Can't use v4 Get()/Set() Array behaviour in v5

[crodwell]

With the AssignTo() and Set() funcs available v4 now gone, scanning works from a pgx Row and also with basic go native types such as string & int but not native slices:

	var result pgtype.Array[string]
	pgxconn.QueryRow(context.Background(), "SELECT ARRAY['test1', 'test2']").Scan(&result)

	 // result:
	 // (pgtype.Array[string]) {
         //       Elements: ([]string) (len=2 cap=2) {
         //       (string) (len=5) "test1",
         //       (string) (len=5) "test2"
         //      },
         //      Dims: ([]pgtype.ArrayDimension) (len=1 cap=1) {
         //       (pgtype.ArrayDimension) {
         //        Length: (int32) 2,
         //        LowerBound: (int32) 1
         //       }
         //      },
         //      Valid: (bool) true
         //     }

	data := []string{"test1", "test2"}

	err := pgtype.NewMap().SQLScanner(data).Scan(&result)
         fmt.Println(err)
         //  cannot scan _text (OID 1009) in text format into []string

How can I do the equivalent of result.Set(data) as I used to be able to do in pgx v4?

[jackc]

What exactly are you trying to do? AssignTo and Set shouldn't be necessary anymore.

You shouldn't need SQLScanner unless you are using database/sql and if you are that's not a valid argument for a sql.Scanner.

Scanning query results directly to Go slices should work out of the box. For example from the test suite:

pgx/pgtype/array_codec_test.go

Lines 326 to 329 in 19039e6

	var actual []string
	err := conn.QueryRow(ctx, `select '{"foo", "NULL", " NULL "}'::text[]`).Scan(&actual)
	require.NoError(t, err)
	require.Equal(t, []string{"foo", "NULL", " NULL "}, actual)

[crodwell]

Effectively in v4 I had a model for each table defined as a struct with each field a pgtype. The API would take a request struct bound from a JSON request - and transform that into an update to one or more models using Set methods. I could then compare changes to model for auditing, and save model state.. A pretty common pattern in an MVC type framework.

Yes, I gathered the Scan is really for scanning database results into the pgtype and there isn't a way to set the pgtype from native go in a generic way anymore .

I'm currently headed down a path of creating wrappers like this which I don't think is ideal.

type TextArray struct {
	pgtype.Array[string]
}

// Set makes up for a non-existent method in the pgx v5 lib that worke in v4
func (dst *TextArray) Set(value []string) {
	// construct dst manually
}

[crodwell]

I was successful using wrapper structs:

import (
	"fmt"
	"strconv"

	"github.com/jackc/pgx/v5/pgtype"
)

// Bool is a wrapper for pgtype.Bool allowing a .Set() method
type Bool struct {
	pgtype.Bool
}

// Set makes up for a non-existent method in the pgx v5 lib that worked in v4
func (dst *Bool) Set(src any) error {

The only negative is my code that used to get a Bool in the following way:

var myBool pgtype.Bool
value := myBool.Bool

now has to be referenced like so due to the conflicting naming of Bool and it's underlying value:

var myBool customTypes.Bool
value := customTypes.Bool.Bool

It would be really nice if Set() and Get() were reinstated in pgx/pgtype v5

[jackc]

It's unlikely that Set and Get will return. They were mostly a consequence of how the type system in v4 combined the concepts of a value container and a value encoder / decoder. That design had deficiencies which were resolved by v5 making Codec is an explicit concept.

However, v5 usually makes it easier to create application or domain types that can be directly used with pgx by exposing per PostgreSQL type interfaces. For your bool example, your type could implement BoolScanner and BoolValuer directly. This would remove the need to wrap pgtype.Bool altogether.