golang interface
背景:
golang的interface是一种satisfied式的。A类只要实现了IA interface定义的方法,A就satisfied了接口IA。更抽象一层,如果某些设计上需要一些更抽象的共性,比如print各类型,这时需要使用reflect机制,reflect实质上就是将interface的实现暴露了一部分给应用代码。要理解reflect,需要深入了解interface。
go的interface是一种隐式的interface,但golang的类型是编译阶段定的,是static的,如:
|
1
2
3
|
type MyInt intvar i intvar j MyInt |
虽然MyInt底层就是int,但在编译器角度看,i的类型是int,j的类型是MyInt,是静态、不一致的。两者要赋值必须要进行类型转换。
即使是interface,就语言角度来看也是静态的。如:
|
1
|
var r io.Reader |
不管r后面用什么来初始化,它的类型总是io.Reader。
更进一步,对于空的interface,也是如此。
记住go语言类型是静态这一点,对于理解interface/reflect很重要。
看一例:
|
1
2
3
4
5
6
|
var r io.Readertty, err := os.OpenFile("/dev/tty", os.O_RDWR, 0)if err != nil { return nil, err}r = tty |
到这里,r的类型是什么?r的类型仍然是interface io.Reader,只是r = tty这一句,隐含了一个类型转换,将tty转成了io.Reader。
interface的实现:
作为一门编程语言,对方法的处理一般分为两种类型:一是将所有方法组织在一个表格里,静态地调用(C++, java);二是调用时动态查找方法(python, smalltalk, js)。
而go语言是两者的结合:虽然有table,但是是需要在运行时计算的table。
如下例:
Binary类实现了两个方法,String()和Get()
|
1
2
3
4
5
6
7
8
|
type Binary uint64func (i Binary) String() string { return strconv.Uitob64(i.Get(), 2)} func (i Binary) Get() uint64 { return uint64(i)} |
因为它实现了String(),按照golang的隐式方法实现来看,Binary satisfied了Stringer接口。因此它可以赋值: s:=Stringer(b)。
以此为例来说明下interface的实现:
interface的内存组织如图:

一个interface值由两个指针组成,第一个指向一个interface table,叫 itable。itable开头是一些描述类型的元字段,后面是一串方法。注意这个方法是interface本身的方法,并非其dynamic value(Binary)的方法。即这里只有String()方法,而没有Get方法。但这个方法的实现肯定是具体类的方法,这里就是Binary的方法。
当这个interface无方法时,itable可以省略,直接指向一个type即可。
另一个指针data指向dynamic value的一个拷贝,这里则是b的一份拷贝。也就是,给interface赋值时,会在堆上分配内存,用于存放拷贝的值。
同样,当值本身只有一个字长时,这个指针也可以省略。
一个interface的初始值是两个nil。比如,
|
1
|
var w io.Writer |
这时,tab和data都是nil。interface是否为nil取决于itable字段。所以不一定data为nil就是nil,判断时要额外注意。
这样,像这样的代码:
|
1
2
3
4
5
6
|
switch v := any.(type) {case int: return strconv.Itoa(v)case float: return strconv.Ftoa(v, 'g', -1)} |
实际上是any这个interface取了 any. tab->type。
而interface的函数调用实际上就变成了:
s.tab->fun[0](s.data)。第一个参数即自身类型指针。
itable的生成:
itable的生成是理解interface的关键。
如刚开始处提的,为了支持go语言这种接口间仅通过方法来联系的特性,是没有办法像C++一样,在编译时预先生成一个method table的,只能在运行时生成。因此,自然的,所有的实体类型都必须有一个包含此类型所有方法的“类型描述符”(type description structure);而interface类型也同样有一个类似的描述符,包含了所有的方法。
这样,interface赋值时,计算interface对象的itable时,需要对两种类型的方法列表进行遍历对比。如后面代码所示,这种计算只需要进行一次,而且优化成了O(m+n)。
可见,interface与itable之间的关系不是独立的,而是与interface具体的value类型有关。即(interface类型, 具体类型)->itable。
1 var any interface{} // initialized elsewhere2 s := any.(Stringer) // dynamic conversion3 for i := 0; i < 100; i++ {4 fmt.Println(s.String())5 } |
itable的计算不需要到函数调用时进行,只需要在interface赋值时进行即可,如上第2行,不需要在第4行进行。
最后,看一些实现代码:
以下是上面图中的两个字段。
143 type iface struct {144 tab *itab // 指南itable145 data unsafe.Pointer // 指向真实数据146 } |
再看itab的实现:
|
1
2
3
4
5
6
7
8
|
617 type itab struct {618 inter *interfacetype619 _type *_type620 link *itab621 bad int32622 unused int32623 fun [1]uintptr // variable sized624 } |
可见,它使用一个疑似链表的东西,可以猜这是用作hash表的拉链。
前两个字段应该是用来表达具体的interface类型和实际拥有的值的类型的,即一个itable的key。(上文提到的(interface类型, 具体类型) )
|
1
2
3
4
5
6
7
8
9
10
11
|
