#table #parameters #traits #sqlite #macro #sql #macro-derive

macro nanosql_macros

Tiny, strongly-typed data mapper for SQLite (procedural macros)

11 releases (breaking)

0.9.1 Aug 28, 2024
0.8.0 Aug 26, 2024
0.3.1 Jul 22, 2024

#1276 in Procedural macros


Used in nanosql

MIT license

73KB
1.5K SLoC

NanoSQL: a tiny, strongly-typed data mapper for SQLite

NanoSQL is a small data mapper library that helps you execute SQL statements with typed parameters and a typed result set. It does not attempt to typecheck your SQL code. Rather, it only ensures that the parameters and results serialize/deserialize to/from the correct shape.

For an overview of SQLite, see the official docs.

Overview

The library is structured around prepared statements. First, you create a description of the interface and implementation of your query. This is realized by the Query trait that contains the input (parameter) and output (result) types as associated types, and a function for building the SQL text. You can imeplement this trait by hand, or use the define_query macro as a convenient shortcut.

Next, you use [Connection::compile()] to compile the Query into a CompiledStatement. This wraps an SQLite prepared statement, but restricts its parameter and return types.

Finally, you call the [CompiledStatement::invoke()] function on your compiled statement to actually query the database:

  • The input of a query is any type that implements the Param trait. These include primitives, optionals of primitives, tuples of primitives or optionals, and structs with fields of primitive or optional types. Nanosql can work with both positional and named arguments, and supports all parameter prefixes accepted by SQLite (?, :, @, and $). This trait can be derived if the derive feature of the crate is activated.
  • The output of a query is a type that implements the ResultSet trait. This is most commonly a collection of some sort (e.g., the standard Vec type), or any other type that aggregates the rows returned from an SQL query into a meaningful data structure. Notably, Option<T> and Single<T> can be used for expecting at most one or exactly one record, respectively. These types implement ResultSet when the type of their wrapped value implements ResultRecord.
  • ResultRecord is a trait that can be implemented by tuple-like and struct-like types for deserializing individual rows. This trait can also be #[derive]d.

Extremely common (basically inevitable) tasks such as creating the schema of a table and inserting records is possible via special helper/extension methods on Connection objects, via the ConnectionExt trait. These in turn use the Table trait for preparing and invoking the corresponding SQL statements, and can be called for convenience.

Examples

The absolute basics - create a table, insert a bunch of records into it, then retrieve them:

use std::fmt::{self, Formatter};
use nanosql::{
    Result, Connection, ConnectionExt, Query,
    ToSql, FromSql, AsSqlTy, Param, ResultRecord, Table
};


/// This type is going to represent our table.
///
/// We derive the `Param` trait for it so that it can be used for
/// binding parameters to the statement when inserting new records,
/// and the `ResultRecord` trait so that we can use it to retrieve
/// results, too.
///
/// We also derive the `Table` trait so that basic operations such as
/// creating the table in the schema and bulk insertion can be performed
/// using the appropriate convenience methods in [`ConnectionExt`].
///
/// The parameter prefix is '$' by default (if not specified via the
/// param_prefix attribute); it may also be one of ':', '@', or '?',
/// the last one being allowed only for tuples and scalar parameters.
///
/// `#[nanosql(rename)]` on a struct renames the table itself, while
/// `#[nanosql(rename_all)]` applies a casing convention to all columns.
#[derive(Clone, PartialEq, Eq, Hash, Debug, Param, ResultRecord, Table)]
#[nanosql(param_prefix = '$')] // optional
#[nanosql(rename = "MyLittlePet", rename_all = "lowerCamelCase")]
struct Pet {
    /// If you don't like the default `AsSqlTy` impl for your column's
    /// type, you can specify a different one. Here we add a non-zero
    /// constraint, but the `id` remains a plain `i64` for convenience.
    ///
    /// You can also add additional `CHECK` constraints, if necessary.
    #[nanosql(sql_ty = core::num::NonZeroI64, check = "id <= 999999")]
    id: i64,
    /// You can apply a `UNIQUE` constraint to any field.
    #[nanosql(unique)]
    nick_name: String,
    /// You can also rename fields/columns one by one
    #[nanosql(rename = "type")]
    kind: PetKind,
}

/// Collective and field-level casing/renaming also works with `enum`s
#[derive(Clone, Copy, PartialEq, Eq, Hash, Debug, ToSql, FromSql, AsSqlTy)]
#[nanosql(rename_all = "UPPER_SNAKE_CASE")]
enum PetKind {
    Dog,
    #[nanosql(rename = "KITTEN")]
    Cat,
    Fish,
}

/// Our first custom query retrieves a pet by its unique ID.
///
/// If you don't want to spell out the impl by hand, you can
/// use the `define_query!{}` macro for a shorter incantation.
struct PetById;

impl Query for PetById {
    /// The type of the parameter(s). This can be a single scalar, a tuple,
    /// a tuple struct of scalars, or a struct with named fields of scalar types.
    type Input<'p> = i64;

    /// The return type of a query can be either a scalar, a single record (struct or
    /// tuple), or an optional of a scalar/record (when it returns either 0 or 1 rows),
    /// or a collection of arbitrarily many scalars/records. Here we choose an `Option`,
    /// because a given ID corresponds to at most one `Pet`.
    type Output = Option<Pet>;

    /// Finally, we create the actual SQL query.
    fn format_sql(&self, formatter: &mut Formatter<'_>) -> fmt::Result {
        formatter.write_str("SELECT id, nickName, type FROM MyLittlePet WHERE id = ?")
    }
}

fn main() -> Result<()> {
    // First, we open a database connection.
    let mut conn = Connection::connect_in_memory()?;
    
