query_builder.rs

//! Runtime query-builder API.

use std::fmt::Display;
use std::fmt::Write;
use std::marker::PhantomData;

use crate::arguments::{Arguments, IntoArguments};
use crate::database::{Database, HasArguments};
use crate::encode::Encode;
use crate::from_row::FromRow;
use crate::query::Query;
use crate::query_as::QueryAs;
use crate::query_scalar::QueryScalar;
use crate::types::Type;
use crate::Either;

/// A builder type for constructing queries at runtime.
///
/// See [`.push_values()`][Self::push_values] for an example of building a bulk `INSERT` statement.
/// Note, however, that with Postgres you can get much better performance by using arrays
/// and `UNNEST()`. [See our FAQ] for details.
///
/// [See our FAQ]: https://github.com/launchbadge/sqlx/blob/master/FAQ.md#how-can-i-bind-an-array-to-a-values-clause-how-can-i-do-bulk-inserts
pub struct QueryBuilder<'args, DB>
where
    DB: Database,
{
    query: String,
    init_len: usize,
    arguments: Option<<DB as HasArguments<'args>>::Arguments>,
}

impl<'args, DB: Database> Default for QueryBuilder<'args, DB> {
    fn default() -> Self {
        QueryBuilder {
            init_len: 0,
            query: String::default(),
            arguments: Some(Default::default()),
        }
    }
}

impl<'args, DB: Database> QueryBuilder<'args, DB>
where
    DB: Database,
{
    // `init` is provided because a query will almost always start with a constant fragment
    // such as `INSERT INTO ...` or `SELECT ...`, etc.
    /// Start building a query with an initial SQL fragment, which may be an empty string.
    pub fn new(init: impl Into<String>) -> Self
    where
        <DB as HasArguments<'args>>::Arguments: Default,
    {
        let init = init.into();

        QueryBuilder {
            init_len: init.len(),
            query: init,
            arguments: Some(Default::default()),
        }
    }

    /// Construct a `QueryBuilder` with existing SQL and arguments.
    ///
    /// ### Note
    /// This does *not* check if `arguments` is valid for the given SQL.
    pub fn with_arguments<A>(init: impl Into<String>, arguments: A) -> Self
    where
        DB: Database,
        A: IntoArguments<'args, DB>,
    {
        let init = init.into();

        QueryBuilder {
            init_len: init.len(),
            query: init,
            arguments: Some(arguments.into_arguments()),
        }
    }

    #[inline]
    fn sanity_check(&self) {
        assert!(
            self.arguments.is_some(),
            "QueryBuilder must be reset before reuse after `.build()`"
        );
    }

    /// Append a SQL fragment to the query.
    ///
    /// May be a string or anything that implements `Display`.
    /// You can also use `format_args!()` here to push a formatted string without an intermediate
    /// allocation.
    ///
    /// ### Warning: Beware SQL Injection Vulnerabilities and Untrusted Input!
    /// You should *not* use this to insert input directly into the query from an untrusted user as
    /// this can be used by an attacker to extract sensitive data or take over your database.
    ///
    /// Security breaches due to SQL injection can cost your organization a lot of money from
    /// damage control and lost clients, betray the trust of your users in your system, and are just
    /// plain embarrassing. If you are unfamiliar with the threat that SQL injection imposes, you
    /// should take some time to learn more about it before proceeding:
    ///
    /// * [SQL Injection on OWASP.org](https://owasp.org/www-community/attacks/SQL_Injection)
    /// * [SQL Injection on Wikipedia](https://en.wikipedia.org/wiki/SQL_injection)
    ///     * See "Examples" for notable instances of security breaches due to SQL injection.
    ///
    /// This method does *not* perform sanitization. Instead, you should use
    /// [`.push_bind()`][Self::push_bind] which inserts a placeholder into the query and then
    /// sends the possibly untrustworthy value separately (called a "bind argument") so that it
    /// cannot be misinterpreted by the database server.
    ///
    /// Note that you should still at least have some sort of sanity checks on the values you're
    /// sending as that's just good practice and prevent other types of attacks against your system,
    /// e.g. check that strings aren't too long, numbers are within expected ranges, etc.
    pub fn push(&mut self, sql: impl Display) -> &mut Self {
        self.sanity_check();

