Firstly a fun fact a lot people don’t know: when passing values between functions, it’s cheaper to pass values instead of pointers in Go. Reason is pointers to objects could be allocated on heap and it takes the computer more efforts to managed heap memory. While values are save on the stack, and stack is cheaper. For example, it’s preferable to:

type Foo struct {
	Bar string
}

func (f Foo) Echo() {
	println(f.Bar)
}

In above example, Echo method will not change the value of Foo, so there is no need to have the method receiver as a pointer.

Obviously Go compiler has gotten cleverer over the years and will do a lot more escape analysis to determine if a pointer to struct can be allocated only on stack. But I think it’s still a good practice to use value instead of pointers whenever you can.

Now here is a subtle bug that if used the value receive pattern incorrectly, can be tricky to track down. It certainly cost me quite some time a while ago and still bites me from time to time. Consider this example:

package main

type Bar struct {
	Baz string
}

func (b *Bar) Update() {
	b.Baz = "hello bar"
}

type Foo struct {
	Bar Bar
}

func (f Foo) UpdateBar() {
	f.Bar.Update()
}

func main() {
	bar := Bar{Baz: "doom"}
	foo := Foo{
		Bar: bar,
	}
	foo.UpdateBar()
	println(bar.Baz)
}

It’ll print doom not hello bar. It’s easy to see why in this simple program, but when it’s nested in structs, it could be a very long debugging process.

Note
If you embed a sync.Mutex in a struct and you attempt to use a value receiver, go vet will warn you about it. Try running go tool vet help copylocks

The main trick I want to share today probably seems very normal to any C or C++ programmers. But it is very useful in practice.

The problem Link to heading

I have a proto generated struct or some struct that I don’t own and I need to use it directly in sql scan. i.e. given the following struct:

type Node struct {
	ID 			int
	Name 		string
	CreatedAt 	timestamppb.Timestamp
}

However in database, I have Name and CreatedAt nullable, and I don’t really want to create a new struct or new variable using sql.NullString just to use in Row.Scan.

My solution Link to heading

type nullString struct {
	s *string
}

func (ts nullString) Scan(value interface{}) error {
	if value == nil {
		*ts.s = "" // nil to empty
		return nil
	}
	switch t := value.(type) {
	case string:
		*ts.s = t
	default:
		return fmt.Errorf("expect string in sql scan, got: %T", value)
	}
	return nil
}

func (n *nullString) Value() (driver.Value, error) {
	if n.s == nil {
		return "", nil
	}
	return *n.s, nil
}

And use it like this:

var node Node

db.QueryRow("select name from node where id=?", id)
	.Scan(nullString(&node.Name))

Note here I don’t need the pointer to the nullString, and I’m setting the value the pointer is pointing to directly. And I didn’t need to create a sql.NullString.

And similarly, I can also create a *timestamppb.Timestamp scanner:

import (
	"database/sql/driver"
	"fmt"
	"time"

	"google.golang.org/protobuf/types/known/timestamppb"
)

type tsScanner struct {
	pt **timestamppb.Timestamp
}

// Scan implements sql.Scanner for protobuf Timestamp.
func (ts tsScanner) Scan(value interface{}) error {
	if value == nil {
		*ts.pt = nil
		return nil
	}
	switch t := value.(type) {
	case time.Time:
		tspb := timestamppb.New(t)
		*ts.pt = tspb
	default:
		return fmt.Errorf("expect time.Time in sql scan got: %T", value)
	}
	return nil
}

// Value implements driver.Valuer for protobuf Timestamp.
func (ts tsScanner) Value() (driver.Value, error) {
	if ts.pt == nil || *ts.pt == nil {
		return nil, nil
	}
	return (*ts.pt).AsTime(), nil
}

And use it:

var node Node

db.QueryRow("select created_at from node where id=?", id)
	.Scan(tsScanner(&node.CreatedAt))

In the timestamppb.Timestamp example, I’m using the pointer of pointer to the struct because I need the set the pointer value, not the actual value.

It can make your head spin the whole pointer to pointer thing. Well, I’ll admit, it still does to me from time to time. Just need a bit more practice I suppose?