Functional programming leads to deep insights into the nature of computation. -- Martin Odersky

形式化

FizzBuzzWhizz详细描述请自行查阅相关资料。此处以3, 5, 7为例,形式化地描述一下问题。

r1
- times(3) -> Fizz
- times(5) -> Buzz
- times(7) -> Whizz
r2
- times(3) && times(5) && times(7) -> FizzBuzzWhizz
- times(3) && times(5) -> FizzBuzz
- times(3) && times(7) -> FizzWhizz
- times(5) && times(7) -> BuzzWhizz
r3
- contains(3) -> Fizz
- the priority of contains(3) is highest
rd
- others -> others

接下来我将使用Scala尝试FizzBuzzWhizz问题的设计和实现。

语义模型

从上面的形式化描述,可以很容易地得到FizzBuzzWhizz问题的语义模型。

Rule: (Int) -> String
Matcher: (Int) -> Boolean
Action: (Int) -> String

其中,Rule存在三种基本的类型:

Rule ::= atom | allof | anyof

三者之间构成了「树型」结构。

atom: (Matcher, Action) -> String
allof(rule1, rule2, ...): rule1 && rule2 && ... 
anyof(rule1, rule2, ...): rule1 || rule2 || ... 

测试用例

借助Java8增强了的「函数式」能力,可抛弃掉很多重复的「样板代码」,使得设计更加简单、漂亮。此外,Java8构造DSL的能力也相当值得称赞,非常直接,简单。测试用例此处选择Spock(基于Groovy语言),可提高用例的可读性和可维护性。

class RuleSpec extends Specification {
  private static def spec() {
    Rule r1_3 = atom(times(3), to("Fizz"))
    Rule r1_5 = atom(times(5), to("Buzz"))
    Rule r1_7 = atom(times(7), to("Whizz"))

    Rule r1 = anyof(r1_3, r1_5, r1_7)

    Rule r2 = anyof(allof(r1_3, r1_5, r1_7),
        allof(r1_3, r1_5),
        allof(r1_3, r1_7),
        allof(r1_5, r1_7))

    Rule r3 = atom(contains(3), to("Fizz"))
    Rule rd = atom(always(true), nop())

    anyof(r3, r2, r1, rd)
  }

  def "fizz buzz whizz"() {
    expect:
    spec().apply(n) == expect

    where:
    n            | expect
    3            | "Fizz"
    5            | "Buzz"
    7            | "Whizz"
    3 * 5 * 7    | "FizzBuzzWhizz"
    3 * 5        | "FizzBuzz"
    3 * 7        | "FizzWhizz"
    (5 * 7) * 2  | "BuzzWhizz"
    13           | "Fizz"
    35 /* 5*7 */ | "Fizz"  /* not "BuzzWhizz" */
    2            | "2"
  }
}

匹配器:Matcher

Matcher是一个「一元函数」,入参为Int,返回值为Boolean,是一种典型的「谓词」。从OO的角度看,always是一种典型的Null Object

import static java.lang.String.valueOf;

@FunctionalInterface
public interface Matcher {
  boolean matches(int n);

  static Matcher times(int n) {
    return x -> x % n == 0;
  }

  static Matcher contains(int n) {
    return x -> valueOf(x).contains(valueOf(n));
  }

  static Matcher always(boolean bool) {
    return n -> bool;
  }
}

执行器:Action

Action也是一个「一元函数」,入参为Int,返回值为String,其本质就是定制常见的map操作,将定义域映射到值域。

@FunctionalInterface
public interface Action {
  String to(int n);

  static Action to(String str) {
    return n -> str;
  }

  static Action nop() {
    return n -> String.valueOf(n);
  }
}

规则:Rule

Composition Everywhere

RuleFizzBuzzWhizz最核心的抽象,也是设计的灵魂所在。从语义上Rule分为2种基本类型,并且两者之间形成了优美的、隐式的「树型」结构,体现了「组合式设计」的强大威力。

    • Atom

    • Compositions: anyof, allof

    Rule是一个「二元函数」,入参为(Int),返回值为String

    @FunctionalInterface
    public interface Rule {
      String apply(int n);
    }

    Rules为一个工厂类,用于生产各种Rule。其中,allof, anyof使用了Java8 StreamAPI

    public final class Rules {
      public static Rule atom(Matcher matcher, Action action) {
        return n -> matcher.matches(n) ? action.to(n) : "";
      }
    
      public static Rule anyof(Rule... rules) {
        return n ->sstream(n, rules)
            .filter(s -> !s.isEmpty())
            .findFirst()
            .orElse("");
      }
    
      public static Rule allof(Rule... rules) {
        return n -> sstream(n, rules)
            .collect(joining());
      }
    
      private static Stream<String> sstream(int n, Rule[] rules) {
        return Arrays.stream(rules)
            .map(r -> r.apply(n));
      }
    
      private Rules() {
      }
    }

    源代码


    horance
    255 声望29 粉丝

    刘光聪,程序员,敏捷教练,开源软件爱好者,具有多年大型遗留系统的重构经验,对OO,FP,DSL等领域具有浓厚的兴趣。