Thinking in Clojure 26-Jul-16 Jumping in We’ll quickly go through Clojure’s data types, some basic functions, and basic syntax Then we’ll get to the good stuff! 2 Clojure’s data types Clojure has: Lists, enclosed in parentheses and separated by spaces or commas: (a 17 "Plenty of parentheses") Functions: (fn [x] (first(rest x)) Numbers: All Java numeric types, plus ratios and exact decimals: 5, 5.3, 5.3e30, 077, 0xFF00FF, 3/5, 5.3M Strings, as in Java: "She said \"Hello\"" Characters: \a, \5, \n, \newline, \tab, etc. The booleans true and false nil, equivalent to Java’s null Symbols, which stand for themselves: :meadow, :CIS-554 Vectors: [5 :a "hi!"] Maps: {:one 1, :two 2} Sets: #{:prolog :clojure} 3 Some basic Clojure functions Syntax of a function call: (function args) Basic operations—sequences (seq) are lists, sets, maps, vectors: (quote arg) or 'arg, to keep arg from being evaluated Basic arithmetic (first seq) is the first element in the sequence (or nil) (rest seq) is what remains when the first element is removed (or nil) (cons arg seq) returns a new sequence with arg as its first element (= args) tests whether its args are equal (empty? seq), (list? seq), (seq? arg), (nil? arg) are more tests (+ args), (- args), (* args), (/ args) , (< args), etc. Basic logic (these aren’t actually functions, but special forms) (and args), (or args), (not arg), (if condition result1 result2) 4 Defining values (def name docstring? value) defines the name to be the given value The doc function will show the documentation string The docstring is an optional (hence the ?) documentation string user=> (def pi "something about a circle" 3.1416) #'user/pi user=> (doc pi) ------------------------user/pi something about a circle nil In the REPL, (find-doc string) will find and display functions whose docstring contains the given string 5 Defining functions Functions are values; (fn [x] (* 3 x)) is a function that will triple its parameter Functions, like other values, can be given names user=> ((fn [x] (* 3 x)) 6) 18 user=> (def triple "Multiply by 3" (fn [x] (* 3 x))) #'user/triple user=> (triple 6) 18 (defn name docstring? [arg1 arg2 … argN] value) is shorthand for (def name docstring? (fn [arg1 arg2 … argN]value)), user=> (defn cube "Raise to the power 3" [x] (* x x x)) #'user/cube user=> (cube 3) 27 6 Functions and special forms The arguments to a function are evaluated before the function is called A special form looks just like a function, but it gets its arguments unevaluated Example: (* 2 (+ 3 4)) The function * is called with the arguments 2 and 7 The special form itself decides when and whether to evaluate its arguments quote does not evaluate its argument if evaluates its first argument, then decides which of the second and third arguments to evaluate Clojure allows you to define your own special forms This means you can define your own control structures 7 A typical Clojure function (defn first-double-letter "Returns the first doubled letter in a string, or nil." [s] (if (< (count s) 2) nil (if (= (first s) (second s)) (first s) (first-double-letter (rest s)) ) ) ) user=> (first-double-letter "Pennsylvania") \n user=> (first-double-letter '(1 2 3 4 3 5 5 4 6)) 5 8 It’s easier with cond cond is an if … then … else if … then … else … construct: (cond test1 result1 test2 result2 … testN resultN) It requires an even number of parameters (one result for each test) The symbol :else may be used as the last test (defn first-double-letter "Returns the first doubled letter in a string, or nil." [s] (cond (< (count s) 2) nil (= (first s) (second s)) (first s) :else (first-double-letter (rest s)) ) ) user=> (first-double-letter "Pennsylvania") \n user=> (first-double-letter '(1 2 3 4 3 5 5 4 6)) 5 9 It’s all about recursion Some rules of doing recursion: 1. 2. 3. 4. Handle the base cases directly (without recursion) Recur only with a simpler case Don’t use global variables Don’t “look down” into the recursion—that will just confuse you In Clojure you are almost always working with a list or some similar sequence Lisp programmers say, “Do something with the head, and recur with the tail” Clojure doesn’t have global variables Clojure’s terms for “head” and “tail” are “first” and “rest” This pretty much covers rules 1 and 2 above This covers rule #3 above Rule 4 always holds. Think about what you are doing now, not what some recursive call is doing 10 Functional programming Clojure is functional—what does that mean? Functions are like functions in math—called with the same arguments, they always return the same result Functions are values, or first-class objects This means: No “global variables,” no dependence on external values, and no side effects! Functions can be passed as parameters to functions, returned as the value of functions, created as needed, stored in data structures, and there are operations on functions that produce new functions The “blub paradox” applies—the value added is substantial, but not obvious to an imperative or object-oriented programmer Data is immutable (like strings in Java) Clojure’s data structures are designed to make this efficient Immutable data greatly simplifies concurrent programming Because data is immutable, loops are unnecessary (use recursion instead!) 11 Costs and benefits Costs of functional programming It’s weird and unfamiliar How can you do anything without objects, mutable variables, or loops? As a manager, functional programmers are hard to find (and expensive!) Clojure, and Lisp dialects generally, have too many parentheses! Benefits of functional programs Easier to write correct programs (Loops are used primarily to change the values of things) “Yeah, right!” – “No, really! All data is local and immutable.” Easier to write unit tests, because function values depend only on inputs Much easier to write concurrent programs Operations on collections make code simpler and more concise The simpler foundation means less syntax and fewer special cases Some operations, such as equality testing, are really fast But it’s still weird! 