OO 第三单元总结
OO 第三单元总结
规格的阅读与实现心得
JML的阅读方法
-
语法上, 可以参考课程组的《JML level0手册》, 涵盖了基本的jml关键词和语法,看不明白的话可以多翻翻,类比着就搞懂了
-
阅读顺序上, 阅读JML可以从一些比较底层的类开始读, 比如
Person
,Message
这种依赖关系比较少的类, 可以先读起来,这样对一个类的功能就可以很快的了解 -
阅读tips :
Idea对于JML高亮支持其实很不好(基本没有啊喂)推荐在vscode上装好插件在上面阅读,
还可以对Idea注释调颜色,这样至少看起来清楚一点不会灰糊糊的一团
JML迭代
每次迭代推荐用Idea 鼠标右键的compare with clipboard, 比较一下差异, 同时注意一下一些坑人限制条件比如1111
测试数据与JML
本单元一开始想试试单元测试的,但咨询了一下学长们,似乎都不太推荐JUNIT, OPENJML等基于JML的测试, 后来经过尝试确实有点麻烦。OpenJml 就是个jml语法检查器(bushi, Junit有点像Verilog的testbench样例还需要自己构造,如果能够有一个可以根据JML直接校验函数的工具就好了
剩下的日子里, 就都在写对拍机了,在构造测试数据时, JML的存在也可以使测试数据的覆盖率提高, 就比如am指令的构造
/*@ public normal_behavior
@ requires !(\exists int i; 0 <= i && i < messages.length; messages[i].equals(message)) &&
@ (message instanceof EmojiMessage) ==> containsEmojiId(((EmojiMessage) message).getEmojiId()) &&
@ (message.getType() == 0) ==> (message.getPerson1() != message.getPerson2());
@ assignable messages;
@ ensures messages.length == \old(messages.length) + 1;
@ ensures (\forall int i; 0 <= i && i < \old(messages.length);
@ (\exists int j; 0 <= j && j < messages.length; messages[j].equals(\old(messages[i]))));
@ ensures (\exists int i; 0 <= i && i < messages.length; messages[i].equals(message));
@ also
@ public exceptional_behavior
@ signals (EqualMessageIdException e) (\exists int i; 0 <= i && i < messages.length;
@ messages[i].equals(message));
@ signals (EmojiIdNotFoundException e) !(\exists int i; 0 <= i && i < messages.length;
@ messages[i].equals(message)) &&
@ (message instanceof EmojiMessage) &&
@ !containsEmojiId(((EmojiMessage) message).getEmojiId());
@ signals (EqualPersonIdException e) !(\exists int i; 0 <= i && i < messages.length;
@ messages[i].equals(message)) &&
@ ((message instanceof EmojiMessage) ==>
@ containsEmojiId(((EmojiMessage) message).getEmojiId())) &&
@ message.getType() == 0 && message.getPerson1() == message.getPerson2();
@*/
## 依据JML构造数据来覆盖所有可能分支 ##
def am():
ty = random.randint(0, 1)
sv = random.choice(extreme_sv_pool)
if random.randint(0, excRate) == 1: # emi
try:
id = random.choice(list(message_pool.keys()))
except:
id = random.randint(idmin, idmax)
if ty == 0:
try: # {id:{type:, pid1:, pid2:, sv:, gid:,}}
if random.randint(0, excRate) == 1:
# 在sm时会触发异常:
# epi & pinf情况, 异常可控
if random.randint(0, 2) == 1:
pid1 = pid2 = random.choice(list(person_pool.keys())) # epi
else:
pid1 = pid2 = random.randint(idmin, idmax) # pinf when send
elif random.randint(0, 2 * excRate) == 1:
pid1 = random.randint(idmin, idmax)
pid2 = random.randint(idmin, idmax)
else:
pid1, pid2 = random.sample(list(person_pool.keys()), 2)
# ADD! 真正加入, 用一点数据结构记录
message_pool[id] = {"type": ty, "sv": sv, "pid1": pid1, "pid2": pid2, "gid": None, "div": "am"}
except:
pid1 = random.randint(idmin, idmax)
pid2 = random.randint(idmin, idmax)
return "am {} {} {} {} {}".format(id, sv, ty, pid1, pid2)
else:
# pinf ginf 情况 覆盖了所有异常,进行测试
if random.randint(0, excRate) == 1:
pid1 = random.randint(idmin, idmax)
gid = random.randint(idmin, idmax)
else:
pid1 = random.choice(list(person_pool.keys()))