310 type imethod struct {311 name nameOff312 ityp typeOff313 }314315 type interfacetype struct {316 typ _type317 pkgpath name318 mhdr []imethod319 }320 |
interfacetype如有若干imethod,可以猜想这是表达interface定义的方法数据结构。
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
|
28 type _type struct {29 size uintptr30 ptrdata uintptr // size of memory prefix holding all pointers31 hash uint3232 tflag tflag33 align uint834 fieldalign uint835 kind uint836 alg *typeAlg37 // gcdata stores the GC type data for the garbage collector.38 // If the KindGCProg bit is set in kind, gcdata is a GC program.39 // Otherwise it is a ptrmask bitmap. See mbitmap.go for details.40 gcdata *byte41 str nameOff42 ptrToThis typeOff43 } |
对于_type,可见里面有gc的东西,应该就是具体的类型了。这里有个hash字段,itable实现就是挂在一个全局的hash table中。hash时用到了这个字段:
22 func itabhash(inter *interfacetype, typ *_type) uint32 {23 // compiler has provided some good hash codes for us.24 h := inter.typ.hash25 h += 17 * typ.hash26 // TODO(rsc): h += 23 * x.mhash ?27 return h % hashSize28 } |
可见,这里有个把interface类型与具体类型之间的信息结合起来做一个hash的过程,这个hash就是上述的itab的存储地点,itab中的link就是hash中的拉链。
回到itab,看取一个itab的逻辑:
如果发生了typeassert或是interface的赋值(强转),需要临时计算一个itab。这时会先在hash表中找,找不到才会真实计算。
44 h := itabhash(inter, typ)4546 // look twice - once without lock, once with.47 // common case will be no lock contention.48 var m *itab49 var locked int50 for locked = 0; locked < 2; locked++ {51 if locked != 0 {52 lock(&ifaceLock)53 }54 for m = (*itab)(atomic.Loadp(unsafe.Pointer(&hash[h]))); m != nil; m = m.link {55 if m.inter == inter && m._type == typ {71 return m // 找到了前面计算过的itab72 }73 }74 }75 // 没有找到,生成一个,并加入到itab的hash中。76 m = (*itab)(persistentalloc(unsafe.Sizeof(itab{})+uintptr(len(inter.mhdr)-1)*sys.PtrSize, 0, &memstats.other_sys))77 m.inter = inter78 m._type = typ79 additab(m, true, canfail) |
这个hash是个全局变量:
13 const (14 hashSize = 100915 )1617 var (18 ifaceLock mutex // lock for accessing hash19 hash [hashSize]*itab20 ) |
最后,看一下如何生成itab:
92 // both inter and typ have method sorted by name, 93 // and interface names are unique, 94 // so can iterate over both in lock step; 95 // the loop is O(ni+nt) not O(ni*nt). // 按name排序过的,因此这里的匹配只需要O(ni+nt) 99 j := 0100 for k := 0; k < ni; k++ {101 i := &inter.mhdr[k]102 itype := inter.typ.typeOff(i.ityp)103 name := inter.typ.nameOff(i.name)104 iname := name.name()109 for ; j < nt; j++ {110 t := &xmhdr[j]111 tname := typ.nameOff(t.name)112 if typ.typeOff(t.mtyp) == itype && tname.name() == iname {118 if m != nil {119 ifn := typ.textOff(t.ifn)120 *(*unsafe.Pointer)(add(unsafe.Pointer(&m.fun[0]), uintptr(k)*sys.PtrSize)) = ifn // 找到匹配,将实际类型的方法填入itab的fun121 }122 goto nextimethod123 }124 }125 }135 nextimethod:136 }140 h := itabhash(inter, typ) //插入上面的全局hash141 m.link = hash[h]142 atomicstorep(unsafe.Pointer(&hash[h]), unsafe.Pointer(m))143 } |
到这里,interface的数据结构的框架。
reflection实质上是将interface背后的实现暴露了一部分给应用代码,使应用程序可以使用interface实现的一些内容。只要理解了interface的实现,reflect就好理解了。如reflect.typeof(i)返回interface i的type,Valueof返回value.


浙公网安备 33010602011771号