12 Easier to write correct programs Programs are easier to write when all data is local Unit testing is easier, because there are no dependencies on functions that may or may not have been called previously When relevant values can be changed elsewhere in the program—possibly in many places—it’s harder to see all the connections Functions in a functional language get all relevant input from the parameter list There is no need for a setUp method Functional programming supports powerful operations on sequences The imperative and object-oriented programming styles have been characterized as “word by word” programming Some sequence operations, such as membership testing, are provided for you In a functional language, any function that can apply to an element of a sequence can be applied to the entire sequence 13 The problem of state Nonfunctional programming language are “stateful” or “have state” The state of a program is given by (1) the values of all the variables throughout the program, and (2) the current locus of execution That can be a huge amount of information to keep track of! Object-oriented programmers try to control complexity by having objects be responsible for their own state, and “loosely coupled” (not very dependent on) other objects Methods often have “side effects,” that is, they modify state Functional languages try to avoid having state at all This isn’t always easy Purely functional languages cannot have side effects Since I/O is a side effect, this is an even more difficult restriction 14 Maintaining state, functionally Sometimes you just need state Consider an adventure game You need to keep track of where you are, where other objects are, what you are holding, which paths are blocked or open, etc.—this is your state You do not need to keep track of permanent, immutable data; for example, most paths between rooms are fixed and unchanging—this isn’t part of the state In a functional program, a “state” is just an immutable data item The data item can be quite complex, such as a dictionary States are immutable, but you can always create a “new” state that is a variation of a given state With carefully designed data structures, not as much storage is required as you might expect So the functional solution to maintaining state is: Pass one state into a function, get a new (and different) state back! 15 Clojure’s I/O compromise A purely functional program has no side effects In Clojure, all functions return a value (print args) and (println args) return nil Clojure allows side effects in two well-defined places: I/O is a side effect Therefore: A purely functional program cannot do I/O! (do args) evaluates all its arguments in order, but returns only the value of the last one When a function (fn argv args) is called, the arguments are evaluated in order, but only the last value is returned Example: (defn powers "Computes cube and square" [] (def n (read)) (println (* n n n)) (println (* n n)) n) 16 Lists are immutable Here is a typical list: my-list A B C w Here is (cons 'w my-list) Here is (rest my-list) Notice that my-list remains unchanged Vectors, hash maps, sorted maps, hash sets, and sorted sets are similarly immutable 17 Functions are just values user=> (cons 'w '(a b c)) (w a b c) user=> (defn swap-args [f x y] (f y x)) #'user/swap-args user=> (swap-args cons '(a b c) 'w) (w a b c) user=> (defn apply-n-times [f x n] "Apply f to x, n times: f(f(f..(n)...))" (if (zero? n) x (apply-n-times f (f x) (dec n)) ) ) #'user/apply-n-times user=> (apply-n-times (fn [x] (* 2 x)) 1 10) 1024 18 Collatz, the hard way Definition: collatz(1) = 1 collatz(n) = collatz(n / 2) if n is even collatz(n) = collatz(3 * n + 1) user=> (defn collatz [n] (let [ do-even (fn [n] (collatz (/ n 2))) do-odd (fn [n] (collatz (inc (* 3 n)))) ] (print n " ") (if (= n 1) 1 (if (even? n) (do-even n) (do-odd n)) ) ) ) #'user/collatz user=> (collatz 7) 7 22 11 34 17 52 26 13 40 20 10 5 16 8 4 2 1 1 19 map, filter, and reduce Here are three powerful functions you will find in almost any functional programming language map – apply a function to every element of a sequence, returning a sequence of results filter – apply a predicate to every element of a sequence, returning a sequence of those that satisfy the predicate user=> (map even? '(3 1 4 1 6)) (false false true false true) user=> (filter even? '(3 1 4 1 6)) (4 6) reduce – use a function to reduce a sequence to a single value user=> (reduce * '(3 1 4 1 6)) 72 20 The real problem with state For decades we’ve been dealing with mutable state Mutable state + concurrency = nondeterminism We use threads and locks and semaphores and so on As Herb Sutter points out in The Free Lunch is Over, we have hit a 3 GHz barrier Since 2003, computers have not gotten faster We still want them faster Concurrency is the only solution Functional languages, with immutable state, provide a partial solution As Martin Odersky points out, you can hide from concurrency for a while yet…but not forever Important consequences: These are complicated, unsafe, and inefficient All newer languages are gaining functional and concurrent features Older languages, such as Java, are also trying to integrate these features “You can run, but you can’t hide!” 21 Oh, and by the way… Clojure has Infinite sequences Lazy sequences Exact decimal arithmetic Function composition Function currying Macros Actors and Software Transactional Memory Access to all the Java methods And lots more No version of Lisp has ever become mainstream Lisp just gets mugged in dark alleys and its ideas stolen! 22 The End 23