        write!(self.query, "{sql}").expect("error formatting `sql`");

        self
    }

    /// Push a bind argument placeholder (`?` or `$N` for Postgres) and bind a value to it.
    ///
    /// ### Note: Database-specific Limits
    /// Note that every database has a practical limit on the number of bind parameters
    /// you can add to a single query. This varies by database.
    ///
    /// While you should consult the manual of your specific database version and/or current
    /// configuration for the exact value as it may be different than listed here,
    /// the defaults for supported databases as of writing are as follows:
    ///
    /// * Postgres and MySQL: 65535
    ///     * You may find sources that state that Postgres has a limit of 32767,
    ///       but that is a misinterpretation of the specification by the JDBC driver implementation
    ///       as discussed in [this Github issue][postgres-limit-issue]. Postgres itself
    ///       asserts that the number of parameters is in the range `[0, 65535)`.
    /// * SQLite: 32766 (configurable by [`SQLITE_LIMIT_VARIABLE_NUMBER`])
    ///     * SQLite prior to 3.32.0: 999
    /// * MSSQL: 2100
    ///
    /// Exceeding these limits may panic (as a sanity check) or trigger a database error at runtime
    /// depending on the implementation.
    ///
    /// [`SQLITE_LIMIT_VARIABLE_NUMBER`]: https://www.sqlite.org/limits.html#max_variable_number
    /// [postgres-limit-issue]: https://github.com/launchbadge/sqlx/issues/671#issuecomment-687043510
    pub fn push_bind<T>(&mut self, value: T) -> &mut Self
    where
        T: 'args + Encode<'args, DB> + Send + Type<DB>,
    {
        self.sanity_check();

        let arguments = self
            .arguments
            .as_mut()
            .expect("BUG: Arguments taken already");
        arguments.add(value);

        arguments
            .format_placeholder(&mut self.query)
            .expect("error in format_placeholder");

        self
    }

    /// Start a list separated by `separator`.
    ///
    /// The returned type exposes identical [`.push()`][Separated::push] and
    /// [`.push_bind()`][Separated::push_bind] methods which push `separator` to the query
    /// before their normal behavior. [`.push_unseparated()`][Separated::push_unseparated] and [`.push_bind_unseparated()`][Separated::push_bind_unseparated] are also
    /// provided to push a SQL fragment without the separator.
    ///
    /// ```rust
    /// # #[cfg(feature = "mysql")] {
    /// use sqlx::{Execute, MySql, QueryBuilder};
    /// let foods = vec!["pizza".to_string(), "chips".to_string()];
    /// let mut query_builder: QueryBuilder<MySql> = QueryBuilder::new(
    ///     "SELECT * from food where name in ("
    /// );
    /// // One element vector is handled correctly but an empty vector
    /// // would cause a sql syntax error
    /// let mut separated = query_builder.separated(", ");
    /// for value_type in foods.iter() {
    ///   separated.push_bind(value_type);
    /// }
    /// separated.push_unseparated(") ");
    ///
    /// let mut query = query_builder.build();
    /// let sql = query.sql();
    /// assert!(sql.ends_with("in (?, ?) "));
    /// # }
    /// ```

    pub fn separated<'qb, Sep>(&'qb mut self, separator: Sep) -> Separated<'qb, 'args, DB, Sep>
    where
        'args: 'qb,
        Sep: Display,
    {
        self.sanity_check();

        Separated {
            query_builder: self,
            separator,
            push_separator: false,
        }
    }

