Posts tagged functional programming

An introduction to typeclass metaprogramming

Typeclass metaprogramming is a powerful technique available to Haskell programmers to automatically generate term-level code from static type information. It has been used to great effect in several popular Haskell libraries (such as the servant ecosystem), and it is the core mechanism used to implement generic programming via GHC generics. Despite this, remarkably little material exists that explains the technique, relegating it to folk knowledge known only to advanced Haskell programmers.

This blog post attempts to remedy that by providing an overview of the foundational concepts behind typeclass metaprogramming. It does not attempt to be a complete guide to type-level programming in Haskell—such a task could easily fill a book—but it does provide explanations and illustrations of the most essential components. This is also not a blog post for Haskell beginners—familiarity with the essentials of the Haskell type system and several common GHC extensions is assumed—but it does not assume any prior knowledge of type-level programming.

Names are not type safety

Haskell programmers spend a lot of time talking about type safety. The Haskell school of program construction advocates “capturing invariants in the type system” and “making illegal states unrepresentable,” both of which sound like compelling goals, but are rather vague on the techniques used to achieve them. Almost exactly one year ago, I published Parse, Don’t Validate as an initial stab towards bridging that gap.

The ensuing discussions were largely productive and right-minded, but one particular source of confusion quickly became clear: Haskell’s newtype construct. The idea is simple enough—the newtype keyword declares a wrapper type, nominally distinct from but representationally equivalent to the type it wraps—and on the surface this sounds like a simple and straightforward path to type safety. For example, one might consider using a newtype declaration to define a type for an email address:

newtype EmailAddress = EmailAddress Text

This technique can provide some value, and when coupled with a smart constructor and an encapsulation boundary, it can even provide some safety. But it is a meaningfully distinct kind of type safety from the one I highlighted a year ago, one that is far weaker. On its own, a newtype is just a name.

And names are not type safety.

Types as axioms, or: playing god with static types

Just what exactly is a type?

A common perspective is that types are restrictions. Static types restrict the set of values a variable may contain, capturing some subset of the space of “all possible values.” Under this worldview, a typechecker is sort of like an oracle, predicting which values will end up where when the program runs and making sure they satisfy the constraints the programmer wrote down in the type annotations. Of course, the typechecker can’t really predict the future, so when the typechecker gets it wrong—it can’t “figure out” what a value will be—static types can feel like self-inflicted shackles.

But that is not the only perspective. There is another way—a way that puts you, the programmer, back in the driver’s seat. You make the rules, you call the shots, you set the objectives. You need not be limited any longer by what the designers of your programming language decided the typechecker can and cannot prove. You do not serve the typechecker; the typechecker serves you.

…no, I’m not trying to sell you a dubious self-help book for programmers who feel like they’ve lost control of their lives. If the above sounds too good to be true, well… I won’t pretend it’s all actually as easy as I make it sound. Nevertheless, it’s well within the reach of the working programmer, and most remarkably, all it takes is a change in perspective.

No, dynamic type systems are not inherently more open

Internet debates about typing disciplines continue to be plagued by a pervasive myth that dynamic type systems are inherently better at modeling “open world” domains. The argument usually goes like this: the goal of static typing is to pin everything down as much as possible, but in the real world, that just isn’t practical. Real systems should be loosely coupled and worry about data representation as little as possible, so dynamic types lead to a more robust system in the large.

This story sounds compelling, but it isn’t true. The flaw is in the premise: static types are not about “classifying the world” or pinning down the structure of every value in a system. The reality is that static type systems allow specifying exactly how much a component needs to know about the structure of its inputs, and conversely, how much it doesn’t. Indeed, in practice static type systems excel at processing data with only a partially-known structure, as they can be used to ensure application logic doesn’t accidentally assume too much.

Parse, don’t validate

Historically, I’ve struggled to find a concise, simple way to explain what it means to practice type-driven design. Too often, when someone asks me “How did you come up with this approach?” I find I can’t give them a satisfying answer. I know it didn’t just come to me in a vision—I have an iterative design process that doesn’t require plucking the “right” approach out of thin air—yet I haven’t been very successful in communicating that process to others.

However, about a month ago, I was reflecting on Twitter about the differences I experienced parsing JSON in statically- and dynamically-typed languages, and finally, I realized what I was looking for. Now I have a single, snappy slogan that encapsulates what type-driven design means to me, and better yet, it’s only three words long:

Parse, don’t validate.

Four months with Haskell

At the end of January of this year, I switched to a new job, almost exclusively because I was enticed by the idea of being able to write Haskell. The concept of using such an interesting programming language every day instead of what I’d been doing before (mostly Rails and JavaScript) was very exciting, and I’m pleased to say that the switch seems to have been well worth it.

Haskell was a language I had played with in the past but never really used for anything terribly practical, but lately I think I can confidently say that it really is an incredible programming language. At the same time, it has some significant drawbacks, too, though probably not the ones people expect. I certainly wasn’t prepared for some of the areas where Haskell would blow me away, nor was I capable of realizing which parts would leave me hopelessly frustrated until I actually sat down and started writing lots and lots of code.