Skip to content

Latest commit

 

History

History
422 lines (276 loc) · 14.7 KB

CHANGELOG.md

File metadata and controls

422 lines (276 loc) · 14.7 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

0.3.4 - 2020-04-10

Fixed

  • [#241] Type name for custom enum is not always attached to TypeInfo in PostgreSQL

  • [#237] [#238] User-defined type name matching is now case-insensitive in PostgreSQL [@qtbeee]

  • [#231] Handle empty queries (and those with comments) in SQLite

  • [#228] Provide MapRow implementations for functions (enables .map(|row| ...) over .try_map(|row| ...))

Added

  • [#234] Add support for NUMERIC in MySQL with the bigdecimal crate [@xiaopengli89]

  • [#227] Support #[sqlx(rename = "new_name")] on struct fields within a FromRow derive [@sidred]

0.3.3 - 2020-04-01

Fixed

Changed

  • [#216] Mark Cursor, Query, QueryAs, query::Map, and Transaction as #[must_use] [@Ace4896]

  • [#213] Remove matches dependency and use matches macro from std [@nrjais]

0.3.2 - 2020-03-31

Fixed

  • [#212] Removed sneaky println! in MySqlCursor

0.3.1 - 2020-03-30

Fixed

  • [#203] Allow an empty password for MySQL

  • [#204] Regression in error reporting for invalid SQL statements on PostgreSQL

  • [#200] Fixes the incorrect handling of raw (r#...) fields of a struct in the FromRow derive [@sidred]

0.3.0 - 2020-03-29

Breaking Changes

  • sqlx::Row now has a lifetime ('c) tied to the database connection. In effect, this means that you cannot store Rows or collect them into a collection. Query (returned from sqlx::query()) has map() which takes a function to map from the Row to another type to make this transition easier.

    In 0.2.x

    let rows = sqlx::query("SELECT 1")
    .fetch_all(&mut conn).await?;

    In 0.3.x

    let values: Vec<i32> = sqlx::query("SELECT 1")
    .map(|row: PgRow| row.get(0))
    .fetch_all(&mut conn).await?;

    To assist with the above, sqlx::query_as() now supports querying directly into tuples (up to 9 elements) or struct types with a #[derive(FromRow)].

    // This extension trait is needed until a rust bug is fixed
use sqlx::postgres::PgQueryAs;

let values: Vec<(i32, bool)> = sqlx::query_as("SELECT 1, false")
    .fetch_all(&mut conn).await?;

  • HasSqlType<T>: Database is now T: Type<Database> to mirror Encode and Decode

  • Query::fetch (returned from query()) now returns a new Cursor type. Cursor is a Stream-like type where the item type borrows into the stream (which itself borrows from connection). This means that using query().fetch() you can now stream directly from the database with zero-copy and zero-allocation.

  • Remove PgTypeInfo::with_oid and replace with PgTypeInfo::with_name

Added

  • Results from the database are now zero-copy and no allocation beyond a shared read buffer for the TCP stream ( in other words, almost no per-query allocation ). Bind arguments still do allocate a buffer per query.

  • [#129] Add support for SQLite. Generated code should be very close to normal use of the C API.

    • Adds Sqlite, SqliteConnection, SqlitePool, and other supporting types

  • [#97] [#134] Add support for user-defined types. [@Freax13]

    • Rust-only domain types or transparent wrappers around SQL types. These may be used transparently inplace of the SQL type.

      #[derive(sqlx::Type)]
#[repr(transparent)]
struct Meters(i32);

    • Enumerations may be defined in Rust and can match SQL by integer discriminant or variant name.

      #[derive(sqlx::Type)]
#[repr(i32)] // Expects a INT in SQL
enum Color { Red = 1, Green = 2, Blue = 3 }
      #[derive(sqlx::Type)]
#[sqlx(rename = "TEXT")] // May also be the name of a user defined enum type
#[sqlx(rename_all = "lowercase")] // similar to serde rename_all
enum Color { Red, Green, Blue } // expects 'red', 'green', or 'blue'

    • Postgres further supports user-defined composite types.

      #[derive(sqlx::Type)]
#[sqlx(rename = "interface_type")]
struct InterfaceType {
    name: String,
    supplier_id: i32,
    price: f64
}

  • [#98] [#131] Add support for asynchronous notifications in Postgres (LISTEN / NOTIFY). [@thedodd]

    • Supports automatic reconnection on connection failure.

    • PgListener implements Executor and may be used to execute queries. Be careful however as if the intent is to handle and process messages rapidly you don't want to be tying up the connection for too long. Messages received during queries are buffered and will be delivered on the next call to recv().

    let mut listener = PgListener::new(DATABASE_URL).await?;

listener.listen("topic").await?;

loop {
    let message = listener.recv().await?;

    println!("payload = {}", message.payload);
}

  • Add unchecked variants of the query macros. These will still verify the SQL for syntactic and semantic correctness with the current database but they will not check the input or output types.

    This is intended as a temporary solution until query_as! is able to support user defined types.

    • query_as_unchecked!
    • query_file_as_unchecked!

