Struct UncheckedTransaction

pub struct UncheckedTransaction<'conn> {
    pub conn: &'conn Connection,
    pub started_at: Instant,
    pub finished: bool,
}

rusqlite, in an attempt to save us from ourselves, needs a mutable ref to a connection to start a transaction. That is a bit of a PITA in some cases, so we offer this as an alternative - but the responsibility of ensuring there are no concurrent transactions is on our head.

This is very similar to the rusqlite Transaction - it doesn’t prevent against nested transactions but does allow you to use an immutable Connection.

FIXME: This currently won’t actually be used most of the time, because rusqlite added [Connection::unchecked_transaction] (and Transaction::new_unchecked, which can be used to reimplement unchecked_transaction_imm), which will be preferred in a call to c.unchecked_transaction(), because inherent methods have precedence over methods on extension traits. The exception here is that this will still be used by code which takes &impl ConnExt (I believe it would also be used if you attempted to call unchecked_transaction() on a non-Connection that implements ConnExt, such as a Safepoint, UncheckedTransaction, or Transaction itself, but such code is clearly broken, so is not worth considering).

The difference is that rusqlite’s version returns a normal rusqlite::Transaction, rather than the UncheckedTransaction from this crate. Aside from type’s name and location (and the fact that rusqlite’s detects slightly more misuse at compile time, and has more features), the main difference is: rusqlite’s does not track when a transaction began, which unfortunately seems to be used by the coop-transaction management in places in some fashion.

There are at least two options for how to fix this:

1. Decide we don’t need this version, and delete it, and moving the transaction timing into the coop-transaction code directly (or something like this).
2. Decide this difference is important, and rename ConnExt::unchecked_transaction to something like ConnExt::transaction_unchecked.

Fields

conn: &'conn Connection

started_at: Instant

finished: bool

Implementations

impl<'conn> UncheckedTransaction<'conn>

pub fn new( conn: &'conn Connection, behavior: TransactionBehavior, ) -> SqlResult<Self>

Begin a new unchecked transaction. Cannot be nested, but this is not enforced by Rust (hence ‘unchecked’) - however, it is enforced by SQLite; use a rusqlite savepoint for nested transactions.

pub fn commit(self) -> SqlResult<()>

Consumes and commits an unchecked transaction.

pub fn rollback(self) -> SqlResult<()>

Consumes and rolls back an unchecked transaction.

Methods from Deref<Target = Connection>

pub fn busy_timeout(&self, timeout: Duration) -> Result<(), Error>

Set a busy handler that sleeps for a specified amount of time when a table is locked. The handler will sleep multiple times until at least “ms” milliseconds of sleeping have accumulated.

Calling this routine with an argument equal to zero turns off all busy handlers.

There can only be a single busy handler for a particular database connection at any given moment. If another busy handler was defined (using busy_handler) prior to calling this routine, that other busy handler is cleared.

Newly created connections currently have a default busy timeout of 5000ms, but this may be subject to change.

pub fn busy_handler( &self, callback: Option<fn(_: i32) -> bool>, ) -> Result<(), Error>

Register a callback to handle SQLITE_BUSY errors.

If the busy callback is None, then SQLITE_BUSY is returned immediately upon encountering the lock. The argument to the busy handler callback is the number of times that the busy handler has been invoked previously for the same locking event. If the busy callback returns false, then no additional attempts are made to access the database and SQLITE_BUSY is returned to the application. If the callback returns true, then another attempt is made to access the database and the cycle repeats.

There can only be a single busy handler defined for each database connection. Setting a new busy handler clears any previously set handler. Note that calling busy_timeout() or evaluating PRAGMA busy_timeout=N will change the busy handler and thus clear any previously set busy handler.

Newly created connections default to a busy_timeout() handler with a timeout of 5000ms, although this is subject to change.

pub fn prepare_cached(&self, sql: &str) -> Result<CachedStatement<'_>, Error>

Prepare a SQL statement for execution, returning a previously prepared (but not currently in-use) statement if one is available. The returned statement will be cached for reuse by future calls to prepare_cached once it is dropped.

fn insert_new_people(conn: &Connection) -> Result<()> {
    {
        let mut stmt = conn.prepare_cached("INSERT INTO People (name) VALUES (?1)")?;
        stmt.execute(["Joe Smith"])?;
    }
    {
        // This will return the same underlying SQLite statement handle without
        // having to prepare it again.
        let mut stmt = conn.prepare_cached("INSERT INTO People (name) VALUES (?1)")?;
        stmt.execute(["Bob Jones"])?;
    }
    Ok(())
}

Failure

Will return Err if sql cannot be converted to a C-compatible string or if the underlying SQLite call fails.

pub fn set_prepared_statement_cache_capacity(&self, capacity: usize)

Set the maximum number of cached prepared statements this connection will hold. By default, a connection will hold a relatively small number of cached statements. If you need more, or know that you will not use cached statements, you can set the capacity manually using this method.

pub fn flush_prepared_statement_cache(&self)

Remove/finalize all prepared statements currently in the cache.

pub fn db_config(&self, config: DbConfig) -> Result<bool, Error>

Returns the current value of a config.