    // Then, we ensure that the table exists in the schema.
    conn.create_table::<Pet>()?;

    // Next, we insert a couple of records so we have test data to work on.
    conn.insert_batch([
        Pet {
            id: 1,
            nick_name: "Fluffy".into(),
            kind: PetKind::Dog,
        },
        Pet {
            id: 2,
            nick_name: "Hello Kitty".into(),
            kind: PetKind::Cat,
        },
        Pet {
            id: 3,
            nick_name: "Nemo".into(),
            kind: PetKind::Fish,
        },
    ])?;

    // We then compile the query into a prepared statement.
    let mut stmt = conn.compile(PetById)?;

    // Finally, we execute the compiled statement, passing parameters, and retrieve the results.
    let result = stmt.invoke(3)?;
    assert_eq!(result, Some(Pet { 
        id: 3,
        nick_name: "Nemo".into(),
        kind: PetKind::Fish,
    }));

    // We can re-use the statement and execute it multiple times
    let result = stmt.invoke(99)?;
    assert_eq!(result, None);

    drop(stmt);

    // Inserting a pet with id = 0 should fail due to the `#[nanosql(sql_ty = ...)]` attribute.
    let insert_id_0_result = conn.insert_batch([
        Pet {
            id: 0,
            nick_name: "Error".into(),
            kind: PetKind::Cat,
        }
    ]);
    assert!(insert_id_0_result.is_err(), "id = 0 violates NonZeroI64's CHECK constraint");

    // Inserting a pet with a high ID is expected to fail due to the CHECK constraint.
    let insert_id_high_result = conn.insert_batch([
        Pet {
            id: 1000000,
            nick_name: "this is unique".into(),
            kind: PetKind::Dog,
        }
    ]);
    assert!(insert_id_high_result.is_err(), "id = 1000000 violates extra CHECK constraint");

    // Inserting a pet with a duplicate name is not allowed due to `#[nanosql(unique)]`.
    let insert_dup_name_result = conn.insert_batch([
        Pet {
            id: 137731,
            nick_name: "Hello Kitty".into(),
            kind: PetKind::Dog,
        }
    ]);
    assert!(insert_dup_name_result.is_err(), "duplicate name violates uniqueness constraint");

    Ok(())
}

See the nanosql/examples directory (and especially realistic.rs) for more advanced and interesting examples.

A note about batch insertion and transactions

The [ConnectionExt::insert_batch()] method wraps the insertion statements in a transaction for improving performance. The exclusiveness of transactions is modeled in rusqlite at the type level by the Connection object being mutably (uniquely) borrowed for the duration of the transaction. This in turn means that insert_batch() also needs to mutably borrow. However, preparing and invoking queries needs an immutable borrow, and prepared statements borrow the Connection for as long as they live. As such, you may get errors like "cannot borrow connection mutably because it is also borrowed as immutable" when mixing batch insertion with other queries. There are two basic solutions to this problem:

  1. Drop the outstanding prepared statements before calling [ConnectionExt::insert_batch()];
  2. Or if you can't do that, then obtain a transaction object that is not checked at compilation time for exclusiveness, using [Connection::unchecked_transaction()], then call the immutably borrowing [TransactionExt::insert_batch()] method on the transaction object instead.

Cargo Features

  • derive: activates procedural macros - mostly custom #[derive]s for commonly-used traits. Enabled by default.
  • expr-check: uses the sqlparser crate to check for syntax errors in raw SQL expressions in derive macro attributes at compile time. This ensures that any generated SQL will be valid and syntax error in user-supplied SQL code will be clearly pinpointed, instead of causing mysterious statement preparation errors at runtime. Enabled by default.
  • not-nan: implements Param and ResultRecord for ordered_float::NotNan. This allows for a more type-safe interface in queries: since SQLite treats NaN as the SQL NULL value, you may run into surprising errors when binding or retrieving an f32::NAN or f64::NAN and the corresponding parameter needs to be NOT NULL, or the source column can be NULL.
  • chrono: adds support for DateTime<Utc | FixedOffset | Local>, by serializing timestamps to and from the RFC-3339 format.
  • uuid: adds support for uuid::Uuid, by representing UUIDs as 16-byte blobs.
  • json: adds support for serde_json::Value (re-exported as JsonValue).
  • pretty-eqp: use the ptree crate to pretty print the results of EXPLAIN QUERY PLAN. This will impl Display for QueryPlan, the return type of [ConnectionExt::explain_query_plan()], which renders the tree in a nice, human-readable format using ASCII art.

Notes on the test suite

  1. The tests try to exercise all features of the library extensively. For this reason, they rely on most or all Cargo features defined in Cargo.toml. Consequently, for successfully compiling and running all tests, you must pass cargo test --all-features to Cargo.

  2. The compiletest_rs crate is used for ensuring that the derive macros detect certain kinds of errors, such as multiple primary keys or table-level constraints that reference non-existent columns.

    Due to the way the compiletest_rs crate is structured, the tests can be somewhat flaky. If you encounter E0464 errors (e.g., "multiple candidates for rlib dependency nanosql found"), then run cargo clean before running cargo test compile_fail --all-features.

TL;DR: the best "lazy" way to run tests is the ./runtests.sh script, which just does:

cargo clean
cargo test --workspace --all-features

Dependencies

~2.5–3.5MB
~71K SLoC