Skip to content

sitr.us

Flow is the JavaScript type checker I have been waiting for

Updated 2016-08-19 (originally posted )Flow4 min read

I am very excited about Flow, a new JavaScript type checker from Facebook. I have put some thought into what a type checker for JavaScript should do - and in my opinion Facebook gets it right. The designers of Flow took great effort to make it work well with JavaScript idioms, and with off-the-shelf JavaScript code. The key features that make that possible are type inference and path-sensitive analysis. I think that Flow has the potential to enable sweeping improvements to the code quality of countless web apps and Node apps.

From the announcement post:

...underlying the design of Flow is the assumption that most JavaScript code is implicitly statically typed; even though types may not appear anywhere in the code, they are in the developer’s mind as a way to reason about the correctness of the code. Flow infers those types automatically wherever possible, which means that it can find type errors without needing any changes to the code at all.

This makes Flow fundamentally different than existing JavaScript type systems (such as TypeScript), which make the weaker assumption that most JavaScript code is dynamically typed, and that it is up to the developer to express which code may be amenable to static typing.

Path-sensitive analysis means that Flow reads control flow to narrow types where appropriate.

This example comes from the announcement post:

The presence of a use of length with a null argument informs Flow that there should be a null check in that function. This version does type-check:

Flow is able to infer that x cannot be null inside the if body.

An alternate fix would be to get rid of any invocations of length where the argument might be null. That would cause Flow to infer a non-nullable type for x.

This capability goes further - here is an example from the Flow documentation:

The type of o is initially null. But Flow is able to determine that the type of o changes when o is reassigned, and that its type is definitely string on the last line.

In addition to null checks, Flow also narrows types when it sees dynamic type checks. This example (which passes the type checker) comes from the documentation:

The inferred return type of foo is string | number. That is a type union, meaning that values returned by foo might be of type string or of type number. The typeof checks in bar narrow the possible types of x and y in the if body to just number. That means that it is safe to multiply values returned by bar - the type checker knows that bar will always return a number.

A coworker told me that what he would like is support for type-checked structs. That would let him add or remove properties from an object in one part of a program, and be assured that the rest of the program is using the new object format correctly. Struct types work using structural types and type aliases:

Notice the type keyword and type annotation on mkPoint - Facebook's literature is a little misleading, in that Flow is really a new language that compiles to JavaScript. But the only differences between Flow and regular JavaScript are the added syntax for type aliases and type annotations, and the type-checking step. Flow can be applied to regular JavaScript code without type annotations.

How well that works will depend on how that code is written. If the JavaScript is cleanly written, you might get a lot of help from Flow without ever needing type annotations. But there are some valid JavaScript idioms that will not type-check (at least not at this time.) For example, optional arguments are not allowed unless you use either Flow's syntax or ECMAScript 6's default parameter syntax.

These are early days for Flow. I am optimistic that over time it will only get better at operating on code that was not written with Flow in mind.

There is a side benefit to using the Flow language: it supports [ECMAScript 6][], but [compiling Flow programs][] produces ECMAScript 5 code that can run in most browsers.

Edit 2016-08-19: Early in Flow's development, Facebook recommended using their own jsx compiler to process code that used Flow. That is what produced ECMAScript 5 code, as mentioned in the previous paragraph. Currently the best practice is to use Babel with the React preset to process Flow code. You still get the benefit of transpiling ECMAScript 6 code to ECMAScript 5, if you have Babel configured to do that.

One of my favorite features of Flow is that null and undefined are not treated as bottom types. A value is only allowed to be null if it has a nullable type. This makes Flow better at catching null pointer exceptions than almost any other object-oriented language.

For example:

A type expression ?T behaves pretty much like T | null | undefined. (From what I can tell, null and undefined are distinct types in Flow - but it does not seem to be possible to use them in type annotations at this time. void is allowed, which seems to be an alias for undefined.)

In the past I have talked to one or two people who said that they would only be interested in type-checked JavaScript if it supported algebraic types. (These are the kind of people I work with :)) That is possible too - here is an example that I wrote:

Thanks to type narrowing, accessing tree.value passes type-checking in the if body where tree instanceof Node is true. Without that check, the type checker would not allow accessing properties that do not exist on both Node and EmptyTree.

There are some details in this example that I want to call out:

  • The class syntax is from ECMAScript 6 - but it is extended in Flow to support type annotations for properties.
  • Tree is a type alias. It has no runtime representation. None is needed, since Node and EmptyTree have no shared behavior. And unlike a superclass, Tree is sealed, meaning that it is not possible to add more subtypes to Tree elsewhere in the codebase.
  • Flow supports parameterized types (a.k.a. generics, a.k.a. parametric polymorphism).
  • You can specify the exact type of function values (in this case, the type of predicate).
  • find might return a value from a tree, or it might return undefined. So the return type is T | void (where void is the type of undefined). Writing ?T would also work.

I have not been sold on prior JavaScript type checkers. They did not seem to "get" JavaScript. Flow is a type checker that I actually plan to use.

Revision history

2016-08-19
Edited to note that the current best practice for handling JSX is to use Babel