try:
gid = random.choice(list(group_pool.keys()))
except:
return ag()
# ADD! 真正的加入
message_pool[id] = {"type": ty, "sv": sv, "pid1": pid1, "pid2": None, "gid": gid, "div": "am"}
return "am {} {} {} {} {}".format(id, sv, ty, pid1, gid)
else:
id = random.randint(idmin, idmax)
if ty == 0:
try:
if random.randint(0, excRate) == 1: # epi
if random.randint(0, 2) == 1:
pid1 = pid2 = random.choice(list(person_pool.keys()))
else:
pid1 = pid2 = random.randint(idmin, idmax)
elif random.randint(0, 2 * excRate) == 1:
pid1 = random.randint(idmin, idmax)
pid2 = random.randint(idmin, idmax)
else:
pid1, pid2 = random.sample(list(person_pool.keys()), 2)
except:
pid1 = random.randint(idmin, idmax)
pid2 = random.randint(idmin, idmax)
message_pool[id] = {"type": ty, "sv": sv, "pid1": pid1, "pid2": pid2, "gid": None, "div": "am"}
return "am {} {} {} {} {}".format(id, sv, ty, pid1, pid2)
else:
pid1 = random.choice(list(person_pool.keys()))
try:
gid = random.choice(list(group_pool.keys()))
message_pool[id] = {"type": ty, "sv": sv, "pid1": pid1, "pid2": None, "gid": gid, "div": "am"}
return "am {} {} {} {} {}".format(id, sv, ty, pid1, gid)
except:
return ag()
容器的选择
一些容器和我的选择, 因为一些容器需要O(1)的查询效率, 那么Hashmap
是最好的选择
第三次作业中, Message
的存储满足的是一个FIFO的规则, 所以用了linkedlist
Exception
private int id; //personId触发了异常
private static int totalExcCnt = 0;
private static HashMap<Integer, Integer> excTable = new HashMap<>();
// 触发的personId - 触发的次数
Group
private HashMap<Integer, Person> people = new HashMap<>();
// id - person
Person
private HashMap<Person, Integer> linkage = new HashMap<>();
// 邻居Person - distance
private LinkedList<Message> messages = new LinkedList<>();
// Person收到的信息
Network
private HashMap<Integer, Person> people = new HashMap<>();
private HashMap<Integer, Group> groups = new HashMap<>();
private HashMap<Integer, Message> messages = new HashMap<>();
private HashMap<Integer, Integer> emojiHeatMap = new HashMap<>();
// 上述皆为<id, 内容> 的Hashmap, O(1)查找不错的
private HashSet<Edge> edges = new HashSet<>();
// Edge是我自己定义的边类, edges记录了所有的边, kruskal的时候会比较方便
private MyDsu<Person> dsu = new MyDsu<>();
// Dsu是我自己写的优化完全的并查集类, 支持不同模板很好用
算法和性能分析
本单元有一些需要注意算法时间复杂度的点
-
维护
valueSum
,ageSum
,ageSquareSum
如果要求
ageVar
那么为了保证精度与jml一致, 自己推一下就好return (timeSum - 2 * ageSum * getAgeMean() + getAgeMean() * getAgeMean() * people.size()) / people.size(); //////////////////////////////////////////////////////////////////// BigInteger part1 = ageSquareSum; BigInteger part2 = (new BigInteger("2")).multiply(ageSum) .multiply(BigInteger.valueOf(getAgeMean())); BigInteger part3 = BigInteger.valueOf(getAgeMean()).multiply(BigInteger.valueOf(getAgeMean())) .multiply(BigInteger.valueOf(people.size())); BigInteger result = (part1.subtract(part2).add(part3)).divide(BigInteger.valueOf(people.size())); return result.intValue();
需要注意的是:
valueSum
是将每对关系计算两次,需要在每次加人时valueSum
加2 * value
。除了在加人删人外,在添加关系时需要遍历所有group
维护valueSum
。(别忘了哦!) -
Kruskal算法, 需要提前维护所有的边集, 然后遍历边集找到所有在联通分量的边。这样可以实现O(ElogE)的复杂度, 如果不存储需要用iscircle再查找可能会慢
-
