CODEWITHSUNDEEP
sqlgo
Don Draper
Don Draper

Reputation: 541

Scan a PostgreSQL field (of ARRAY type) into a slice of Go structs

Let's say I have:

type User struct {
    ID int64 `json:"id"
    Posts []Post `json:"posts"
}

type Post struct {
    ID int64 `json:"id"
    Text string `json:"text"
}

The SQL query:

WITH temp AS (SELECT u.id AS user_id, p.id AS post_id, p.text AS post_text FROM users u JOIN posts p ON u.id=p.user_id)

SELECT user_id, ARRAY_AGG(ARRAY[post_id::text, post_text]) 
FROM temp
GROUP BY user_id
)

What I want is to scan rows from the query above into a slice of User objects:

 import (
    "context"
    "fmt"
    "github.com/jackc/pgx/v4/pgxpool"
    "github.com/lib/pq"
 )

 var out []User

    rows, _ := client.Query(context.Background(), query) // No error handling for brevity

    for rows.Next() {

        var u User

        if err := rows.Scan(&u.ID, pq.Array(&u.Posts)); err != nil {
            return
        }

        out = append(out, u)

    }

Pretty much expectedly, the code above fails with:

pq: cannot convert ARRAY[4][2] to StringArray

This makes sense, but is there a way to read the SQL output into my slice of users?

Upvotes: 1

Views: 2576

Answers (1)

mkopriva
mkopriva

Reputation: 38203

Scanning of multi-dimensional arrays of arbitrary types, like structs, is not supported by lib/pq. If you want to scan such an array you'll have to parse and decode it yourself in a custom sql.Scanner implementation.

For example:

type PostList []Post

func (ls *PostList) Scan(src any) error {
    var data []byte
    switch v := src.(type) {
    case string:
        data = []byte(v)
    case []byte:
        data = v
    }

    // The data var holds the multi-dimensional array value,
    // something like: {{"1","foo"}, {"2","bar"}, ...}
    // The above example is easy to parse but too simplistic,
    // the array is likely to be more complex and therefore
    // harder to parse, but not at all impossible if that's
    // what you want.

    return nil
}

If you want to learn more about the PostgreSQL array representation syntax, see:

An approach that does not require you to implement a parser for PostgreSQL arrays would be to build and pass JSON objects, instead of PostgreSQL arrays, to array_agg. The result of that would be a one-dimensional array with jsonb as the element type.

SELECT user_id, array_agg(jsonb_build_object('id', post_id, 'text', post_text)) 
FROM temp
GROUP BY user_id

Then the implementation of the custom sql.Scanner just needs to delegate to lib/pq.GenericArray and another, element-specific sql.Scanner, would delegate to encoding/json.

type PostList []Post

func (ls *PostList) Scan(src any) error {
    return pq.GenericArray{ls}.Scan(src)
}

func (p *Post) Scan(src any) error {
    var data []byte
    switch v := src.(type) {
    case string:
        data = []byte(v)
    case []byte:
        data = v
    }
    return json.Unmarshal(data, p)
}

type User struct {
    ID    int64    `json:"id"`
    Posts PostList `json:"posts"`
}

Upvotes: 4

Related Questions