• SQLITE_DBCONFIG_ENABLE_FKEY: return false or true to indicate whether FK enforcement is off or on
• SQLITE_DBCONFIG_ENABLE_TRIGGER: return false or true to indicate whether triggers are disabled or enabled
• SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER: return false or true to indicate whether fts3_tokenizer are disabled or enabled
• SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE: return false to indicate checkpoints-on-close are not disabled or true if they are
• SQLITE_DBCONFIG_ENABLE_QPSG: return false or true to indicate whether the QPSG is disabled or enabled
• SQLITE_DBCONFIG_TRIGGER_EQP: return false to indicate output-for-trigger are not disabled or true if it is

pub fn set_db_config( &self, config: DbConfig, new_val: bool, ) -> Result<bool, Error>

Make configuration changes to a database connection

• SQLITE_DBCONFIG_ENABLE_FKEY: false to disable FK enforcement, true to enable FK enforcement
• SQLITE_DBCONFIG_ENABLE_TRIGGER: false to disable triggers, true to enable triggers
• SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER: false to disable fts3_tokenizer(), true to enable fts3_tokenizer()
• SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE: false (the default) to enable checkpoints-on-close, true to disable them
• SQLITE_DBCONFIG_ENABLE_QPSG: false to disable the QPSG, true to enable QPSG
• SQLITE_DBCONFIG_TRIGGER_EQP: false to disable output for trigger programs, true to enable it

pub fn create_scalar_function<F, T>( &self, fn_name: &str, n_arg: i32, flags: FunctionFlags, x_func: F, ) -> Result<(), Error>

where F: FnMut(&Context<'_>) -> Result<T, Error> + Send + UnwindSafe + 'static, T: SqlFnOutput,

Attach a user-defined scalar function to this database connection.

fn_name is the name the function will be accessible from SQL. n_arg is the number of arguments to the function. Use -1 for a variable number. If the function always returns the same value given the same input, deterministic should be true.

The function will remain available until the connection is closed or until it is explicitly removed via remove_function.

Example

fn scalar_function_example(db: Connection) -> Result<()> {
    db.create_scalar_function(
        "halve",
        1,
        FunctionFlags::SQLITE_UTF8 | FunctionFlags::SQLITE_DETERMINISTIC,
        |ctx| {
            let value = ctx.get::<f64>(0)?;
            Ok(value / 2f64)
        },
    )?;

    let six_halved: f64 = db.query_row("SELECT halve(6)", [], |r| r.get(0))?;
    assert_eq!(six_halved, 3f64);
    Ok(())
}

Failure

Will return Err if the function could not be attached to the connection.

pub fn create_aggregate_function<A, D, T>( &self, fn_name: &str, n_arg: i32, flags: FunctionFlags, aggr: D, ) -> Result<(), Error>

where A: RefUnwindSafe + UnwindSafe, D: Aggregate<A, T> + 'static, T: SqlFnOutput,

Attach a user-defined aggregate function to this database connection.

Failure

Will return Err if the function could not be attached to the connection.

pub fn create_window_function<A, W, T>( &self, fn_name: &str, n_arg: i32, flags: FunctionFlags, aggr: W, ) -> Result<(), Error>

where A: RefUnwindSafe + UnwindSafe, W: WindowAggregate<A, T> + 'static, T: SqlFnOutput,

Attach a user-defined aggregate window function to this database connection.

See https://sqlite.org/windowfunctions.html#udfwinfunc for more information.

pub fn remove_function(&self, fn_name: &str, n_arg: i32) -> Result<(), Error>

Removes a user-defined function from this database connection.

fn_name and n_arg should match the name and number of arguments given to create_scalar_function or create_aggregate_function.