    // Most of the `QueryBuilder` API is purposefully very low-level but this was a commonly
    // requested use-case so it made sense to support.
    /// Push a `VALUES` clause where each item in `tuples` represents a tuple/row in the clause.
    ///
    /// This can be used to construct a bulk `INSERT` statement, although keep in mind that all
    /// databases have some practical limit on the number of bind arguments in a single query.
    /// See [`.push_bind()`][Self::push_bind] for details.
    ///
    /// To be safe, you can do `tuples.into_iter().take(N)` where `N` is the limit for your database
    /// divided by the number of fields in each tuple; since integer division always rounds down,
    /// this will ensure that you don't exceed the limit.
    ///
    /// ### Notes
    ///
    /// If `tuples` is empty, this will likely produce a syntactically invalid query as `VALUES`
    /// generally expects to be followed by at least 1 tuple.
    ///
    /// If `tuples` can have many different lengths, you may want to call
    /// [`.persistent(false)`][Query::persistent] after [`.build()`][Self::build] to avoid
    /// filling up the connection's prepared statement cache.
    ///
    /// Because the `Arguments` API has a lifetime that must live longer than `Self`, you cannot
    /// bind by-reference from an iterator unless that iterator yields references that live
    /// longer than `Self`, even if the specific `Arguments` implementation doesn't actually
    /// borrow the values (like `MySqlArguments` and `PgArguments` immediately encode the arguments
    /// and don't borrow them past the `.add()` call).
    ///
    /// So basically, if you want to bind by-reference you need an iterator that yields references,
    /// e.g. if you have values in a `Vec` you can do `.iter()` instead of `.into_iter()`. The
    /// example below uses an iterator that creates values on the fly
    /// and so cannot bind by-reference.
    ///
    /// ### Example (MySQL)
    ///
    /// ```rust
    /// # #[cfg(feature = "mysql")]
    /// # {
    /// use sqlx::{Execute, MySql, QueryBuilder};
    ///
    /// struct User {
    ///     id: i32,
    ///     username: String,
    ///     email: String,
    ///     password: String,
    /// }
    ///
    /// // The number of parameters in MySQL must fit in a `u16`.
    /// const BIND_LIMIT: usize = 65535;
    ///
    /// // This would normally produce values forever!
    /// let users = (0..).map(|i| User {
    ///     id: i,
    ///     username: format!("test_user_{i}"),
    ///     email: format!("test-user-{i}@example.com"),
    ///     password: format!("Test!User@Password#{i}"),
    /// });
    ///
    /// let mut query_builder: QueryBuilder<MySql> = QueryBuilder::new(
    ///     // Note the trailing space; most calls to `QueryBuilder` don't automatically insert
    ///     // spaces as that might interfere with identifiers or quoted strings where exact
    ///     // values may matter.
    ///     "INSERT INTO users(id, username, email, password) "
    /// );
    ///
    /// // Note that `.into_iter()` wasn't needed here since `users` is already an iterator.
    /// query_builder.push_values(users.take(BIND_LIMIT / 4), |mut b, user| {
    ///     // If you wanted to bind these by-reference instead of by-value,
    ///     // you'd need an iterator that yields references that live as long as `query_builder`,
    ///     // e.g. collect it to a `Vec` first.
    ///     b.push_bind(user.id)
    ///         .push_bind(user.username)
    ///         .push_bind(user.email)
    ///         .push_bind(user.password);
    /// });
    ///
    /// let mut query = query_builder.build();
    ///
    /// // You can then call `query.execute()`, `.fetch_one()`, `.fetch_all()`, etc.
    /// // For the sake of demonstration though, we're just going to assert the contents
    /// // of the query.
    ///
    /// // These are methods of the `Execute` trait, not normally meant to be called in user code.
    /// let sql = query.sql();
    /// let arguments = query.take_arguments().unwrap();
    ///
    /// assert!(sql.starts_with(
    ///     "INSERT INTO users(id, username, email, password) VALUES (?, ?, ?, ?), (?, ?, ?, ?)"
    /// ));
    ///
    /// assert!(sql.ends_with("(?, ?, ?, ?)"));
    ///
    /// // Not a normally exposed function, only used for this doctest.
    /// // 65535 / 4 = 16383 (rounded down)
    /// // 16383 * 4 = 65532
    /// assert_eq!(arguments.len(), 65532);
    /// # }
    /// ```
    pub fn push_values<I, F>(&mut self, tuples: I, mut push_tuple: F) -> &mut Self
    where
        I: IntoIterator,
        F: FnMut(Separated<'_, 'args, DB, &'static str>, I::Item),
    {
        self.sanity_check();

        self.push("VALUES ");

        let mut separated = self.separated(", ");

        for tuple in tuples {
            separated.push("(");

