Python 建構子與set與迴圈

[FakeWood]

國慶連假打python的時候，本來以為是自己邏輯上出錯的bug，後來發現好像跟python本身的運作比較有關，但我不知道怎麼會這樣所以上來尋求一下幫助，內容如下

先寫一個叫ObjWithSet的class，裡面裝著1個set
寫建構子讓物件被實例化的時候，在set裡放入1個東東

class ObjWithSet:
    s = set()
    def __init__( self, element ):
        self.s.add( element )

用for loop跑一下，創造3個物件

for i in range( 3 ):
    var = ObjWithSet( i )
    print( f'var{i} is at {id(var)}, var{i}.s is at {id(var.s)} : {var.s}' )

執行結果:

var0 is at [3130823073056], var0.s is at [3130823139488] : {0}
var1 is at [3130823077328], var1.s is at [3130823139488] : {0, 1}
var2 is at [3130823073056], var2.s is at [3130823139488] : {0, 1, 2}

我們可以看到，三個物件都在同一個位址上，這沒問題，因為他們都是局部變數而已
但是物件裡面的set也在同一個位址上，並且沒有被重置???
雖然剛剛在建構子裡沒有先把set用 .clear() 清掉，但我沒想到它不會自己清空

要命的是，要是用一個list來裝for loop裡所創建的物體...

vars = []
for i in range( 3,6 ):
    vars.append( ObjWithSet( i ) )
    print( f'var{i} is at [{id(vars[-1])}], var{i}.s is at [{id(vars[-1].s)}] : {vars[-1].s}' )

執行結果：

var3 is at [2991306104096], var3.s is at [2991306170528] : {3}
var4 is at [2991306108368], var4.s is at [2991306170528] : {3, 4}
var5 is at [2991306108512], var5.s is at [2991306170528] : {3, 4, 5}

雖然看似創造出了3個不同物件，但裡面的set是共用同一個，debug的時候差點哭出來

如果先在建構子裡面用 .clear() 清空set呢?

class ObjWithSet:
    s = set()
    def __init__( self, element ):
        self.s.clear()
        self.s.add( element )

一樣創3個物件，最後查看第一個物件裡的set

vars = []
for i in range( 3,6 ):
    vars.append( ObjWithSet( i ) )
    print( f'var{i} is at [{id(vars[-1])}], var{i}.s is at [{id(vars[-1].s)}] : {vars[-1].s}' )
print( f'var3.s: {vars[0].s}' )

執行結果：

var3 is at [2630846635296], var3.s is at [2630846701728] : {3}
var4 is at [2630846639568], var4.s is at [2630846701728] : {4}
var5 is at [2630846639712], var5.s is at [2630846701728] : {5}
var3.s: {5}

因為這三個物件裡的set是同一個，所以之前的var3.s果然還是被改掉了

那該如何讓每個物件裡的set都是獨一無二的呢?

把建構子寫成這樣，直接賦值而不用 .add()

class ObjWithSet2:
    s = set()
    def __init__( self, element ):
        self.s = { element }

一樣創3個物件

var1s = []
for i in range( 9,12 ):
    var1s.append( ObjWithSet2( i ) )
    print( f'var{i} is at [{id(var1s[-1])}], var{i}.s is at [{id(var1s[-1].s)}] : {var1s[-1].s}' )
print( f'var9: {var1s[0].s}' )

執行結果：

var9 is at [1352874391248], var9.s is at [1352876837568] : {9}
var10 is at [1352874393552], var10.s is at [1352876837120] : {10}
var11 is at [1352874390768], var11.s is at [1352876836896] : {11}
var9: {9}

可喜可賀，終於3個物件有著一個獨立的set了

但我想問的是，為什麼建構子用第一種寫法不行，第二種就可以呢??

下面是所有code:

# 這寫法會導致不同物件裡面的屬性的位址一樣，原因不明
class ObjWithSet:
    s = set()
    def __init__( self, element ):
        self.s.clear()
        self.s.add( element )

# # 直接迴圈會導致創建的物體位址一樣，不過這無所謂
for i in range( 3 ):
    var = ObjWithSet( i )
    print( f'var{i} is at [{id(var)}], var{i}.s is at [{id(var.s)}] : {var.s}' )

print()

# 先建一個容器才不會迴圈裡間的物體位址一樣，但因為Set寫法的關係，裡面的屬性位址一樣
vars = []
for i in range( 3,6 ):
    vars.append( ObjWithSet( i ) )
    print( f'var{i} is at [{id(vars[-1])}], var{i}.s is at [{id(vars[-1].s)}] : {vars[-1].s}' )


# 這寫法才能在物件裡創出獨一無二的set（應該吧
class ObjWithSet2:
    s = set()
    def __init__( self, element ):
        self.s = { element }

print()

# 但一樣要在外面有個容器存
for i in range( 6,9 ):
    var = ObjWithSet2( i )
    print( f'var{i} is at [{id(var)}], var{i}.s is at [{id(var.s)}] : {var.s}' )

print()

# 終於位址都不一樣了
var1s = []
for i in range( 9,12 ):
    var1s.append( ObjWithSet2( i ) )
    print( f'var{i} is at [{id(var1s[-1])}], var{i}.s is at [{id(var1s[-1].s)}] : {var1s[-1].s}' )

# 總之，建構子裡要直接丟值新創一個容器？

[ATREE01]

你的宣告方式錯誤了
要把set宣告在init內才會是屬於instance的元素，你的寫法會變成屬於class的元素所以不同instance之間會是同一個。

class test:
    def __init__(self, idx):
        self.s = set()
        self.s.add(idx)
        
for i in range(3):
    tmp = test(i)
    print(id(tmp), id(tmp.s), tmp.s)

[lazywulf]

大哥(或大姊)，你所宣告的 set 為 class variable 而非 instance variable，當然所有物件會共用同一個 s 的實例。如果你想要每個人都不一樣，請加 self。

class ObjWithSet:
    def __init__( self, element ):
        self.s = set()
        self.s.add( element )

[FakeWood]

原來如此! 感謝各位!
第一次用python寫class，不太習慣這樣把instance variable寫在建構子裡的方式哈哈