Cornucopia

Generate type checked Rust from your SQL

Install | Example

Cornucopia is a small CLI utility resting on tokio-postgres and designed to facilitate PostgresQL workflows in Rust.

Cornucopia aims to get out of your way, transpiling your queries to Rust on demand without requiring you to maintain a live database connection. Each query is prepared against your schema, ensuring that the statement is valid SQL. These prepared statements are then be used to generate properly typed rust code for this query. Keep reading for more info, or take a look at the examples folder for a quickstart 🚀

---

Install

Docker

The CLI spawns a postgres container when it has to generate Rust modules. Thus, you need to have a working docker. Note that on Linux non-sudo users need to be in the docker group. Read the official installation and post-installation steps.

Dependencies

Cornucopia will generate queries powered by the tokio runtime through tokio-postgres and deadpool-postgres, so you will need add the latest version of these to your dependencies. You might need more dependencies depending on which features you inted to use, The code block below shows an example of what your dependencies might look like with every feature that cornucopia supports:

# Cargo.toml
[dependencies]
tokio = { version = "1.17.0", features = ["full"] }
deadpool-postgres = { version = "0.10.2", features = ["serde"] }
tokio-postgres = { version = "0.7.5", features = [
    "with-serde_json-1",
    "with-time-0_3",
    "with-uuid-0_8",
] }
serde = { version = "1.0.136", features = ["derive"] }
serde_json = "1.0.79"
time = "0.3.9"
uuid = "0.8.2"

You can omit tokio-postgres feature flags for json, time and uuid and their corresponding crates if you don't need them.

Cornucopia CLI

Aside from the dependencies, you will need the lightweight cornucopia cli to generate your Rust modules. This can be done via a simple cargo install cornucopia which will pull the latest binary and install it in your cargo path.

Concepts

This section explain a bit more about how cornucopia works. If you just want to get started, take a look at the examples folder.

Cornucopia is pretty simple to use. Your migrations and queries should each reside in a dedicated folder, and from there the CLI takes care of the rest for you. In the next sections, we'll explore the basic usage, but feel free to explore the CLI's whole interface using the --help option at any point.

Migrations

New migrations can be added using the command cornucopia migration new. Cornucopia will automatically manage migrations when it generates your Rust modules, but you can also use the command cornucopia migration run to run migrations on your production database too if you so desire.

Queries

Each .sql file in your query directory will be converted into a Rust module containing functions corresponding to these queries. These functions are fully typed so you know exactly what parameters it takes, and what it returns. Queries are augmented by special comments allowing you you quickly fine-tune them.

Generated modules

Assuming you have the following migration

CREATE TABLE Authors (
    Id SERIAL NOT NULL,
    Name VARCHAR(70) NOT NULL,
    Country VARCHAR(100) NOT NULL,
    PRIMARY KEY(Id)
);

then, the following query

--! authors()*
SELECT * FROM Authors;

will be turned by cornucopia into

pub async fn authors(client: &Client) -> Result<Vec<(i32, String, String)>, Error> {
    let stmt = client
        .prepare_typed_cached(
            "SELECT * FROM Authors;", &[],
        )
        .await?;

    let res = client.query(&stmt, &[]).await?;

    let return_value = res
        .iter()
        .map(|res| {
            let return_value_0: i32 = res.get(0);
            let return_value_1: String = res.get(1);
            let return_value_2: String = res.get(2);
            (return_value_0, return_value_1, return_value_2)
        })
        .collect::<Vec<(i32, String, String)>>();
    Ok(return_value)
}

Not bad! The generated function uses prepared statements, a statement cache, and strong typing (Notice how the returned rows' types have been inferred!). This is only a taste of what you can achieve, but should be fairly representative of what's going on under the hood.

Meta query syntax

As you may have noticed from the previous section, this little comment --! authors()* is doing a lot of heavy-lifting for us. It tells cornucopia to generate a function name authors which takes no parameters. Since there is no specified return, cornucopia lets Postgres infer the types itself, which it is usually pretty good at. Then, there's the asterisk * which signals that this query will return zero or more results. That's how we ended up with a Vec in the generated query in the section above.

Note that comments that do not start with --! are simply ignored by cornucopia, so feel free to use them as you usually would.

So, what else can we do with those annotations? The grammar can be summed up as:

<NAME> (<PARAMS>) <RETURN_TYPE> <QUANTIFIER>

In the next sections we'll explore a bit more what these options mean and what you can do with them. The regexp-esque notation used in this section to describe the grammar is for illustrative purposes only, The full grammar is available in the grammar.pest file.

Name

The name of the generated function. Has to be a valid Rust identifier.

Params

The parameters of the prepared statement, separated by commas, with an optional trailing comma.

The order in which parameters are given corresponds to the parameter number (e.g. the first parameter is $1 in the statement). Every PostgresQL parameter $i must have a corresponding parameter in the meta parameter list . If the param type is ambiguous, you can specify it using the syntax <IDENT> : <TYPE> where <TYPE> is a PostgresQL type supported by cornycopia. These are called override parameters. Otherwise, a param consisys simply of an identifier: the type is inferred from the prepared statements. These are called inferred parameters. Override parameters must come before inferred parameters. Again, we can sum this up as

<PARAMS> = <OVERRIDE>?, <INFERRED>? where <OVERRIDE> = <IDENT> : <TYPE> and <INFERRED> = <IDENT>.

Return type

There are two kinds of returns, implicit and explicit.

Implicit return

Implicit returns don't name the returned columns. The column types are inferred using prepared statements. To make a return implicit, simply omit it (you don't have to write anything).

Implicit returns are further categorized into void, scalar, and tuple types dependenging on the number of columns returned. For example,

• A query returning no column would result in ()
• A query returning a single TEXT column would result in String,
• A query returning a TEXT and a INTEGER would result in (String, i32)

Explicit return

Explicit returns give a name to the returned columns. The column types are inferred using prepared statements. To make a return explicit, list the returned column names inside brackets, in the same order as they are returned in the statement, separated by commas, with an optional trailing comma. There must be exactly as many names in the explicit return the as there are returned columns.

Each query that has an explicit return will generate a Rust struct to hold the query data. For example, this query

--! example_query() {name, country} *
SELECT Name, Country FROM Authors;

would result in this struct being generated

pub struct ExampleQuery {
    pub name: String,
    pub country: String
}

pub async fn authors(client: &Client) -> Result<Vec<ExampleQuery>, Error> {
    /* ....omitted for brevity... */
}

Quantifier

The quantifier indicates the expected number of rows to be returned by a query. If no quantifier is specified, the it is assumed that only one record is to be returned. Using * and ? (corresponding to the "zero or more" and "zero or one" quantifiers) will wrap the resulting rust type in a Vec and Option respectively. To sum it up:

• no quantifier results in T
• * results in Vec<T>
• ? results in Option<T>

Transactions

Cornucopia actually generates two versions of your queries, one that accepts a regular client, while the other version (named with a *_tx suffix) accepts a transaction.

Supported types

PostgrsQL type	Rust type
bool, boolean	bool
char, character	i8
smallint, int2, smallserial, serial2	i16
int, int4, serial, serial4	i32
bigint, int8, bigserial, serial8	i64
real, float4	f32
double precision, float8	f64
text	String
varchar	String
bytea	Vec
timestamp without time zone, timestamp	time::PrimitiveDateTime
timestamp with time zone, timestamptz	time::OffsetDateTime
date	time::Date
time	time::Time
json	serde_json::Value
jsonb	serde_json::Value
uuid	uuid::Uuid

License

Licensed under the MIT license.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.