            // use a `Separated` with a separate (hah) internal state
            push_tuple(separated.query_builder.separated(", "), tuple);

            separated.push_unseparated(")");
        }

        separated.query_builder
    }

    /// Creates `((a, b), (..)` statements, from `tuples`.
    ///
    /// This can be used to construct a bulk `SELECT` statement like this:
    /// ```sql
    /// SELECT * FROM users WHERE (id, username) IN ((1, "test_user_1"), (2, "test_user_2"))
    /// ```
    ///
    /// Although keep in mind that all
    /// databases have some practical limit on the number of bind arguments in a single query.
    /// See [`.push_bind()`][Self::push_bind] for details.
    ///
    /// To be safe, you can do `tuples.into_iter().take(N)` where `N` is the limit for your database
    /// divided by the number of fields in each tuple; since integer division always rounds down,
    /// this will ensure that you don't exceed the limit.
    ///
    /// ### Notes
    ///
    /// If `tuples` is empty, this will likely produce a syntactically invalid query
    ///
    /// ### Example (MySQL)
    ///
    /// ```rust
    /// # #[cfg(feature = "mysql")]
    /// # {
    /// use sqlx::{Execute, MySql, QueryBuilder};
    ///
    /// struct User {
    ///     id: i32,
    ///     username: String,
    ///     email: String,
    ///     password: String,
    /// }
    ///
    /// // The number of parameters in MySQL must fit in a `u16`.
    /// const BIND_LIMIT: usize = 65535;
    ///
    /// // This would normally produce values forever!
    /// let users = (0..).map(|i| User {
    ///     id: i,
    ///     username: format!("test_user_{i}"),
    ///     email: format!("test-user-{i}@example.com"),
    ///     password: format!("Test!User@Password#{i}"),
    /// });
    ///
    /// let mut query_builder: QueryBuilder<MySql> = QueryBuilder::new(
    ///     // Note the trailing space; most calls to `QueryBuilder` don't automatically insert
    ///     // spaces as that might interfere with identifiers or quoted strings where exact
    ///     // values may matter.
    ///     "SELECT * FROM users WHERE (id, username, email, password) in"
    /// );
    ///
    /// // Note that `.into_iter()` wasn't needed here since `users` is already an iterator.
    /// query_builder.push_tuples(users.take(BIND_LIMIT / 4), |mut b, user| {
    ///     // If you wanted to bind these by-reference instead of by-value,
    ///     // you'd need an iterator that yields references that live as long as `query_builder`,
    ///     // e.g. collect it to a `Vec` first.
    ///     b.push_bind(user.id)
    ///         .push_bind(user.username)
    ///         .push_bind(user.email)
    ///         .push_bind(user.password);
    /// });
    ///
    /// let mut query = query_builder.build();
    ///
    /// // You can then call `query.execute()`, `.fetch_one()`, `.fetch_all()`, etc.
    /// // For the sake of demonstration though, we're just going to assert the contents
    /// // of the query.
    ///
    /// // These are methods of the `Execute` trait, not normally meant to be called in user code.
    /// let sql = query.sql();
    /// let arguments = query.take_arguments().unwrap();
    ///
    /// assert!(sql.starts_with(
    ///     "SELECT * FROM users WHERE (id, username, email, password) in ((?, ?, ?, ?), (?, ?, ?, ?), "
    /// ));
    ///
    /// assert!(sql.ends_with("(?, ?, ?, ?)) "));
    ///
    /// // Not a normally exposed function, only used for this doctest.
    /// // 65535 / 4 = 16383 (rounded down)
    /// // 16383 * 4 = 65532
    /// assert_eq!(arguments.len(), 65532);
    /// }
    /// ```
    pub fn push_tuples<I, F>(&mut self, tuples: I, mut push_tuple: F) -> &mut Self
    where
        I: IntoIterator,
        F: FnMut(Separated<'_, 'args, DB, &'static str>, I::Item),
    {
        self.sanity_check();

        self.push(" (");

        let mut separated = self.separated(", ");

        for tuple in tuples {
            separated.push("(");

            push_tuple(separated.query_builder.separated(", "), tuple);

            separated.push_unseparated(")");
        }
        separated.push_unseparated(") ");

        separated.query_builder
    }

    /// Produce an executable query from this builder.
    ///
    /// ### Note: Query is not Checked
    /// It is your responsibility to ensure that you produce a syntactically correct query here,
    /// this API has no way to check it for you.
    ///
    /// ### Note: Reuse
    /// You can reuse this builder afterwards to amortize the allocation overhead of the query
    /// string, however you must call [`.reset()`][Self::reset] first, which returns `Self`
    /// to the state it was in immediately after [`new()`][Self::new].
    ///
    /// Calling any other method but `.reset()` after `.build()` will panic for sanity reasons.
    pub fn build(&mut self) -> Query<'_, DB, <DB as HasArguments<'args>>::Arguments> {
        self.sanity_check();

