[Erlang 0117] 当我们谈论Erlang Maps时,我们谈论什么 Part 2

 

  声明:本文讨论的Erlang Maps是基于17.0-rc2,时间2014-3-4.后续Maps可能会出现语法或函数API上的有所调整,特此说明.

 

     

      前情提要: [Erlang 0116] 当我们谈论Erlang Maps时,我们谈论什么 Part 1 

      继续昨天的话题,在Erlang Factory SF Bay Area 2013有一个议题:"Where are we on the Map?" [PDF ],这个Talk基本上就是选取了EEP43的要点,有兴趣的同学可以FQ观看视频 Where are We on the Map? - Kenneth Lundin - YouTube 如果是腿脚不利索的,可以看墙内的.仔细阅读EEP43,其信息量巨大,包括Maps的设计演变来龙去脉,各种取舍,也是我们学习设计的极佳范例.下面我将按照自己的逻辑顺序重新解读EEP43,先从如何使用开始,直观上感受一下区别,然后再回答"何必有我"的问题.

 

Maps Basic

 

      EEP43 给出了Map比较规范的定义, Map M包含一定数量的键值对,实现从K1..Kn到V1..Vn的映射,其中没有两个Key是相等的(equal). equal指的是K1==K2,matching指的是K1 =:= K2. erlang:is_map(M)用于判断数据是否map类型.不过按照现在的情况,当出现1.0和1做key的时候,结果和EEP43中设计的结果不同,还是要看下一个版本是怎么处理的,这个不小心就是个坑:

 

Eshell V6.0  (abort with ^G)
1> M=#{1=>a}.   %% Construction syntax
#{1 => a}
2> M#{1.0 => b}.
#{1 => a,1.0 => b}
3> M#{1 => b}.
#{1 => b}
4> M#{1 := b}.
#{1 => b}
5> M#{1.0 := b}.
** exception error: bad argument
     in function  maps:update/3
        called as maps:update(1.0,b,#{1 => a})
     in call from erl_eval:'-expr/5-fun-0-'/2 (erl_eval.erl, line 249)
     in call from lists:foldl/3 (lists.erl, line 1261)
6> M2= #{1=>a,1=>b,1.0 =>c}.
#{1 => b,1.0 => c}
7> 1 == 1.0.
true
8> #{1.0 =>a ,1 =>b}.
#{1 => b,1.0 => a}
 

　　

     构造Map的时候我们重点要验证的就是"Maps in Erlang are ordered, Important!!!! – Maps with the same set of keys are always presented in the same way":

 

 

Eshell V6.0  (abort with ^G)
1> #{a=>124,b=>1024,c=>23}.
#{a => 124,b => 1024,c => 23}
2> #{b=>1024,c=>23,a=>124}.
#{a => 124,b => 1024,c => 23}
3> #{b=>1024,a=>124,c=>23}.
#{a => 124,b => 1024,c => 23}
4> M=#{b=>1024,a=>124,c=>23}.
#{a => 124,b => 1024,c => 23}

　　

 

看下Map的基本操作,构造,更新,模式匹配.注意下面代码中 #{f:=F,a:={A,B}} =  M.做匹配的时候,前面的部分key是顺序无关的.

 

Eshell V6.0  (abort with ^G)
1> M=#{a=>{1,2},b=>23,<<"OK">> =>ok, f=>fun()->receive a ->"Got a!" end end}.
#{a => {1,2},b => 23,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok}
2> M#{b=>1024}.
#{a => {1,2},b => 1024,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok}
3> M#{b2=>1023}.
#{a => {1,2},b => 23,b2 => 1023,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok}
4> M#{b2 :=1024}.
** exception error: bad argument
     in function  maps:update/3
        called as maps:update(b2,1024,
                              #{a => {1,2},b => 23,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok})
     in call from erl_eval:'-expr/5-fun-0-'/2 (erl_eval.erl, line 249)
     in call from lists:foldl/3 (lists.erl, line 1261)
5> M#{b :=1024}.
#{a => {1,2},b => 1024,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok}
6>  #{f:=F,a:={A,B}} =  M.
#{a => {1,2},b => 23,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok}
7> b().
A = 1
B = 2
F = fun() ->
           receive
               a ->
                   "Got a!"
           end
    end
M = #{a => {1,2},b => 23,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok}
ok
8> 

 

 

2> M=#{a=>123,b=>234}.
#{a => 123,b => 234}
3> M#{a=>1000}.
#{a => 1000,b => 234}
4> M#{not_key=>1000}.
#{a => 123,b => 234,not_key => 1000}
5> M#{a :=1024}.
#{a => 1024,b => 234}
6> M#{not_key :=1024}.
** exception error: bad argument
     in function  maps:update/3
        called as maps:update(not_key,1024,#{a => 123,b => 234})
     in call from erl_eval:'-expr/5-fun-0-'/2 (erl_eval.erl, line 249)
     in call from lists:foldl/3 (lists.erl, line 1261)
7> P=#{name=>"zen",id=>2002}.
#{id => 2002,name => "zen"}
8> P#{naem=>1024}.
#{id => 2002,naem => 1024,name => "zen"}

　　 

  上面更新字段名称的方式,是用=> 还是:= ?从工程层面考虑,我会选择 := 运算,因为当给一个不存在的key进行赋值的时候,会抛出错误bad argument.而上面第7~8行代码,本意是更新name字段,但是由于拼写错误原来的字段值没有修改,只增加了一错误字段.在工程代码中显然:=会大大降低排错的成本.

