How do I implement Queryable and Insertable for custom field types in Diesel?

I have an SQL table that I want to work with through Diesel:

CREATE TABLE records (
    id BIGSERIAL PRIMARY KEY,
    record_type SMALLINT NOT NULL,
    value DECIMAL(10, 10) NOT NULL
)

This table generates the following schema:

table! {
    records (id) {
        id -> Int8,
        record_type -> Int2,
        value -> Numeric,
    }
}

Diesel exports decimals as bigdecimal::BigDecimal, but I'd like to work with decimal::d128 instead. I also want to map record_type to an enum, so I declare my model like this:

use decimal::d128;

pub enum RecordType {
    A,
    B,
}

pub struct Record {
    pub id: i64,
    pub record_type: RecordType,
    pub value: d128,
}

I can't use #derive(Queryable, Insertable) because of non-standard type mapping, so I try to implement these traits myself:

impl Queryable<records::SqlType, Pg> for Record {
    type Row = (i64, i16, BigDecimal);

    fn build(row: Self::Row) -> Self {
        Record {
            id: row.0,
            record_type: match row.1 {
                1 => RecordType::A,
                2 => RecordType::B,
                _ => panic!("Wrong record type"),
            },
            value: d128!(format!("{}", row.2)),
        }
    }
}

I can't figure out how to implement Insertable. What is the Values associated type? Diesel's documentation is not very clear on this.

Maybe there's a better way to achieve what I'm trying to do?

Cargo.toml:

[dependencies]
bigdecimal = "0.0.10"
decimal = "2.0.4"
diesel = { version = "1.1.1", features = ["postgres", "bigdecimal", "num-bigint", "num-integer", "num-traits"] }
dotenv = "0.9.0"

Solution

I find it more convenient to create newtype wrappers that implement ToSql and FromSql. You can then build with these basic blocks to create larger types that can derive Queryable / Insertable.

This example only shows how to perform the mapping of the enum to and from a SmallInt, but the case for the decimal would be the same. The only difference would be in how you perform the transformations:

#[macro_use]
extern crate diesel;

mod types {
    use diesel::sql_types::*;
    use diesel::backend::Backend;
    use diesel::deserialize::{self, FromSql};
    use diesel::serialize::{self, ToSql, Output};
    use std::io;

    table! {
        records (id) {
            id -> BigInt,
            record_type -> SmallInt,
        }
    }

    #[derive(Debug, Copy, Clone, AsExpression, FromSqlRow)]
    #[sql_type = "SmallInt"]
    pub enum RecordType {
        A,
        B,
    }

    impl<DB: Backend> ToSql<SmallInt, DB> for RecordType
    where
        i16: ToSql<SmallInt, DB>,
    {
        fn to_sql<W>(&self, out: &mut Output<W, DB>) -> serialize::Result
        where
            W: io::Write,
        {
            let v = match *self {
                RecordType::A => 1,
                RecordType::B => 2,
            };
            v.to_sql(out)
        }
    }

    impl<DB: Backend> FromSql<SmallInt, DB> for RecordType
    where
        i16: FromSql<SmallInt, DB>,
    {
        fn from_sql(bytes: Option<&DB::RawValue>) -> deserialize::Result<Self> {
            let v = i16::from_sql(bytes)?;
            Ok(match v {
                1 => RecordType::A,
                2 => RecordType::B,
                _ => return Err("replace me with a real error".into()),
            })
        }
    }

    #[derive(Insertable, Queryable, Debug)]
    #[table_name = "records"]
    pub struct Record {
        pub id: i64,
        pub record_type: RecordType,
    }
}

There's a draft guide describing all the derives and their annotations, but it doesn't yet mention #[sql_type] for an entire type. This lets Diesel know what kind of underlying storage is needed inside of the database.