Failure

Will return Err if the function could not be removed.

pub fn limit(&self, limit: Limit) -> i32

Returns the current value of a [Limit].

pub fn set_limit(&self, limit: Limit, new_val: i32) -> i32

Changes the [Limit] to new_val, returning the prior value of the limit.

pub fn pragma_query_value<T, F>( &self, schema_name: Option<DatabaseName<'_>>, pragma_name: &str, f: F, ) -> Result<T, Error>

where F: FnOnce(&Row<'_>) -> Result<T, Error>,

Query the current value of pragma_name.

Some pragmas will return multiple rows/values which cannot be retrieved with this method.

Prefer PRAGMA function introduced in SQLite 3.20: SELECT user_version FROM pragma_user_version;

pub fn pragma_query<F>( &self, schema_name: Option<DatabaseName<'_>>, pragma_name: &str, f: F, ) -> Result<(), Error>

where F: FnMut(&Row<'_>) -> Result<(), Error>,

Query the current rows/values of pragma_name.

Prefer PRAGMA function introduced in SQLite 3.20: SELECT * FROM pragma_collation_list;

pub fn pragma<F, V>( &self, schema_name: Option<DatabaseName<'_>>, pragma_name: &str, pragma_value: V, f: F, ) -> Result<(), Error>

where F: FnMut(&Row<'_>) -> Result<(), Error>, V: ToSql,

Query the current value(s) of pragma_name associated to pragma_value.

This method can be used with query-only pragmas which need an argument (e.g. table_info('one_tbl')) or pragmas which returns value(s) (e.g. integrity_check).

Prefer PRAGMA function introduced in SQLite 3.20: SELECT * FROM pragma_table_info(?1);

pub fn pragma_update<V>( &self, schema_name: Option<DatabaseName<'_>>, pragma_name: &str, pragma_value: V, ) -> Result<(), Error>

where V: ToSql,

Set a new value to pragma_name.

Some pragmas will return the updated value which cannot be retrieved with this method.

pub fn pragma_update_and_check<F, T, V>( &self, schema_name: Option<DatabaseName<'_>>, pragma_name: &str, pragma_value: V, f: F, ) -> Result<T, Error>

where F: FnOnce(&Row<'_>) -> Result<T, Error>, V: ToSql,

Set a new value to pragma_name and return the updated value.

Only few pragmas automatically return the updated value.

pub fn unchecked_transaction(&self) -> Result<Transaction<'_>, Error>

Begin a new transaction with the default behavior (DEFERRED).

Attempt to open a nested transaction will result in a SQLite error. Connection::transaction prevents this at compile time by taking &mut self, but Connection::unchecked_transaction() may be used to defer the checking until runtime.

See [Connection::transaction] and [Transaction::new_unchecked] (which can be used if the default transaction behavior is undesirable).

Example

fn perform_queries(conn: Rc<Connection>) -> Result<()> {
    let tx = conn.unchecked_transaction()?;

    do_queries_part_1(&tx)?; // tx causes rollback if this fails
    do_queries_part_2(&tx)?; // tx causes rollback if this fails

    tx.commit()
}

Failure

Will return Err if the underlying SQLite call fails. The specific error returned if transactions are nested is currently unspecified.

pub fn transaction_state( &self, db_name: Option<DatabaseName<'_>>, ) -> Result<TransactionState, Error>

Determine the transaction state of a database

pub fn execute_batch(&self, sql: &str) -> Result<(), Error>

Convenience method to run multiple SQL statements (that cannot take any parameters).

Example

fn create_tables(conn: &Connection) -> Result<()> {
    conn.execute_batch(
        "BEGIN;
         CREATE TABLE foo(x INTEGER);
         CREATE TABLE bar(y TEXT);
         COMMIT;",
    )
}

Failure

Will return Err if sql cannot be converted to a C-compatible string or if the underlying SQLite call fails.

pub fn execute<P>(&self, sql: &str, params: P) -> Result<usize, Error>

where P: Params,

Convenience method to prepare and execute a single SQL statement.

On success, returns the number of rows that were changed or inserted or deleted (via sqlite3_changes).

Example

With positional params

fn update_rows(conn: &Connection) {
    match conn.execute("UPDATE foo SET bar = 'baz' WHERE qux = ?1", [1i32]) {
        Ok(updated) => println!("{} rows were updated", updated),
        Err(err) => println!("update failed: {}", err),
    }
}

With positional params of varying types

fn update_rows(conn: &Connection) {
    match conn.execute(
        "UPDATE foo SET bar = 'baz' WHERE qux = ?1 AND quux = ?2",
        params![1i32, 1.5f64],
    ) {
        Ok(updated) => println!("{} rows were updated", updated),
        Err(err) => println!("update failed: {}", err),
    }
}

With named params

fn insert(conn: &Connection) -> Result<usize> {
    conn.execute(
        "INSERT INTO test (name) VALUES (:name)",
        &[(":name", "one")],
    )
}

Failure

Will return Err if sql cannot be converted to a C-compatible string or if the underlying SQLite call fails.

pub fn path(&self) -> Option<&str>

Returns the path to the database file, if one exists and is known.