 

 然后就是常用的 Function Header Match,这里可以看到新增了两个Guard  erlang:is_map erlang:map_size

 

-module(a).
-compile(export_all).
test(#{state := {ok,State}} =S) ->
  State.
 
Eshell V6.0  (abort with ^G)
1> a:test().
** exception error: undefined function a:test/0
2> a:test(#{state => {ok,200}}).
200
3> a:test(#{state => {error,200}}).
** exception error: no function clause matching a:test(#{state => {error,200}}) (a.erl, line 7)
4> 
 

 


-module(a).
-compile(export_all).

a(#{state := {ok,State}} = S) when erlang:is_map(S) ->
  State.

b(#{state := {ok,State}} = S) when erlang:map_size(S) >=1 ->
  State.

test(#{state := {ok,State}} =S) when erlang:is_map(S) andalso erlang:map_size(S) >=1 ->
  State.

Eshell V6.0  (abort with ^G)
1> a:b(#{state => {ok,200},id=>123,dd=>11}).
200
2> a:a(#{state => {ok,200},id=>123,dd=>11}).
200
3> a:test(#{state => {ok,200},id=>123,dd=>11}).
200
4>
4> 

　　 

maps模块方面不知道变动还会有多少,就目前提供的函数和EEP中的描述还是有不少细微差别的,比如foldr foldl全部整合为fold函数;这个代码太长了,折叠一下:

  

Eshell V6.0  (abort with ^G)

1> M=#{a=>123,b=>{ok,200},{c,d}=> <<"dvd">>}.
#{a => 123,b => {ok,200},{c,d} => <<"dvd">>}
2> maps:new().
#{}
3> maps:remove(b,M).
#{a => 123,{c,d} => <<"dvd">>}
4> maps:get(b,M).
{ok,200}
5> maps:keys(M).
[a,b,{c,d}]
6> maps:get(not_key,M).
** exception error: bad_key
     in function  maps:get/2
        called as maps:get(not_key,#{a => 123,b => {ok,200},{c,d} => <<"dvd">>})
7> maps:find(not_key,M).
error
8> maps:foldr(fun({K,V},Acc)->Acc++[{K,V}] end,[],M).
** exception error: undefined function maps:foldr/3
9> maps:foldl(fun({K,V},Acc)->Acc++[{K,V}] end,[],M).
** exception error: undefined function maps:foldl/3
10> maps:fold(fun({K,V},Acc)->Acc++[{K,V}] end,[],M).
** exception error: no function clause matching maps:fold(#Fun<erl_eval.12.101568567>,[],
                                                          #{a => 123,b => {ok,200},{c,d} => <<"dvd">>}) (maps.erl, line 168)
11> maps:fold(fun({K,V},Acc)->Acc++[{K,V}] end,[],M).
** exception error: no function clause matching maps:fold(#Fun<erl_eval.12.101568567>,[],
                                                          #{a => 123,b => {ok,200},{c,d} => <<"dvd">>}) (maps.erl, line 168)
12> maps:fold(fun(K,V,Acc)->Acc++[{K,V}] end,[],M).
[{a,123},{b,{ok,200}},{{c,d},<<"dvd">>}]
13> maps:to_list(M).
[{a,123},{b,{ok,200}},{{c,d},<<"dvd">>}]
14> maps:from_list(v(12)).
#{a => 123,b => {ok,200},{c,d} => <<"dvd">>}
16> maps:put(a,1984,M).
#{a => 1984,b => {ok,200},{c,d} => <<"dvd">>}
17> maps:values(M).
[123,{ok,200},<<"dvd">>]
18> maps:put(king,1984,M).
#{a => 123,b => {ok,200},king => 1984,{c,d} => <<"dvd">>}
19> maps:update(a,1984,M).
#{a => 1984,b => {ok,200},{c,d} => <<"dvd">>}
20> maps:update(king,1984,M).
** exception error: bad argument
     in function  maps:update/3
        called as maps:update(king,1984,#{a => 123,b => {ok,200},{c,d} => <<"dvd">>})
21> maps:merge(M,#{name=>"zen"}).
#{a => 123,b => {ok,200},name => "zen",{c,d} => <<"dvd">>}
22> maps:is_key(king,M).
false
23> maps:is_key(a,M).
true
24> maps:without([a,b,c],M).
#{{c,d} => <<"dvd">>}
25>

　　

比较!比较!

