33 releases (breaking)

0.36.0 Aug 22, 2024
0.35.0 Feb 19, 2024
0.34.0 Nov 14, 2023
0.33.1 May 18, 2023
0.11.0 Jun 2, 2017

#120 in Encoding

Download history 3067/week @ 2024-08-21 2507/week @ 2024-08-28 2111/week @ 2024-09-04 2281/week @ 2024-09-11 2251/week @ 2024-09-18 2173/week @ 2024-09-25 1989/week @ 2024-10-02 2069/week @ 2024-10-09 2224/week @ 2024-10-16 2834/week @ 2024-10-23 2699/week @ 2024-10-30 2157/week @ 2024-11-06 2296/week @ 2024-11-13 2468/week @ 2024-11-20 4270/week @ 2024-11-27 3345/week @ 2024-12-04

12,760 downloads per month
Used in 19 crates (13 directly)

MIT/Apache

68KB
1.5K SLoC

serde_rusqlite

Documentation

See full documentation

Usage

Add this to your Cargo.toml:

[dependencies]
serde_rusqlite = "0.36.0"

Maintenance Build Status Documentation

Serde Rusqlite

This crate provides convenience functions to bridge serde and rusqlite. With their help you can "deserialize" rusqlite Row's into serde Deserialize types and "serialize" types implementing Serialize into bound query arguments (positional or named) that rusqlite expects.

Serialization of named bound arguments is only supported from structs and maps because other serde types lack column name information. Likewise, serialization of positional bound arguments is only supported from tuples, sequences and primitive non-iterable types. In the latter case the result will be single-element vector. Each serialized field or element must implement rusqlite::types::ToSql.

For deserialization you can use two families of functions: from_*() and from_*_with_columns(). The most used one is the former. The latter allows you to specify column names for types that need them, but don't supply them. This includes different Map types like HashMap. Specifying columns for deserialization into e.g. struct doesn't have any effect as the field list of the struct itself will be used in any case.

SQLite only supports 5 types: NULL (None), INTEGER (i64), REAL (f64), TEXT (String) and BLOB (Vec<u8>). Corresponding rust types are inside brackets.

Some types employ non-trivial handling, these are described below:

  • Serialization of u64 will fail if it can't be represented by i64 due to sqlite limitations.

  • Simple enums will be serialized as strings so:

    enum Gender {
       M,
       F,
    }
    

    will have two possible TEXT options in the database "M" and "F". Deserialization into enum from TEXT is also supported.

  • bools are serialized as INTEGERs 0 or 1, can be deserialized from INTEGER and REAL where 0 and 0.0 are false, anything else is true.

  • f64 and f32 values of NaN are serialized as NULLs. When deserializing such value Option<f64> will have value of None and f64 will have value of NaN. The same applies to f32.

  • Bytes, ByteBuf from serde_bytes are supported as optimized way of handling BLOBs.

  • unit serializes to NULL.

  • Only sequences of u8 are serialized and deserialized, BLOB database type is used. It's more optimal though to use Bytes and ByteBuf from serde_bytes for such fields.

  • unit_struct serializes to struct name as TEXT, when deserializing the check is made to ensure that struct name coincides with the string in the database.

Examples

use serde_derive::{Deserialize, Serialize};
use serde_rusqlite::*;

#[derive(Serialize, Deserialize, Debug, PartialEq)]
struct Example {
   id: i64,
   name: String,
}

let connection = rusqlite::Connection::open_in_memory().unwrap();
connection.execute("CREATE TABLE example (id INT, name TEXT)", []).unwrap();

// using structure to generate named bound query arguments
let row1 = Example { id: 1, name: "first name".into() };
connection.execute("INSERT INTO example (id, name) VALUES (:id, :name)", to_params_named(&row1).unwrap().to_slice().as_slice()).unwrap();
// and limiting the set of fields that are to be serialized
let row2 = Example { id: 10, name: "second name".into() };
connection.execute("INSERT INTO example (id, name) VALUES (2, :name)", to_params_named_with_fields(&row2, &["name"]).unwrap().to_slice().as_slice()).unwrap();

// using tuple to generate positional bound query arguments
let row2 = (3, "third name");
connection.execute("INSERT INTO example (id, name) VALUES (?, ?)", to_params(&row2).unwrap()).unwrap();

// deserializing using query() and from_rows(), the most efficient way
let mut statement = connection.prepare("SELECT * FROM example").unwrap();
let mut res = from_rows::<Example>(statement.query([]).unwrap());
assert_eq!(res.next().unwrap().unwrap(), row1);
assert_eq!(res.next().unwrap().unwrap(), Example { id: 2, name: "second name".into() });

// deserializing using query_and_then() and from_row(), incurs extra overhead in from_row() call
let mut statement = connection.prepare("SELECT * FROM example").unwrap();
let mut rows = statement.query_and_then([], from_row::<Example>).unwrap();
assert_eq!(rows.next().unwrap().unwrap(), row1);
assert_eq!(rows.next().unwrap().unwrap(), Example { id: 2, name: "second name".into() });

// deserializing using query_and_then() and from_row_with_columns(), better performance than from_row()
let mut statement = connection.prepare("SELECT * FROM example").unwrap();
let columns = columns_from_statement(&statement);
let mut rows = statement.query_and_then([], |row| from_row_with_columns::<Example>(row, &columns)).unwrap();
assert_eq!(rows.next().unwrap().unwrap(), row1);
assert_eq!(rows.next().unwrap().unwrap(), Example { id: 2, name: "second name".into() });

// deserializing using query() and from_rows_ref()
let mut statement = connection.prepare("SELECT * FROM example").unwrap();
let mut rows = statement.query([]).unwrap();
{
   // only first record is deserialized here
   let mut res = from_rows_ref::<Example>(&mut rows);
   assert_eq!(res.next().unwrap().unwrap(), row1);
}
// the second record is deserialized using the original Rows iterator
assert_eq!(from_row::<Example>(&rows.next().unwrap().unwrap()).unwrap(), Example { id: 2, name: "second name".into() });

License: MIT OR Apache-2.0

Dependencies

~22MB
~425K SLoC