Hands-on with Clojure

:: Clojure, Hacker School

Until now, Clojure has been an “armchair” language for me. I’ve read a lot about it, and I’ve read a fair amount of code written in it. But aside from typing a few things into an online REPL like Try Clojure, I’ve not really used it hands-on.

A couple days ago I started my 12-week batch at Hacker School. First on my list is to get some real experience with Clojure.

And I’m going to blog about the experience.


A couple caveats:

  1. Hacker School encourages you to blog early, blog often. This post won’t necessarily be super well thought-out. Expect me to write some really dumb Clojure code, and state some silly opinions about Clojure.

  2. I’m coming from a few years of using Racket — and especially during the last year feeling very comfortable and reasonably proficient. Obviously Clojure is not Racket, in ways that are probably some mix of better, worse, and just different. But I may get cranky about a few things. I expect some of that will turn out to be “early days” and I’ll feel better later.

In other words: I welcome comments that help me understand how to do something better or more idiomatically. However please cut me some slack on my subjective reactions, because I’m not trying to flame, here.


JVM

Back when there was that security issue with the JVM on OS X I removed the JVM from my MacBook. The fact that Clojure uses the JVM is a huge plus for some people. For me, it’s meh. Anyway, no choice. First order of business is to install it.

I decide to download JDK 8. Later I hear that wasn’t necessarily the best choice:

Amusingly I mistakenly choose to download to ~/Dropbox not ~/Downloads. Although I think oracle.com was slow, it probably didn’t help matters that I was re-uploading the 200 MB back to Dropbox. Derp.

While waiting for the download to complete, I spend time re-working through the early chapters of Learn You a Haskell for Great Good, since I plan to dive into Haskell and/or Idris in later weeks.

After I installed the JDK I tried:

1
2
3
4
$ java -version
java version "1.8.0_20"
Java(TM) SE Runtime Environment (build 1.8.0_20-b26)
Java HotSpot(TM) 64-Bit Server VM (build 25.20-b23, mixed mode)

Erm. What. Version 1.8 doesn’t sound like 8. Also which and ll show file dates that don’t seem new enough.

So apparently I already had JRE on here, despite removing it (I thought) back around the time of the OS X security bug. Or, is JDK 8 the “8” in JRE 1.8? Huh??

I spend some time trying to figure this out, with no luck.

Eventually I need to find where JDK 8 actually was installed, and make sure it’s in my PATH.

But meanwhile I decide to try ignoring it, and forge ahead and install lein and Clojure.

I followed part of A Brief Beginner’s Guide to Clojure. I went through its examples using lein to create a new project, start a REPL, and so on. Worked fine.

Clojure and Emacs

Next, how to get this working in Emacs?

I found Emacs and Clojure, a guide for getting set up with Clojure on Emacs. Following that worked out pretty well. Just a couple problems of my own creation.

Initially, cider-jack-in would complain that pretty-regexp-alist wasn’t defined. Some searching turned up issue 833. Yep, I had an old clojure-mode from doinking around with just editing clojure code in the past. So I updated clojure-mode.

Now, C-c M-j cider-jack-in is working. I find it weird that I must both 1. C-c C-k to compile the buffer1, then 2. C-c M-n to enter the namespace… and only then can I use the REPL. This seems awkward compared to just F5 in racket-mode. Is there a good reason for separating the two, compiling vs. entering the ns? Probably. Anyway, I suppose I could bind F5 in clojure-mode to the composition of compile and enter. For now, I just learn to type both when needed.

So things are mostly working, except two things:

Well, maybe I should have paid attention to those warnings about mismatched versions. I had ended up with cider 0.8.0-SNAPSHOT from MELPA, but cider-nrepl 0.7.0 from copy-pasta. Eventually I figured out how to create a ~/lein/profiles.clj and set it to

1
{:user {:plugins [[cider/cider-nrepl "0.8.0-SNAPSHOT"]]}}

Now all seems well. Overall Emacs and Clojure really helped me understand the pieces. Big thanks to the author, Gareth Jones, and the other people who contributed it.