  • Add support for many more types in Postgres

    • JSON, JSONB [@oeb25]
    • INET, CIDR [@PoiScript]
    • Arrays [@oeb25]
    • Composites ( Rust tuples or structs with a #[derive(Type)] )
    • NUMERIC [@abonander]
    • OID (u32)
    • "CHAR" (i8)
    • TIMESTAMP, TIMESTAMPTZ, etc. with the time crate [@utter-step]
    • Enumerations ( Rust enums with a #[derive(Type)] ) [@Freax13]

Changed

  • Query (and QueryAs; returned from query(), query_as(), query!(), and query_as!()) now will accept both &mut Connection or &Pool where as in 0.2.x they required &mut &Pool.

  • Executor now takes any value that implements Execute as a query. Execute is implemented for Query and QueryAs to mean exactly what they've meant so far, a prepared SQL query. However, Execute is also implemented for just &str which now performs a raw or unprepared SQL query. You can further use this to fetch Rows from the database though it is not as efficient as the prepared API (notably Postgres and MySQL send data back in TEXT mode as opposed to in BINARY mode).

    use sqlx::Executor;

// Set the time zone parameter
conn.execute("SET TIME ZONE LOCAL;").await

// Demonstrate two queries at once with the raw API
let mut cursor = conn.fetch("SELECT 1; SELECT 2");
let row = cursor.next().await?.unwrap();
let value: i32 = row.get(0); // 1
let row = cursor.next().await?.unwrap();
let value: i32 = row.get(0); // 2

Removed

  • Query (returned from query()) no longer has fetch_one, fetch_optional, or fetch_all. You must map the row using map() and then you will have a query::Map value that has the former methods available.

    let values: Vec<i32> = sqlx::query("SELECT 1")
    .map(|row: PgRow| row.get(0))
    .fetch_all(&mut conn).await?;

Fixed

  • [#62] [#130] [#135] Remove explicit set of IntervalStyle. Allow usage of SQLx for CockroachDB and potentially PgBouncer. [@bmisiak]

  • [#108] Allow nullable and borrowed values to be used as arguments in query! and query_as!. For example, where the column would resolve to String in Rust (TEXT, VARCHAR, etc.), you may now use Option<String>, Option<&str>, or &str instead. [@abonander]

  • [#108] Make unknown type errors far more informative. As an example, trying to SELECT a DATE column will now try and tell you about the chrono feature. [@abonander]

    optional feature `chrono` required for type DATE of column #1 ("now")

0.2.6 - 2020-03-10

Added

Fixed

0.2.5 - 2020-02-01

Fixed

  • Fix decoding of Rows containing NULLs in Postgres #104

  • After a large review and some battle testing by @ianthetechie of the Pool, a live leaking issue was found. This has now been fixed by @abonander in #84 which included refactoring to make the pool internals less brittle (using RAII instead of manual work is one example) and to help any future contributors when changing the pool internals.

  • Passwords are now being precent decoding before being presented to the server [@repnop]

  • [@100] Fix FLOAT and DOUBLE decoding in MySQL

Added

  • [#72] Add PgTypeInfo::with_oid to allow simple construction of PgTypeInfo which enables HasSqlType to be implemented by downstream consumers of SQLx [@jplatte]

  • [#96] Add support for returning columns from query! with a name of a rust keyword by using raw identifiers [@yaahc]

  • [#71] Implement derives for Encode and Decode. This is the first step to supporting custom types in SQLx. [@Freax13]

0.2.4 - 2020-01-18

Fixed

  • Fix decoding of Rows containing NULLs in MySQL (and add an integration test so this doesn't break again)

0.2.3 - 2020-01-18

Fixed

  • Fix query! when used on a query that does not return results

0.2.2 - 2020-01-16

Added

  • [#57] Add support for unsigned integers and binary types in query! for MySQL [@mehcode]

Fixed

  • Fix stall when requesting TLS from a Postgres server that explicitly does not support TLS (such as postgres running inside docker) [@abonander]

  • [#66] Declare used features for tokio in sqlx-macros explicitly

0.2.1 - 2020-01-16

Fixed

  • [#55] Use a shared tokio runtime for the query! macro compile-time execution (under the runtime-tokio feature) [@udoprog]

0.2.0 - 2020-01-15

Fixed

Added

  • Support Tokio through an optional runtime-tokio feature.

  • Support SQL transactions. You may now use the begin() function on Pool or Connection to start a new SQL transaction. This returns sqlx::Transaction which will ROLLBACK on Drop or can be explicitly COMMIT using commit().

  • Support TLS connections.

0.1.4 - 2020-01-11

Fixed

Added

0.1.3 - 2020-01-06

Fixed

0.1.2 - 2020-01-03

Added

  • Support for Authentication in MySQL 5+ including the newer authentication schemes now default in MySQL 8: mysql_native_password, sha256_password, and caching_sha2_password.

  • Chrono support for MySQL was only partially implemented (was missing NaiveTime and DateTime<Utc>).

  • Vec<u8> (and [u8]) support for MySQL (BLOB) and Postgres (BYTEA).