如何在没有泛型的情况下对此复合类型层次结构建模

Question

I have a system that parses a logfile which contains changesets of mysql tables, think of something like a binlog. There can be updates and inserts, deletes we ignore for now. The function of my module gets an input like this:

type Changeset struct {
    Table string // which table was affected
    Type string // INSERT or UPDATE
    OldData map[string]string // these 2 fields contain all columns of a table row
    NewData map[string]string
}

OldData is empty when it's an INSERT changeset, when it's an UPDATE changeset, OldData and NewData are filled (the data before and after the update).

Now I don't want to work with untyped data like this in my module, as I need to model some domain and it would be nicer to have some type safety. However, I need to still retain the knowledge if a change was an insert or an update for that domain logic (like, if it's an update, I will validate that some fields didn't change, as an example).

Assume I have two tables (let's say they only have one field named Id, but in reality they have more and different ones). So I modeled these objects like so:

type Foo struct { // foo table
    Id string
    // ... imagine more fields  here ...
}

type Bar struct { // bar table
    Id string
    // ... imagine more fields  here ...
}

Now I can map the map[string][string] from Changeset.OldData and Changeset.NewData, but then I don't know anymore if the change was an insert or an update. I was thinking a bit back and forth, but the best I came up with was:

type FooInsert struct {
    New Foo
}

type FooUpdate struct {
    New Foo
    Old Foo
}

type BarInsert struct {
    New Bar
}

type BarUpdate struct {
    New Bar
    Old Bar
}

And the mapping code looks like this:

func doMap(c Changeset) interface{} {
    if c.Table == "foo" {
        switch c.Type {
            case "UPDATE":
                return FooUpdate{Old: Foo{Id: c.OldData["id"]}, New: Foo{Id: c.NewData["id"]}}

            case "INSERT":
                return FooInsert{New: Foo{Id: c.NewData["id"]}}
        }
    }

    if c.Table == "bar" {
        switch c.Type {
                // ... almost same as above, but return BarUpdate/BarInsert ...
        }
    }

    return nil
}

The upside is, it enables me to write do do a typeswitch on the result of this mapping function like so:

insertChangeset := Changeset{
    Table: "foo",
    Type: "INSERT",
    NewData: map[string]string{"id": "1"},
}

o := doMap(insertChangeset)

switch o.(type) {
    case BarUpdate:
        println("Got an update of table bar")

    case FooUpdate:
        println("Got an update of table foo")

    case BarInsert:
        println("Got an insert to table bar")

    case FooInsert:
        println("Got an insert to table foo")           
}

The typeswitch is what I would need to have in the end (different types per change changeset type and per entity.) But:

• the mapping code as seen in doMap is very ugly and repetitive.
• for every new entity X I introduce, I need to create two more types XInsert and XUpdate.

Is there any way around this mess? In other programming languages I might have thought of something like:

type Update<T> {
    T Old
    T New
}

type Insert<T> {
    T New
}

But not sure how to model this in Go. I created also a playground sample that shows the whole code in one program: https://play.golang.org/p/ZMnB5K7RaI

Answer 1

have a look at this solution. It is one possible solution.

Generally: you want to work with interfaces here. In the sample I use the interface DataRow to store data of a row of any table. All table structs have to implement 2 functions as you can see in my example. (Also see my note about a general function in a base class with generics)

Here the code again:

package main

import "fmt"

type Foo struct {
    Id string
}

func (s *Foo) Fill(m map[string]string) {
    // If you want to build a general Fill you can build a base struct for Foo, Bar, etc. that works with reflect. 
    // Note that it will be slower than implementing the function here! Ask me if you want one I built recently.

    s.Id = m["id"]
}

func (s *Foo) GetRow() interface{} {
    return nil
}

type Bar struct {
    Id string
}

func (s *Bar) Fill(m map[string]string) {
    s.Id = m["id"]
}

func (s *Bar) GetRow() interface{} {
    return nil
}

type DataRow interface {
    Fill(m map[string]string)
    GetRow() interface{}
}

type Changeset struct {
    Table   string
    Type    string
    OldData map[string]string
    NewData map[string]string
}

type ChangesetTyped struct {
    Table   string
    Type    string
    OldData DataRow
    NewData DataRow
}

func doMap(c Changeset) ChangesetTyped {
    ct := ChangesetTyped{
        Table:   c.Table,
        Type:    c.Type,
        OldData: parseRow(c.Table, c.OldData),
    }

    if c.Type == "UPDATE" {
        ct.NewData = parseRow(c.Table, c.NewData)
    }

    return ct
}

func parseRow(table string, data map[string]string) (row DataRow) {
    if table == "foo" {
        row = &Foo{}
    } else if table == "bar" {
        row = &Bar{}
    }

    row.Fill(data)
    return
}

func main() {
    i := Changeset{
        Table:   "foo",
        Type:    "INSERT",
        NewData: map[string]string{"id": "1"},
    }

    u1 := Changeset{
        Table:   "foo",
        Type:    "UPDATE",
        OldData: map[string]string{"id": "20"},
        NewData: map[string]string{"id": "21"},
    }

    u2 := Changeset{
        Table:   "bar",
        Type:    "UPDATE",
        OldData: map[string]string{"id": "30"},
        NewData: map[string]string{"id": "31"},
    }

    m1 := doMap(i)
    m2 := doMap(u1)
    m3 := doMap(u2)

    fmt.Println(m1, m1.OldData)
    fmt.Println(m2, m2.OldData, m2.NewData)
    fmt.Println(m3, m3.OldData, m3.NewData)
}

If you want to get the actual row from DataRow cast to the correct type use (of type Foo in this example):

foo, ok := dt.GetRow().(Foo)
if !ok {
    fmt.Println("it wasn't of type Foo after all")
}

Hope this helps you in you golang quest!