        Query {
            statement: Either::Left(&self.query),
            arguments: self.arguments.take(),
            database: PhantomData,
            persistent: true,
        }
    }

    /// Produce an executable query from this builder.
    ///
    /// ### Note: Query is not Checked
    /// It is your responsibility to ensure that you produce a syntactically correct query here,
    /// this API has no way to check it for you.
    ///
    /// ### Note: Reuse
    /// You can reuse this builder afterwards to amortize the allocation overhead of the query
    /// string, however you must call [`.reset()`][Self::reset] first, which returns `Self`
    /// to the state it was in immediately after [`new()`][Self::new].
    ///
    /// Calling any other method but `.reset()` after `.build()` will panic for sanity reasons.
    pub fn build_query_as<'q, T: FromRow<'q, DB::Row>>(
        &'q mut self,
    ) -> QueryAs<'q, DB, T, <DB as HasArguments<'args>>::Arguments> {
        QueryAs {
            inner: self.build(),
            output: PhantomData,
        }
    }

    /// Produce an executable query from this builder.
    ///
    /// ### Note: Query is not Checked
    /// It is your responsibility to ensure that you produce a syntactically correct query here,
    /// this API has no way to check it for you.
    ///
    /// ### Note: Reuse
    /// You can reuse this builder afterwards to amortize the allocation overhead of the query
    /// string, however you must call [`.reset()`][Self::reset] first, which returns `Self`
    /// to the state it was in immediately after [`new()`][Self::new].
    ///
    /// Calling any other method but `.reset()` after `.build()` will panic for sanity reasons.
    pub fn build_query_scalar<'q, T>(
        &'q mut self,
    ) -> QueryScalar<'q, DB, T, <DB as HasArguments<'args>>::Arguments>
    where
        DB: Database,
        (T,): for<'r> FromRow<'r, DB::Row>,
    {
        QueryScalar {
            inner: self.build_query_as(),
        }
    }

    /// Reset this `QueryBuilder` back to its initial state.
    ///
    /// The query is truncated to the initial fragment provided to [`new()`][Self::new] and
    /// the bind arguments are reset.
    pub fn reset(&mut self) -> &mut Self {
        self.query.truncate(self.init_len);
        self.arguments = Some(Default::default());

        self
    }

    /// Get the current build SQL; **note**: may not be syntactically correct.
    pub fn sql(&self) -> &str {
        &self.query
    }

    /// Deconstruct this `QueryBuilder`, returning the built SQL. May not be syntactically correct.
    pub fn into_sql(self) -> String {
        self.query
    }
}

/// A wrapper around `QueryBuilder` for creating comma(or other token)-separated lists.
///
/// See [`QueryBuilder::separated()`] for details.
#[allow(explicit_outlives_requirements)]
pub struct Separated<'qb, 'args: 'qb, DB, Sep>
where
    DB: Database,
{
    query_builder: &'qb mut QueryBuilder<'args, DB>,
    separator: Sep,
    push_separator: bool,
}

impl<'qb, 'args: 'qb, DB, Sep> Separated<'qb, 'args, DB, Sep>
where
    DB: Database,
    Sep: Display,
{
    /// Push the separator if applicable, and then the given SQL fragment.
    ///
    /// See [`QueryBuilder::push()`] for details.
    pub fn push(&mut self, sql: impl Display) -> &mut Self {
        if self.push_separator {
            self.query_builder
                .push(format_args!("{}{}", self.separator, sql));
        } else {
            self.query_builder.push(sql);
            self.push_separator = true;
        }

        self
    }

    /// Push a SQL fragment without a separator.
    ///
    /// Simply calls [`QueryBuilder::push()`] directly.
    pub fn push_unseparated(&mut self, sql: impl Display) -> &mut Self {
        self.query_builder.push(sql);
        self
    }

