Use the REPL, Luke

There was an online discussion about iteration times in Haskell and whether and why they are slow. For me, it’s not slow. I do all my Haskell development using a REPL. Here are some tips I wrote up in that discussion.

Prepare for GHCi use

The first thing you want to do before writing anything for your project is make sure you can load your code in the REPL; GHCi. Sometimes you have special configuration options or whatnot (cabal repl and stack ghci make this much easier than in the past). The sooner you start the better. It can be a PITA to load some projects that expect to just be a “start, run and die” process, they often launch threads without any clean-up procedure; in this way the REPL makes you think about cleaner architecture.

Make sure it scales

Learn how to make GHCi fast for your project so that you don’t hit a wall as your project scales. Loading code with byte-code is much faster than object code, but loading with object code has a cache so that in a 100 module project if you only need to reload one, it’ll just load one. Make sure this is happening for you, when you need it. Dabble with the settings.

Write small, parametrized functions

Code that is good for unit tests is code that is good for the REPL. Write small functions that take state as arguments (dependency injection) rather than loading their own state, then they can be ran in the REPL and used in a test suite easily. Regard functions that you can’t just call directly with suspicion.

Test work-in-progress implementations in the REPL

While writing, test your function in the REPL with typical arguments it will expect, rather than implementing a function and then immediately using it in the place you want to ultimately use it. You can skip this for trivial “glue” functions, but it’s helpful for non-trivial functions.

Setup/teardown helpers

Write helpful setup/teardown code for your tests and REPL code. For example, if you have a function that needs a database and application configuration to do anything, write a function that automatically and conveniently gets you a basic development config and database connection for running some action.

Make data inspectable

Make sure to include Show instances for your data types, so that you can inspect them in the REPL. Treat Show as your development instance, it’s for you, don’t use it for “real” serialization or for “user-friendly” messages. Develop a distaste for data structures that are hard to inspect.

Figure out the fastest iteration for you

Use techniques like :reload to help you out. For example, if I’m working on hindent, then I will test a style with HIndent.test chrisDone "x = 1", for example, in the REPL, and I’ll see the output pretty printed as Haskell in my Emacs REPL. But I work on module HIndent.Style.ChrisDone. So I first :load HIndent and then for future work I use :reload to reload my .ChrisDone changes and give me the HIndent environment again.

Configuration

Make sure you know about the .ghci file which you can put in your ~/ and also in the project directory where GHCi is run from. You can use :set to set regular GHC options including packages (-package foo) and extensions (-XFoo), and any special include directories (-ifoo).

More advanced tricks

Consider tricks like live reloading; if you can support it. I wrote an IRC server and I can run it in the REPL, reload the code, and update the handler function without losing any state. If you use foreign-store you can make things available, like the program’s state, in an IORef or MVar.

This trick is a trick, so don’t use it in production. But it’s about as close as we can get to Lisp-style image development.

In summary

Haskell’s lucky to have a small REPL culture, but you have to work with a Lisp or Smalltalk to really know what’s possible when you fully “buy in”. Many Haskellers come from C++ and “stop program, edit file, re-run compiler, re-run whole program” cycles and don’t have much awareness or interest in it. If you are such a person, the above probably won’t come naturally, but try it out.