Returns Some("") for a temporary or in-memory database.

Note that in some cases PRAGMA database_list is likely to be more robust.

pub fn last_insert_rowid(&self) -> i64

Get the SQLite rowid of the most recent successful INSERT.

Uses sqlite3_last_insert_rowid under the hood.

pub fn query_row<T, P, F>(&self, sql: &str, params: P, f: F) -> Result<T, Error>

where P: Params, F: FnOnce(&Row<'_>) -> Result<T, Error>,

Convenience method to execute a query that is expected to return a single row.

Example

fn preferred_locale(conn: &Connection) -> Result<String> {
    conn.query_row(
        "SELECT value FROM preferences WHERE name='locale'",
        [],
        |row| row.get(0),
    )
}

If the query returns more than one row, all rows except the first are ignored.

Returns Err(QueryReturnedNoRows) if no results are returned. If the query truly is optional, you can call .optional() on the result of this to get a Result<Option<T>>.

Failure

Will return Err if sql cannot be converted to a C-compatible string or if the underlying SQLite call fails.

pub fn query_row_and_then<T, E, P, F>( &self, sql: &str, params: P, f: F, ) -> Result<T, E>

where P: Params, F: FnOnce(&Row<'_>) -> Result<T, E>, E: From<Error>,

Convenience method to execute a query that is expected to return a single row, and execute a mapping via f on that returned row with the possibility of failure. The Result type of f must implement std::convert::From<Error>.

Example

fn preferred_locale(conn: &Connection) -> Result<String> {
    conn.query_row_and_then(
        "SELECT value FROM preferences WHERE name='locale'",
        [],
        |row| row.get(0),
    )
}

If the query returns more than one row, all rows except the first are ignored.

Failure

Will return Err if sql cannot be converted to a C-compatible string or if the underlying SQLite call fails.

pub fn prepare(&self, sql: &str) -> Result<Statement<'_>, Error>

Prepare a SQL statement for execution.

Example

fn insert_new_people(conn: &Connection) -> Result<()> {
    let mut stmt = conn.prepare("INSERT INTO People (name) VALUES (?1)")?;
    stmt.execute(["Joe Smith"])?;
    stmt.execute(["Bob Jones"])?;
    Ok(())
}

Failure

Will return Err if sql cannot be converted to a C-compatible string or if the underlying SQLite call fails.

pub fn prepare_with_flags( &self, sql: &str, flags: PrepFlags, ) -> Result<Statement<'_>, Error>

Prepare a SQL statement for execution.

Failure

Will return Err if sql cannot be converted to a C-compatible string or if the underlying SQLite call fails.

pub unsafe fn load_extension_enable(&self) -> Result<(), Error>

Enable loading of SQLite extensions from both SQL queries and Rust.

You must call [Connection::load_extension_disable] when you’re finished loading extensions (failure to call it can lead to bad things, see “Safety”), so you should strongly consider using [LoadExtensionGuard] instead of this function, automatically disables extension loading when it goes out of scope.

Example

fn load_my_extension(conn: &Connection) -> Result<()> {
    // Safety: We fully trust the loaded extension and execute no untrusted SQL
    // while extension loading is enabled.
    unsafe {
        conn.load_extension_enable()?;
        let r = conn.load_extension("my/trusted/extension", None);
        conn.load_extension_disable()?;
        r
    }
}

Failure

Will return Err if the underlying SQLite call fails.

Safety

TLDR: Don’t execute any untrusted queries between this call and [Connection::load_extension_disable].

Perhaps surprisingly, this function does not only allow the use of [Connection::load_extension] from Rust, but it also allows SQL queries to perform the same operation. For example, in the period between load_extension_enable and load_extension_disable, the following operation will load and call some function in some dynamic library:

SELECT load_extension('why_is_this_possible.dll', 'dubious_func');

This means that while this is enabled a carefully crafted SQL query can be used to escalate a SQL injection attack into code execution.

Safely using this function requires that you trust all SQL queries run between when it is called, and when loading is disabled (by [Connection::load_extension_disable]).

pub fn load_extension_disable(&self) -> Result<(), Error>

Disable loading of SQLite extensions.

