Roman numerals kata in Clojure, discovering the algorithm by TDD. IIRC the original description mentions the argument would not be greater than 3000 (as in a year A.D.), this is not checked. Arguments > 9999 will not be handled correctly.

roman.clj

(ns tonvanbart.roman
  (:use clojure.test)
  (:require [clojure.string :as str]))

(defn convert-step
  "one step converting a number to roman numerals"
  [n unitsymbol fivesymbol tensymbol]
  (cond
    (< n 4) (apply str (repeat n unitsymbol))
    (< n 9) (str/join
              (concat
                  (repeat (- 5 n) unitsymbol)
                  fivesymbol
                  (repeat (- n 5) unitsymbol)))
    (= n 9) (str unitsymbol tensymbol)))

(defn convert
  "convert a number to roman numerals"
  ([n] (convert n ""))
  ([n result]
      (cond
        (> n 1000) (let [part (convert-step (quot n 1000) "m" "" "")]
                     (recur (rem n 1000) part))
        (> n 100) (let [part (str result (convert-step (quot n 100) "c" "d" "m"))]
                    (recur (rem n 100) part))
        (> n 10)  (let [part (str result (convert-step (quot n 10) "x" "l" "c"))]
                    (recur (rem n 10) part))
        :else (str result (convert-step n "i" "v" "x")))))

; tests

(deftest conversions
  (are [number converted] (= converted (convert number))
    1 "i"
    2 "ii"
    3 "iii"
    4 "iv"
    8 "viii"
    9 "ix"
    12 "xii"
    19 "xix"
    39 "xxxix"
    49 "xlix"
    208 "ccviii"
    1958 "mcmlviii"
    2008 "mmviii"
    2112 "mmcxii"))

(run-tests)