英文:
Can I optimize this Go reflect function so it isn't as slow?
问题
我正在进行一个项目,需要为所有支持的方法构建部分更新。每个部分更新都需要一个不同的结构体,具有不同的字段和字段数量,并且不知道哪些字段会存在或不存在。我决定遍历每个结构体字段,如果存在则将其添加到一个数组中,在最后返回该数组。我还花了一些时间对几个看起来最符合实际的函数进行了基准测试,以帮助做出决策。
所有结构体字段都将是指针。在这个前提下,这是我编写的函数。
注意:我无法为此创建一个playground示例,因为它不支持基准测试。我将链接完整的类,并将解释放在它们上面。
1)为每个部分更新结构体创建一个映射函数,我将单独检查每个字段。如果字段不为nil
,我将以[key,value]的格式将值放入一个二维数组中。在处理完结构体后,返回该数组。
2)创建一个使用泛型和反射的单个映射函数,执行与上述相同的操作。
这是我运行的基准测试方法:
// main_test.go
package main
import (
"testing"
"time"
)
func BenchmarkBoth(b *testing.B) {
field1 := "testfield1"
field2 := "testfield2"
field3 := "testfield3"
field4 := "testfield4"
field5 := "testfield5"
date1, _ := time.Parse(time.RFC3339, "2004-10-16T12:40:53.00Z")
str := &updateRequest{
FieldOne: &field1,
FieldTwo: &field2,
FieldThree: &field3,
FieldFour: &field4,
FieldFive: &field5,
FieldSix: &date1,
}
b.Run("ManualUpdateRequestMapping", func(b *testing.B) {
for i := 0; i < b.N; i++ {
_ = ManualUpdateRequestMapping(str)
}
})
b.Run("ReflectUpdateRequestMapping", func(b *testing.B) {
for i := 0; i < b.N; i++ {
_ = ReflectUpdateRequestMapping(str)
}
})
}
下面是测试结果和使用的CPU:
cpu: 12th Gen Intel(R) Core(TM) i9-12900KF
BenchmarkBoth/ManualUpdateRequestMapping-24 3560083 331.9 ns/op 368 B/op 8 allocs/op
BenchmarkBoth/ReflectUpdateRequestMapping-24 1393377 866.7 ns/op 648 B/op 21 allocs/op
PASS
ok com.example.stack 3.745s
我预计反射函数会慢一些,但不会慢到大约2.5倍。它似乎还分配了大约2.5倍的资源。我在上面的代码中搞砸了吗,还是反射真的慢那么多?
如果有任何建议可以使上面的代码更高效,我愿意接受所有建议。我已经使用Go语言工作了大约3个月,所以如果我在上面的代码中犯了任何错误,请轻一点。:)
英文:
I am working on a project where I need to build out the partial updates for all of the methods that will be supported. Each partial update will require a different struct, with different fields and number of fields, and not knowing which ones will be present or not. I decided on going over each struct field, and if it's present adding it to an array to return at the end. I also took some time to Benchmark a couple of functions that seemed most realistic to approach this problem, and to help make a decision.
All of the structs fields will be pointers. With that in mind, these are the functions I wrote.
Note: I can't create a playground example for this, because it doesn't support Benchmarks. I'll link the full classes, and put the explanation above them.
- Create a mapping function for each partial update struct, where I would check each field separately. If the field is not
nil
, I will put the value in a 2D array storing in a [key,value] format. After the struct has been processed, return the array. - Create a single mapping function that uses Generics and Reflection to do the same as above.
// main.go
package main
import (
"reflect"
"strings"
"time"
)
type updateRequest struct {
FieldOne *string `json:"field_one,omitempty"`
FieldTwo *string `json:"field_two,omitempty"`
FieldThree *string `json:"field_three,omitempty"`
FieldFour *string `json:"field_four,omitempty"`
FieldFive *string `json:"field_five,omitempty"`
FieldSix *time.Time `json:"field_six,omitempty" time_format:"2006-01-02"`
}
// Mapping function that would need to be recreated for each partial update struct.
