I said that we were using Flow other developers asked me: "But why not TypeScript"? And every time I had to explain my choice. Going deep into details, history, and tooling. Sometimes even explaining the state of current projects and our workflow. This was not a pleasant experience when you are not advocating for something. I was just a user. I do not really prefer one to another. And it felt really strange to had this conversation over and over again.

Moreover, we make services for our clients, and we hire other developers. And some of them literally wants to work with TypeScript and does not want to work with Flow. Why? Because they heard about TypeScript and how awesome it is. While the hype train around Flow is not so massive.

Me: Hi, we use X, Y, Z, and Flow for this project

Dev: Flow? Why not TypeScript?

Me: Oh, are kidding me? 🤦

If you do not want to fight this hype train but make it work for you it is better to step aside and use whatever is hyped. Keeping in mind that there's no massive change for your current workflow.

Infrastructure

Vue 3.0 will support TypeScript out of the box, Nuxt already supports TypeScript. There are a lot of projects that ship types together with the source code. axios, vuex, vue-router, lodash to name a few.

What about Flow support? Vue currently uses Flow for typing (switching to TypeScript starting from 3.0), but these types are development only. You cannot take them and use in your own projects.

Maybe there are different types? Yes, Flow has its own repository for types. The problem is that installing types is a whole new extra step. You have to configure postinstall hook to make sure that types are also rebased after each npm install call (yes, they use git rebase).

When you will dig into flow-typed packages you will find that it is mostly React oriented. Flow even has a lot of React primitives in the standard library. Which I find really strange.

But what about Vue specific types (since we use Vue a lot)? Yes, you can find @vue-flow-type package maintained by a single person. The sad thing is that I am this single person. I am really tired of maintaining types for several popular projects. And as you can imagine there are bugs, type changes, new releases, etc.

TypeScript wins this one for me. Its distribution system does not force me to do extra work. Just install something and it will work. Because types/ subfolder is shipped together with the source code via npm. No extra steps are required. Library authors maintain types/ folder together with the rest of codebase. They can be sure that everything works correctly.

IDE

Let's discuss IDE support for Flow projects. Or I would say "no IDE support".

That's not a big thing, I can write my code using just nano. But I spend a lot of my life inside text editors and I want them to be friendly. Sadly, Flow plugins for all major IDEs (and text editors) are buggy and unreliable. For example, VSCode plugin does not work at all.

And at the same time, VSCode is known for its first-class TypeScript support. With intellisense, type-checking, and autocomplete out of the box.

vetur plugin for vscode

Have a look at our VSCode + TS + Vue setup.

With this simple feature, your development workflow starts to feel more responsive, and the feedback loop time is significantly decreased.

Unfixed bugs

The other thing that was ruining my Flow experience is the number of unfixed bugs in Flow itself.

For example, when you install Vuex all Vue components are extended with an extra property which can be accessed with this.$store. The thing is there's no way to tell Flow that Vuex was added. And this bug is opened since 2015, for 4 years now!

Of course, you can still write your own types:

// @flow

import Vue from 'vue'
import type { Store } from 'vuex'

import type { State } from '~/types/vuex'

/**
* Represents our extended Vue instance.
*
* We just use the annotations here, since properties are already injected.
* You will need to add new annotations in case you will extend Vue with new
* plugins.
*/
export default class CustomVue extends Vue {
  $store: Store<State>
}
Enter fullscreen mode Exit fullscreen mode

But now you have to maintain your own types by yourself. Do you want to add this.$router property? Please, add it yourself. Nuxt specific types? You are on your own.

Compare it with the standard TypeScript approach:

import Vue, { ComponentOptions } from "vue";
import { Store } from "./index";

declare module "vue/types/options" {
  interface ComponentOptions<V extends Vue> {
    store?: Store<any>;
  }
}

declare module "vue/types/vue" {
  interface Vue {
    $store: Store<any>;
  }
}
Enter fullscreen mode Exit fullscreen mode

Existing types can be extended with special declarations. And library authors do that for you. Remember what I said about types distribution? This feature makes the distribution even better.

The second well-known bug from 2015 is that you cannot annotate this even if you have to. Some libraries have strange APIs. With Flow you just cannot do anything, typing is lost there. But with TypeScript you can annotate what this means in every context. And it is great for a lot of use-cases.

Why these bugs are not fixed? I don't know. They brought a lot of attention during the years. A lot of people wanted these things, but Flow team does not share their vision on the project. And they release things that they want, not the community.

Releases

Talking about releases I must mention their policy: "just release things and make users fix their code". Here's the release history and what it has done to my project. Almost every release breaks my code. Considering that it is a template with almost no code – it is really scary.

By the way, Flow team does not follow SemVer, they just release incrementally. After one release jsx inside .vue files stopped working. I was not able to fix it ever again on new versions. I took the lazy path: pinned the version and just ignored the updates after this incident.

TypeScript has clear release policy, SemVer, and wide attention to the community. It is much better to maintain in the long run.

Conclusion

We have made our choice and said "Good bye" to Flow. Now all our projects and our project template supports TypeScript. And we regret nothing!

By the way, our template is truly awesome. It supports:

  • Nuxt for server-side rendering and boilerplate isolation
  • TypeScript everywhere: code, tests, configuration
  • Jest for unit tests, TestCafe for E2E tests
  • Docker for development and production
  • Awesome documentation that covers every aspect of the project

Try it out!

. . . . . . . . . . . . . . . . . . . . .