pcopy.Copy
主要用于两个类型间的深度拷贝, 前身是deepcopy
新加预热函数。Copy时打开加速开关,达到性能提升4-10倍的效果。
警告:
高性能的同时可能会有些bug, 如果发现bug可以去掉pcopy.WithUsePreheat()
试下, 结果不一致,可以提issue。
- 高性能, 相对第一个版本提升4-10倍的性能
- 支持异构结构体拷贝, dst和src可以是不同的类型,会拷贝dst和src交集的部分
- 多类型支持struct/map/slice/array/int...int64/uint...uint64/ 等等
go get github.com/antlabs/pcopy
package main
import (
"fmt"
"github.com/antlabs/pcopy"
)
type dst struct {
ID int
Result string
}
type src struct{
ID int
Text string
}
func main() {
d, s := dst{}, src{ID:3}
pcopy.Preheat(&dst{}, &src{}) // 一对类型只要预热一次
pcopy.Copy(&d, &s, pcopy.WithUsePreheat())
fmt.Printf("%#v\n", d)
}
package main
import (
"fmt"
"github.com/antlabs/pcopy"
)
func main() {
i := []int{1, 2, 3, 4, 5, 6}
var o []int
pcopy.Preheat(&o, &i)
pcopy.Copy(&o, &i, pcopy.WithUsePreheat())
fmt.Printf("%#v\n", o)
}
package main
import (
"fmt"
"github.com/antlabs/pcopy"
)
func main() {
i := map[string]int{
"cat": 100,
"head": 10,
"tr": 3,
"tail": 44,
}
var o map[string]int
pcopy.Preheat(&o, &i)
pcopy.Copy(&o, &i, pcopy.WithUsePreheat())
fmt.Printf("%#v\n", o)
}
经常看到,对同一个结构体的,有值更新操作,都是一堆手工if 然后赋值的代码。不仅容易出错,还累。快使用pcopy解放双手。
type option struct {
Int int
Float64 float64
S string
}
func main() {
var a, b option
if b.Int != 0 {
a.Int = b.Int
}
if b.Float64 != 0.0 {
a.Float64 = b.Float64
}
if b.S != "" {
a.S = b.S
}
pcopy.Preheat(&a, &b) //只要预热一次
//可以约化成
pcopy.Copy(&a, &b, pcopy.WithUsePreheat())
}
从零实现的pcopy相比json序列化与反序列化方式拥有更好的性能
goos: darwin
goarch: arm64
pkg: benchmark
Benchmark_Use_reflectValue_MiniCopy-8 334728 3575 ns/op
Benchmark_Use_reflectValue_DeepCopy-8 595302 1956 ns/op
Benchmark_Use_reflectValue_Copier-8 203574 5860 ns/op
Benchmark_Use_Ptr_jsoniter-8 821113 1477 ns/op
Benchmark_Use_Ptr_pcopy-8 3390382 354.0 ns/op
Benchmark_Use_Ptr_coven-8 1414197 848.7 ns/op
PASS
ok benchmark 9.771s
从下面的压测数据可以看到,基本提供了4-10倍的性能提升
goos: darwin
goarch: arm64
pkg: github.com/antlabs/pcopy
Benchmark_BaseMap_Unsafe_Pcopy-8 529747 2343 ns/op
Benchmark_BaseMap_miniCopy-8 62181 19212 ns/op
Benchmark_BaseMap_Reflect-8 93810 12756 ns/op
Benchmark_BaseSlice_Unsafe_Pcopy-8 2013764 595.1 ns/op
Benchmark_BaseSlice_miniCopy-8 154918 7728 ns/op
Benchmark_BaseSlice_Reflect-8 188720 6393 ns/op
Benchmark_BaseType_Unsafe_Pcopy-8 4872112 243.8 ns/op
Benchmark_BaseType_MiniCopy-8 517814 2278 ns/op
Benchmark_BaseType_Pcopy-8 635156 1886 ns/op
Benchmark_CompositeMap_Unsafe_Pcopy-8 486253 2409 ns/op
Benchmark_CompositeMap_miniCopy-8 229674 5173 ns/op
Benchmark_CompositeMap_Reflect-8 475243 2490 ns/op
Benchmark_GetLikeFavorited_Unsafe_Pcopy2-8 446907 2662 ns/op
Benchmark_GetLikeFavorited_Unsafe_Pcopy-8 470217 2572 ns/op
Benchmark_GetLikeFavorited_MiniCopy-8 85674 13989 ns/op
Benchmark_GetLikeFavorited_Reflect_Pcopy-8 121603 9856 ns/op
Benchmark_GetRedPoint_Unsafe_Pcopy-8 1626688 736.1 ns/op
Benchmark_GetRedPoint_MiniCopy-8 650004 1871 ns/op
Benchmark_GetRedPoint_Reflect_Pcopy-8 1669778 722.0 ns/op
Benchmark_Interface_Unsafe_Pcopy-8 2869022 421.3 ns/op
Benchmark_Interface_MiniCopy-8 413936 2704 ns/op
Benchmark_Interface_Pcopy-8 440250 2688 ns/op
Benchmark_Interface_BaseSlice_Unsafe_Pcopy-8 1266501 947.4 ns/op
Benchmark_Interface_BaseSlice_MiniCopy-8 141610 8422 ns/op
Benchmark_Interface_BaseSlice_Pcopy-8 203906 5917 ns/op
Benchmark_Ptr_BaseType1_Unsafe_Pcopy-8 910153 1310 ns/op
Benchmark_Ptr_BaseType1_Reflect_Pcopy-8 391117 3026 ns/op
Benchmark_Ptr_BaseSlice_Unsafe_Pcopy-8 698156 1704 ns/op
Benchmark_Ptr_BaseSlice_Reflect_Pcopy-8 219999 5415 ns/op
Benchmark_SliceWithStruct_Unsafe_Pcopy-8 1395982 860.3 ns/op
Benchmark_SliceWithStruct_miniCopy-8 163154 7298 ns/op
Benchmark_SliceWithStruct_Reflect_Pcopy-8 190728 6213 ns/op