isCircle方法, 要用路径压缩+合并秩优化的并查集, 鉴于网上似乎还没有java版的,我给一个吧嘿嘿
import java.util.HashMap; public class MyDsu<T> { private HashMap<T, T> fatherMap = new HashMap<>(); private HashMap<T, Integer> rankMap = new HashMap<>(); public MyDsu() { } public void addElement(T a) { fatherMap.put(a, a); rankMap.put(a, 1); } public T findRoot(T a) { T r = a; while (fatherMap.get(r) != r) { r = fatherMap.get(r); } T i = a; T tmp; while (i != r) { tmp = fatherMap.get(i); fatherMap.put(i, r); i = tmp; } return r; } public void unionSet(T a, T b) { if (a == null || b == null) { return; } T root1 = findRoot(a); T root2 = findRoot(b); if (rankMap.get(root1) == rankMap.get(root2)) { fatherMap.put(root2, root1); //root 1 is fa rankMap.put(root1, rankMap.get(root1) + 1); } else { if (rankMap.get(root1) < rankMap.get(root2)) { fatherMap.put(root1, root2); } else { fatherMap.put(root2, root1); } } } }
-
DeleteColdEmoji
如果是按着jml的逻辑, 先遍历
heatmap
找到对应待删除的emojiMessage
, 那么针对每一个emoji,都要遍历一遍整个messages
显然复杂度较大。正常想法就是先遍历messages
再遍历heatmap
, 那么就可以O(n)解决问题 -
dijkstra算法
注意可以使用堆优化, 并使用一个内部类
node
来记录相关节点信息public class Dijkstra { public class Node implements Comparable<Node> { private int dis = 0; private int id = 0; public int getDis() { return dis; } public int getId() { return id; } public Node(int id, int dis) { this.dis = dis; this.id = id; } @Override public int compareTo(Node o) { return Integer.compare(this.getDis(), o.getDis()); } } // <id, dis> private HashMap<Integer, Integer> dis = new HashMap<>(); // <id, done> private HashMap<Integer, Boolean> done = new HashMap<>(); // <id, person> private HashMap<Integer, Person> people; private PriorityQueue<Node> queue = new PriorityQueue<>(); public Dijkstra(HashMap<Integer, Person> people) { this.people = people; for (Map.Entry<Integer, Person> personEntry : people.entrySet()) { int id = personEntry.getKey(); dis.put(id, (int) 1e9); done.put(id, false); } } public int run(int fromId, int toId) { //算法本体 } }
Network 扩展考虑
题目要求
假设出现了几种不同的Person
- Advertiser:持续向外发送产品广告
- Producer:产品生产商,通过Advertiser来销售产品
- Customer:消费者,会关注广告并选择和自己偏好匹配的产品来购买 -- 所谓购买,就是直接通过Advertiser给相应Producer发一个购买消息
- Person:吃瓜群众,不发广告,不买东西,不卖东西
如此Network可以支持市场营销,并能查询某种商品的销售额和销售路径等 请讨论如何对Network扩展,给出相关接口方法,并选择3个核心业务功能的接口方法撰写JML规格(借鉴所总结的JML规格模式)
首先我设计:
- 在Message下新增加
Advertisement
类, 其中包含了attraction
代表广告吸引力,其余属性类似于message分群发、私发等 - 新增加Product类用于买卖,宣传, 内置属性
storage
,salesAmount
- productList[]保存着当前的product
- 每个person(作为吃瓜群众,也会收到广告) 其中可以通过getInterest(Product p) 来获取对p的a兴趣
/*@ public normal_behavior
@ requires containsMessage(id) && getMessage(id).getType() == 0 &&
@ getMessage(id).getPerson1().isLinked(getMessage(id).getPerson2()) &&
@ getMessage(id).getPerson1() != getMessage(id).getPerson2();
@ assignable messages, emojiHeatList; +
@ assignable getMessage(id).getPerson1().socialValue, getMessage(id).getPerson1().money;
@ assignable getMessage(id).getPerson2().messages, getMessage(id).getPerson2().socialValue, getMessage(id).getPerson2().money,
getMessage(id).getPerson2().interests;
