Skip to content

sitr.us

Flow Cookbook: Unpacking JSON API data

Flow, Flow Cookbook4 min read

This recipe is part of the Flow Cookbook series.

Hacker News provides a public API. One of the endpoints of that API accepts an ID and responds with an item:

Here is an example of a response:

An "item" might be a story, a comment, a question, a job posting, a poll, or a voting option in a poll. Each item type has different properties - for example stories have a title, but comments do not. If you don't have context for the ID, the only way to know what you have fetched is to check the type property of the response at runtime. To add type safety to a call to this API, it is necessary to describe a type that encompasses all of the possible shapes that the returned data might take. That is going to be a union type, which will look something like this:

There are not a huge number of properties in the API responses - but if we list out all of the properties in every branch the result will be too big and dense for light reading. So let's start by factoring out common properties into helper types.

All of the different item types have by, id, and time properties. So we can put those all into one type:

Those Username and ID types are just aliases that I defined for primitive types:

I think that using aliases like these helps to provide clarity on the purpose of each property. If we had a property with the type by: string it would not be obvious whether the value of that property is an ID that happens to be a string, or a human-readable value. Using the Username type alias makes it obvious that the property will contain a value that might be suitable for display to users. Otherwise the types string and Username are interchangeable.

There is more common structure in Hacker News item types: the story, ask, job, and poll response types all represent top-level submissions, which have several properties in common:

With those helper types in place, we can produce a type that describes all possible items:

An intersection type like { type: 'story', kids: ID[], url: URL } & ItemCommon & TopLevel is essentially a shorthand for an object type with a type property that is always equal to 'story', combined with all of the properties listed in the ItemCommon and TopLevel types. Each branch of the union type contains a type intersection that combines common properties with the properties that are particular to each item type.1

type Item = ItemCommon & (/* union type */). That would put the union type inside of the intersection type. But due to a quirk in Flow as of version 0.36 the union type must be the outermost layer of of composition for type narrowing to work.

We don't have to do anything special to parse incoming data into that type. Flow types are duck types - Item is just an alias for plain Javascript objects with a certain structure. We just need to declare that API responses have the Item type. That is done with with the return type in this function signature:

You might have noticed something a little strange about the types of the type properties in Item. We used string literals where there should have been type expressions! For example, we gave 'story' as a type in the first branch of the union type. In fact string literals are types. In a type expression, the literal 'story' is a type with exactly one possible value: the string 'story'. ('story' is a subtype of the more general type, string.) This is useful because it signals to Flow which branch of the union type is applicable inside the body of a case or if statement.

Consider this function, which does not type-check:

Flow can narrow the type of a variable in certain contexts. A runtime comparison with a static string literal does the trick:

Hacker News does provide endpoints for fetching recent submissions of a specific item type (e.g., the latest stories). But to demonstrate the flexibility of the Item type, let's write some code that fetches and displays the latest items of any type. We will need to switch on the type property of each item to display it properly:

Flow is able to infer which item type is given in each case body. This is just like how type-narrowing worked in the if body in the getTitle function.

Flow's checking has an added bonus: if you have a case body with no return or break statement, execution falls through into the next case body. When switching on item.type, a fall-through would result in a situation where a case body might be executed with any of several different item types. For example:

Flow allows this, because all of the types listed in that example have a title property. But if a case body did something not compatible with all of the different item types that could fall-through into it, then Flow would report an error.

Next up are functions to determine which items to fetch, and to make the necessary requests:

And finally, some code to set everything running:

Refining the model

Later on we may realize that it would be useful to be able to refer to each item type individually. To do that, we can create a named alias for each item type:

Then we can replace the earlier definition of Item with a simpler one:

This well let us write specialized functions, such as a function that specifically formats a poll with its options.

So how do we get to a point where we can call a function that accepts only polls? The answer is, once again, type-narrowing:

Notice the use of flatMap in fetchPollOpts. This filters results to check that the results are actually poll options. At the same time, Flow is able to infer that the filtered results all have the PollOpt type. This uses a custom definition for flatMap:

If you trust that all of the items that are fetched will be of the right type, and you do not want to bother with a runtime check, then you could use a type-cast instead:

Finally, here is a function that feeds fetched items to the new-and-improved item formatting function:

The code from this article is available at https://github.com/hallettj/hacker-news-example. I encourage you to check out the code to tinker with it. Try building more functionality, and see how type-checking affects the way you write code.


  1. top-level union