func ManualUpdateRequestMapping(req *updateRequest) [][]string {
vals := make([][]string, 0, 6)
if req.FieldOne != nil {
vals = append(vals, []string{"field_one", *req.FieldOne})
}
if req.FieldTwo != nil && req.FieldThree != nil {
vals = append(vals, []string{"field_two", *req.FieldTwo}, []string{"field_three", *req.FieldThree})
}
if req.FieldFour != nil {
vals = append(vals, []string{"field_four", *req.FieldFour})
}
if req.FieldFive != nil {
vals = append(vals, []string{"field_five", *req.FieldFive})
}
if req.FieldSix != nil {
vals = append(vals, []string{"field_six", req.FieldSix.Format(time.RFC3339)})
}
return vals
}
// Generics and Reflection function
func ReflectUpdateRequestMapping[T *updateRequest](str T) [][]string {
valHolder := reflect.ValueOf(*str)
if valHolder.Kind() != reflect.Struct {
return nil
}
vals := make([][]string, 0, valHolder.NumField())
for i := 0; i < valHolder.NumField(); i++ {
if valHolder.Field(i).IsNil() {
continue
}
spl := strings.Split(valHolder.Type().Field(i).Tag.Get("json"), ",")
if valHolder.Field(i).Elem().Type() != reflect.TypeOf(time.Time{}) {
vals = append(vals, []string{spl[0], valHolder.Field(i).Elem().String()})
} else {
vals = append(vals, []string{spl[0], valHolder.Field(i).Interface().(*time.Time).Format(time.RFC3339)})
}
}
return vals
}
This is the benchmark method I ran with:
// main_test.go
package main
import (
"testing"
"time"
)
func BenchmarkBoth(b *testing.B) {
field1 := "testfield1"
field2 := "testfield2"
field3 := "testfield3"
field4 := "testfield4"
field5 := "testfield5"
date1, _ := time.Parse(time.RFC3339, "2004-10-16T12:40:53.00Z")
str := &updateRequest{
FieldOne: &field1,
FieldTwo: &field2,
FieldThree: &field3,
FieldFour: &field4,
FieldFive: &field5,
FieldSix: &date1,
}
b.Run("ManualUpdateRequestMapping", func(b *testing.B) {
for i := 0; i < b.N; i++ {
_ = ManualUpdateRequestMapping(str)
}
})
b.Run("ReflectUpdateRequestMapping", func(b *testing.B) {
for i := 0; i < b.N; i++ {
_ = ReflectUpdateRequestMapping(str)
}
})
}
Below is the CPU used and the results that come from the test:
cpu: 12th Gen Intel(R) Core(TM) i9-12900KF
BenchmarkBoth/ManualUpdateRequestMapping-24 3560083 331.9 ns/op 368 B/op 8 allocs/op
BenchmarkBoth/ReflectUpdateRequestMapping-24 1393377 866.7 ns/op 648 B/op 21 allocs/op
PASS
ok com.example.stack 3.745s
I expected the Reflection function to be slower, but not ~2.5x slower. It also seems to allocate ~2.5x more resources per iteration. Did I botch something in the code above, or is Reflection just that much slower?
If there are any recommendations to make the code above more efficient, I am open to all suggestions. I've been working with Go for about 3 months now, so please go easy on me if I committed any Golang treason in the code above.
答案1
得分: 1
反射比非反射代码慢。这里有一个改进方法。一些注意事项:
- 通过将字段作为常规类型值获取并从那里进行操作,减少反射调用。
- 不需要那种新潮的类型参数东西。
- 说到Golang的背叛,语言的名称是Go。
解决了这些问题,下面是代码:
func UpdateRequestMapping(p any) [][]string {
v := reflect.ValueOf(p).Elem()
if v.Kind() != reflect.Struct {
return nil
}
t := v.Type()
result := make([][]string, t.NumField())
for i := 0; i < t.NumField(); i++ {
var ok bool
var value string
switch f := v.Field(i).Interface().(type) {
case *string:
if f != nil {
ok = true
value = *f
}
case *time.Time:
if f != nil {
ok = true
value = (*f).Format(time.RFC3339)
}
}
if ok {
name, _, _ := strings.Cut(t.Field(i).Tag.Get("json"), ",")
result[i] = []string{name, value}
}
}
return result
}
英文:
Reflection is slower than non-reflection code. Here's an improvement. Some notes:
- Reduce reflect calls by getting the field as a regular typed value and working from there.
- There's no need for that new-fangled type parameter stuff.
- Speaking of Golang treason, the name of the language is Go.
With that out of the way, here's the code:
func UpdateRequestMapping(p any) [][]string {
v := reflect.ValueOf(p).Elem()
if v.Kind() != reflect.Struct {
return nil
}
t := v.Type()
result := make([][]string, t.NumField())
for i := 0; i < t.NumField(); i++ {
var ok bool
var value string
switch f := v.Field(i).Interface().(type) {
case *string:
if f != nil {
ok = true
value = *f
}
case *time.Time:
if f != nil {
ok = true
value = (*f).Format(time.RFC3339)
}
}
if ok {
name, _, _ := strings.Cut(t.Field(i).Tag.Get("json"), ",")
result[i] = []string{name, value}
}
}
return result
}
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库,让每个人都能够通过互相帮助和分享经验来进步。
评论