@ ensures (\old(getMessage(id)) instance of Advertisement) ==>
@ (\old(getMessage(id)).getPerson1().getInterest(Advertisement.getProduct()) ==
@ \old(getMessage(id).getPerson1().getInterest(Advertisement.getProduct())) + ((Advertisement)\old(getMessage(id))).getAttraction()
@ ensures (!(\old(getMessage(id)) instanceof Advertisement)) ==> (\not_assigned(people[*].getInterest(Product[*]));
......略
@
@ also
@ public normal_behavior
@ requires containsMessage(id) && getMessage(id).getType() == 1 &&
@ getMessage(id).getGroup().hasPerson(getMessage(id).getPerson1());
@ assignable people[*].socialValue, people[*].money, people[*].interests
messages, emojiHeatList;
......略
@ ensures (\old(getMessage(id)) instance of Advertisement) ==>
@ (\exists int i; i == ((Advertisement)\old(getMessage(id))).getAttraction();
@ (\forall Person p; \old(getMessage(id)).getGroup().hasPerson(p) && p != \old(getMessage(id)).getPerson1();
@ p.getInterest(Advertisement.getProduct()) == \old(p.getInterest(Advertisement.getProduct())) + i));
@ ensures (!(\old(getMessage(id)) instanceof Advertisement)) ==> (\not_assigned(people[*].getInterest(Product[*]));
......略
@ also
@ public exceptional_behavior
@ signals (MessageIdNotFoundException e) !containsMessage(id);
@ signals (RelationNotFoundException e) containsMessage(id) && getMessage(id).getType() == 0 &&
@ !(getMessage(id).getPerson1().isLinked(getMessage(id).getPerson2()));
@ signals (PersonIdNotFoundException e) containsMessage(id) && getMessage(id).getType() == 1 &&
@ !(getMessage(id).getGroup().hasPerson(getMessage(id).getPerson1()));
@*/
public void sendMessage(int id) throws
RelationNotFoundException, MessageIdNotFoundException, PersonIdNotFoundException;
// Customer
/*@ public normal_behavior
@ requires contains(producerId);
@ requires !(\exists int i; 0 <= i && i < productList.length; productList[i] == id);
@ ensures (\exists int i; 0 <= i && i < productList.length; productList[i] == product);
@ ensures productList.length == \old(productList.length) + 1;
@ also
@ public normal_behavior
@ requires contains(producerId);
@ requires (\exists int i; 0 <= i && i < productList.length; productList[i] == id);
@ ensures (\forall int i; 0 <= i && i < productList.length; productList[i] == \old(productList[i]));
@ ensures productList.length == \old(productList.length);
@ ensures (\exists int i; 0 <= i && i < productList.length; productList[i] == id
&& productList[i].storage == \old(productList[i].storage) + 1);
@*/
public void produceProduct(Product product, int producerId);
/*@ public normal_behavior
@ requires containsProductId(productId);
@ ensures \result == productList(id).getSalesAmount();
@ also
@ public exceptional_behavior
@ signals (ProductNotFoundException e) !containsProduct(productId);
@*/
public int querySaleAmount(int productId) throws ProductNotFoundException;
架构分析
本单元作业架构已经由课程组给出,无需多分析,重要的是阅读jml和采取合适的数据结构和算法来实现。
体会感想
整个单元其实期待还是很高的。但是整个单元学下来, 似乎所有的时间都用在读jml上了, 图论算法也不算难,感谢助教的怜悯,其实希望有个jml直接对接检查代码的说,结果似乎没有,感觉jml的威力瞬间减少了不少。 三次作业难得全满挺不错的,希望下次继续努力!