 

  Map的设计定位是data-type,那就存在Map与其它数据类型的比较规则,以及Map数据之间的比较规则.

 

[1] 与其它数据类型之间的比较

 

number < atom < reference < fun < port < pid < tuple <map < list < bit string

 

11> M= #{a=>123,b=>100}.
#{a => 123,b => 100}
12> M>12.
true
13> M> <<"12">>.
false
14> M> {a,b}.
true
15> M  > [1,2].
false

　　

 

[2] Map之间比较

 

     Two different maps M1 and M2 are sorted first after size and secondly after their Key=>Value pairs.
      lists:sort([list_to_tuple(maps:to_list(M)) || M <- [M1,M2]). 

 

运算符优先级

 

   没有应用在Map之间的运算符,只有两个内部的运算符=> :=; => 用于创建和更新k-v, := 用于更新已经存在的k-v, :=在匹配过程用来从指定的key中提取Value值;

 

Why Maps?

 

      我们已经有了record,dicts,gb_trees,ets,proplists,为什么还要Maps?Maps和现有的数据结构相比,最大的优势就是充分发挥Erlang模式匹配的威力. 我还关心的是之前的问题是否解决了:

  

1.可以把record的name用作参数吗?

#RecordName{} 可以吗? 因为没有RecordName的限制了,所以这个问题自然消失;

2.可以把record的filed作为参数使用吗?

 

Eshell V6.0  (abort with ^G)
1> M=#{a=>{1,2},b=>23,<<"OK">> =>ok, f=>fun()->receive a ->"Got a!" end end}.
#{a => {1,2},b => 23,f => #Fun<erl_eval.20.101568567>,<<"OK">> => ok}
2> N=a.
a
3> #{N := Data} =M.
* 1: illegal use of variable 'N' in map
4>

　　

而在代码模块中,下面的代码也会报错:illegal use of variable 'N' in map

test3(N,#{N := Data}=M)->
  Data.

3. a.b.c.d.e.f 能实现吗?

 

 

上次提到的a.b.c.d.e这种fluent 式的写法吗?这是肯定被毙掉的方案.问题就出在点号上,一方面它是语句结束,一方面它还是浮点数中的小数点;下面就是一个非常悲剧的例子:

 

1> M = #{ 1.1 => a, 1 => #{ 1 => b } }.
#{ 1 => #{ 1 => b }, 1.1 => a }.

2> #M.1.1.
a | b ?
 

　　

4.record转proplists proplists转record

 

现在已经没有必要转换了

5.key只能是atom

 

Maps中的Key可以是任意Erlang terms .

6.record往往要定义在hrl中

不需要了.

 

 

那就现在的情况,Maps会替代Record吗?

 

  EEP43 的重要依据是Richard O'Keefes 的 No more need for records (fifth draft).可以说Maps缘起record替换方案.而Maps最终的设计目标是"Maps does not claim to be an replacement to records as the frames proposal does. Instead maps targets a larger usage domain and wishes to be a complement to records and supersede them where suitable."

  从语言长远发展看,Map如果提供足够的便利,以及性能保障,淘汰掉record是一个开发者主动选择的自然过程,是一个"Maps不杀record,record却因Maps而死"过程.对于开发者倒不必有什么恐慌,record不会一夜之间消失,那么多的项目哪会在一朝一夕之间完成过渡?顺其自然就好.

  

悬而未决的功能

 

  有些语法特性在EEP43中提到了,但是在当前版本(17.0-rc2)并没有提供;首当其冲的就是"Accessing a single value",要达到这个目的可以通过模式匹配完成,也可以通过调用maps:get方法完成,所以我对这个方法的期待度并不大.

 

Eshell V6.0  (abort with ^G)
1> M=#{a=>12,b=>200,c=>234}.
#{a => 12,b => 200,c => 234}
2> #{b := B}=M.
#{a => 12,b => 200,c => 234}
3> B.
200
4> #{b := B,c := C}=M.
#{a => 12,b => 200,c => 234}
5> #{b := B,c := C,d:=D}=M.
** exception error: no match of right hand side value #{a => 12,b => 200,c => 234}

　 

其次就是Map comprehension ,我个人非常喜欢list comprehension,所以对这个功能还是非常期待的.

M1 = #{ E0 => E1 || K := V <- M0  }

 

 OK,今天就到这里,期待Erlang新版本的发布.

 

  最后感谢支付宝付款的小伙伴们,昨天早晨老婆跟我说"短信通知有人在支付宝给你打钱了",我还说"这是什么新诈骗手段啊",验证之后真的是很惊喜,谢谢你们的认可和支持,我会继续努力的.