    /// Push the separator if applicable, then append a bind argument.
    ///
    /// See [`QueryBuilder::push_bind()`] for details.
    pub fn push_bind<T>(&mut self, value: T) -> &mut Self
    where
        T: 'args + Encode<'args, DB> + Send + Type<DB>,
    {
        if self.push_separator {
            self.query_builder.push(&self.separator);
        }

        self.query_builder.push_bind(value);
        self.push_separator = true;

        self
    }

    /// Push a bind argument placeholder (`?` or `$N` for Postgres) and bind a value to it
    /// without a separator.
    ///
    /// Simply calls [`QueryBuilder::push_bind()`] directly.
    pub fn push_bind_unseparated<T>(&mut self, value: T) -> &mut Self
    where
        T: 'args + Encode<'args, DB> + Send + Type<DB>,
    {
        self.query_builder.push_bind(value);
        self
    }
}

#[cfg(all(test, feature = "postgres"))]
mod test {
    use crate::postgres::Postgres;

    use super::*;

    #[test]
    fn test_new() {
        let qb: QueryBuilder<'_, Postgres> = QueryBuilder::new("SELECT * FROM users");
        assert_eq!(qb.query, "SELECT * FROM users");
    }

    #[test]
    fn test_push() {
        let mut qb: QueryBuilder<'_, Postgres> = QueryBuilder::new("SELECT * FROM users");
        let second_line = " WHERE last_name LIKE '[A-N]%;";
        qb.push(second_line);

        assert_eq!(
            qb.query,
            "SELECT * FROM users WHERE last_name LIKE '[A-N]%;".to_string(),
        );
    }

    #[test]
    #[should_panic]
    fn test_push_panics_when_no_arguments() {
        let mut qb: QueryBuilder<'_, Postgres> = QueryBuilder::new("SELECT * FROM users;");
        qb.arguments = None;

        qb.push("SELECT * FROM users;");
    }

    #[test]
    fn test_push_bind() {
        let mut qb: QueryBuilder<'_, Postgres> =
            QueryBuilder::new("SELECT * FROM users WHERE id = ");

        qb.push_bind(42i32)
            .push(" OR membership_level = ")
            .push_bind(3i32);

        assert_eq!(
            qb.query,
            "SELECT * FROM users WHERE id = $1 OR membership_level = $2"
        );
    }

    #[test]
    fn test_build() {
        let mut qb: QueryBuilder<'_, Postgres> = QueryBuilder::new("SELECT * FROM users");

        qb.push(" WHERE id = ").push_bind(42i32);
        let query = qb.build();

        assert_eq!(
            query.statement.unwrap_left(),
            "SELECT * FROM users WHERE id = $1"
        );
        assert_eq!(query.persistent, true);
    }

    #[test]
    fn test_reset() {
        let mut qb: QueryBuilder<'_, Postgres> = QueryBuilder::new("");

        let _query = qb
            .push("SELECT * FROM users WHERE id = ")
            .push_bind(42i32)
            .build();

        qb.reset();

        assert_eq!(qb.query, "");
    }

    #[test]
    fn test_query_builder_reuse() {
        let mut qb: QueryBuilder<'_, Postgres> = QueryBuilder::new("");

        let _query = qb
            .push("SELECT * FROM users WHERE id = ")
            .push_bind(42i32)
            .build();

        qb.reset();

        let query = qb.push("SELECT * FROM users WHERE id = 99").build();

        assert_eq!(
            query.statement.unwrap_left(),
            "SELECT * FROM users WHERE id = 99"
        );
    }

    #[test]
    fn test_query_builder_with_args() {
        let mut qb: QueryBuilder<'_, Postgres> = QueryBuilder::new("");

        let query = qb
            .push("SELECT * FROM users WHERE id = ")
            .push_bind(42i32)
            .build();

        let mut qb: QueryBuilder<'_, Postgres> =
            QueryBuilder::new_with(query.sql(), query.take_arguments());
        let query = qb.push("OR membership_level =").push_bind(3i32).build();

        assert_eq!(
            query.sql(),
            "SELECT * FROM users WHERE id = $1 OR membership_level = $2"
        );
    }
}