My First Project, and require

From my armchair Clojure I suspected I’d have some “uncanny valley” moments, where Clojure and Racket seem the same but are subtly different. I wasn’t disappointed. Right off the bat I struggled with require. I was trying to write penultimate. At first I thought, oh in Racket I’d use match for this. So I tried to require core.match. What I didn’t realize is that Clojure’s :require imports things under a name with a namespace prefix, by default.

I added it to :dependencies in my project.clj:

1
2
  :dependencies [[org.clojure/clojure "1.6.0"]
                 [org.clojure/core.match "0.2.1"]]

That’s fine. But trying to require it in my core.clj:

1
2
(ns my-lib.core
  (:require [clojure.core.match :only (match)]))

Gives an error:

=> #<CompilerException java.io.FileNotFoundException: Could not locate clojure/core/match__init.class or clojure/core/match.clj on classpath: , compiling:(my_lib/core.clj:1:1)>

And I have no idea really what the hell this means or what to do about it.

I gave up on core.match, realizing I could do what I want with Clojure’s destructuring let. Which is awesome sauce. But then, I wanted to make some unit tests. And I had the same problem trying to use clojure.test.

Eventually I realized that if I wanted to refer to deftest and is imported from the clojure.test namespace, I had to do it either of two ways:

  1. Use (:require [clojure.test :refer :all]) so I could refer to deftest and is without any namespace prefix.

  2. Use (:require clojure.test) and prefix e.g. clojure.test/deftest and clojure.test/is.

In other words, by default, Clojure’s require is backwards from Racket’s with respect to name prefixes. I’m not sure I really like Clojure’s choice here. In my not-insubstantial experience with Racket, name collisions are infrequent. They seem like the special case that should require (no pun) special work — not the other way around. (When name collisions do occur in Racket, it’s easy to use Racket’s only-in, and/or use prefix-in to create some prefix. You could even do (prefix-in foo/) to make prefixes in a superficial Clojure style).

Tests

After I managed to get a deftest in my core.clj file, I noticed the ../test/my_lib/core_test.clj file created by Leinengen. Ah. The tests are supposed to go there. And if I put them there, lein test and cider-run-all-tests will work.

So that’s good, but I don’t love is that I must put tests in a separate file. I’ve really grown to like Racket’s submodules, which enable things like putting things in a submodule named test, in the same file as what’s being tested.2

Code

Here’s what I ended up with right before it was time to head out for dinner:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
(ns my-lib.core
  (:require [clojure.test :refer :all]))

;; A (probably dubious) implementation of penultimate.
(defn penultimate-dubious [xs]
  (let [[x & xs] xs]
    (if xs
      (let [[y & xs] xs]
        (if-not xs
          ;; y is last, therefore x must be next-to-last
          x
          ;; keep trying
          (recur (cons y xs))))
      nil)))

(defn penultimate [xs]
  (let [[x & [y & ys]] xs]
    (if ys
      (recur (cons y ys))
      (and y x))))

;; See ../test/my_lib/core_test.clj for tests. Unfortunately nothing
;; like Racket's test submodules.

Although penultimate seems OK to me, I’m curious about other ways to express that in Clojure.

Conclusions and next steps

What conclusions can we draw from all this? Nothing too important. This is just a brain dump of my experience and opinions. It’s probably fair to say that the “onboarding” for Racket is simpler, in terms of getting a working development environment, writing and evaluating code. Beyond that, I have some knee-jerk reactions that probably don’t mean much.

The 4clojure.com problems are great, and I will work on more today. However I’m already feeling the urge to pick some small “real” project, and work on that instead of (or at least in addition to) problem sets. The trick will be to pick something that’s not too big or small to tackle.

  1. Also, C-c C-x “refreshes” REPL with code from current buffer. At the moment this feels like a distinction without a difference. 

  2. To clarify, the neat thing about a test submodule in Racket is that it exists at “test time”. In other words, the submodule can live in the same source file, and access its surrounding parent module. But it is not loaded or evaluated at run time — just at “test time”. Likewise you can use submodules for “documentation time” or other concepts.