Template Haskell

zw3rk
zw3rk
May 23, 2017 · 4 min read

After using Haskell for a while, running into Template Haskell become more and more likely. A reverse dependency lookup for template-haskell lists over one thousand packages.

While I believe many have used Template Haskell via aeson, language-c-inline, lens, wreq, yesod, or any of the other packages, my gut feeling is that fewer have actually written Template Haskell code.

This post is not intended to advocate the use of Template Haskell, but rather to foster a better understanding for what it is, and how it can be used.

Template Haskell is described as a facility that adds compile-time metaprogramming to Haskell. Metaprogramming is the ability of a language to treat code as data, and write code to produce other code to be used in the program. Other languages have (limited) metaprogramming facilities as well: e.g., C has macros via the C preprocessor, that allow to expand code at compile time as well, C++ has preprocessor macros and templates, and the languages of the Lisp family usually have macro systems and are known for their homoiconicity and metaprogramming capabilities.

The C preprocessor

To use metaprogramming facilities we need a way to say how we want to generate code, and a way to specify where to generate code. The C preprocessor does this on a textual basis with a separate language. We can define a macro:

#define ADD_ONE(x) ((x)+1)

this will simply replace all occurrences of ADD_ONE(x) with (x)+1 prior to handing off the code to the compiler. For example:

int x = ADD_ONE(1);

becomes

int x = (1)+1;

Template Haskell Quotes

Contrary to the C preprocessor, Template Haskell operates on the abstract syntax tree (AST) level. Again we need a way to describe what we want the compiler to generate, and a directive to describe where to generate it. GHC expects a function that generates code to live in the Q monad. The Q monad governs its capabilities. Thus our add_one function needs to be of the form:

add_one :: Int -> Q a

We are free to generate different kinds of Haskell syntax, e.g., Types, Declarations, Patterns, Expression, …; here we clearly want an Expression:

add_one :: Int -> Q Exp

The first interesting question is, how to build this Exp Expression type. The template-haskell Haddocks for expressions are a good start. We end up with:

add_one :: Int -> Q Exp
add_one x = pure $ InfixE (Just (LitE (IntegerL x)))
(VarE '(+))
(Just (LitE (IntegerL 1)))

We translate x :: Int into a literal expression with an integer literal of x, + into a variable expression, and then combine those using the InfixE constructor for infix expression.

There has to be an easier way, otherwise no one would bother writing Template Haskell in the first place. And of course there is. Template Haskell comes with quotation support. The idea is to quote some expression (or type, declaration, pattern, …) and obtain the above expression as simple as:

add_one :: Int -> Q Exp
add_one x = [| x + 1 |]

There are also quoter for declarations, type, patterns, … or you can even define your own quoters using -XQuasiQuotes.

See also: The GHC User Guide on Template Haskell

Template Haskell Splices

So far we learned how to create a function that can generate a Q-action. Ultimately we want to convert the generated expressions to actual code in our program. GHC calls this splicing. We splice the result of calling the function (e.g. add_one) into our program. A splice is written as $x or $(...), where x is an identifier or ... is an arbitrary expression. Thus if our add_one function lives in our Lib module, we can build the following Consumer module, which makes use of add_one, as follows:

{-# LANGUAGE TemplateHaskell #-}
module Consumer where

import Lib (add_one)

two :: Int
two = $(add_one 1)

GHC will at compile time evaluate the splice $(add_one 1), and put the result of evaluating add_one 1 in the Q monad into its place.

The Stage Restriction

You might be curious as to why we (had to) placed add_one into its own Lib module. This is often referred to as the stage restriction. The functions we want to call in our splices need to be defined outside of the module where they are used. This is necessary since they need to be compiled first. GHC compiles the module which contains the functions we want to use in our splice first to machine code, and when evaluating the splice, builds a tiny bit of interpreted bytecode that wraps the arguments and calls the function.

The Q Monad

I’ve mentioned the Q monad earlier and said it governs the metaprogramming capabilities of the functions we call in our splices. The Quasi type class describes the interface that we can make use of when constructing Template Haskell functions.

Of special note should be the qRunIO :: IO a -> m a. Which is wrapped by runIO :: IO a -> Q a. This function allows to execute arbitrary IO actions at compile time! This allows capabilities like reading (e.g. file-embed) and writing arbitrary files or running commands/processes (e.g. gitrev). While this power can provide great utility, I believe we should be well aware of this fact when compiling arbitrary third party code.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch

Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore

Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store