See [Connection::load_extension_enable] for an example.

Failure

Will return Err if the underlying SQLite call fails.

pub unsafe fn load_extension<P>( &self, dylib_path: P, entry_point: Option<&str>, ) -> Result<(), Error>

where P: AsRef<Path>,

Load the SQLite extension at dylib_path. dylib_path is passed through to sqlite3_load_extension, which may attempt OS-specific modifications if the file cannot be loaded directly (for example converting "some/ext" to "some/ext.so", "some\\ext.dll", …).

If entry_point is None, SQLite will attempt to find the entry point. If it is not None, the entry point will be passed through to sqlite3_load_extension.

Example

fn load_my_extension(conn: &Connection) -> Result<()> {
    // Safety: we don't execute any SQL statements while
    // extension loading is enabled.
    let _guard = unsafe { LoadExtensionGuard::new(conn)? };
    // Safety: `my_sqlite_extension` is highly trustworthy.
    unsafe { conn.load_extension("my_sqlite_extension", None) }
}

Failure

Will return Err if the underlying SQLite call fails.

Safety

This is equivalent to performing a dlopen/LoadLibrary on a shared library, and calling a function inside, and thus requires that you trust the library that you’re loading.

That is to say: to safely use this, the code in the extension must be sound, trusted, correctly use the SQLite APIs, and not contain any memory or thread safety errors.

pub unsafe fn handle(&self) -> *mut sqlite3

Get access to the underlying SQLite database connection handle.

Warning

You should not need to use this function. If you do need to, please open an issue on the rusqlite repository and describe your use case.

Safety

This function is unsafe because it gives you raw access to the SQLite connection, and what you do with it could impact the safety of this Connection.

pub fn get_interrupt_handle(&self) -> InterruptHandle

Get access to a handle that can be used to interrupt long running queries from another thread.

pub fn changes(&self) -> u64

Return the number of rows modified, inserted or deleted by the most recently completed INSERT, UPDATE or DELETE statement on the database connection.

See https://www.sqlite.org/c3ref/changes.html

pub fn is_autocommit(&self) -> bool

Test for auto-commit mode. Autocommit mode is on by default.

pub fn is_busy(&self) -> bool

Determine if all associated prepared statements have been reset.

pub fn cache_flush(&self) -> Result<(), Error>

Flush caches to disk mid-transaction

pub fn is_readonly(&self, db_name: DatabaseName<'_>) -> Result<bool, Error>

Determine if a database is read-only

Trait Implementations

impl ConnExt for UncheckedTransaction<'_>

fn conn(&self) -> &Connection

The method you need to implement to opt in to all of this.

fn set_pragma<T>(&self, pragma_name: &str, pragma_value: T) -> SqlResult<&Self>

where T: ToSql, Self: Sized,

Set the value of the pragma on the main database. Returns the same object, for chaining.

fn prepare_maybe_cached<'conn>( &'conn self, sql: &str, cache: bool, ) -> SqlResult<MaybeCached<'conn>>

Get a cached or uncached statement based on a flag.

fn execute_all(&self, stmts: &[&str]) -> SqlResult<()>

Execute all the provided statements.

fn execute_one(&self, stmt: &str) -> SqlResult<()>

Execute a single statement.

fn execute_cached<P: Params>(&self, sql: &str, params: P) -> SqlResult<usize>

Equivalent to Connection::execute but caches the statement so that subsequent calls to execute_cached will have improved performance.

fn query_one<T: FromSql>(&self, sql: &str) -> SqlResult<T>

Execute a query that returns a single result column, and return that result.

fn exists<P: Params>(&self, sql: &str, params: P) -> SqlResult<bool>

Return true if a query returns any rows

fn try_query_one<T: FromSql, P: Params>( &self, sql: &str, params: P, cache: bool, ) -> SqlResult<Option<T>>

where Self: Sized,

Execute a query that returns 0 or 1 result columns, returning None if there were no rows, or if the only result was NULL.

fn query_row_and_then_cachable<T, E, P, F>( &self, sql: &str, params: P, mapper: F, cache: bool, ) -> Result<T, E>

where Self: Sized, P: Params, E: From<Error>, F: FnOnce(&Row<'_>) -> Result<T, E>,

Equivalent to rusqlite::Connection::query_row_and_then but allows passing a flag to indicate that it’s cached.

fn query_rows_and_then<T, E, P, F>( &self, sql: &str, params: P, mapper: F, ) -> Result<Vec<T>, E>

where Self: Sized, P: Params, E: From<Error>, F: FnMut(&Row<'_>) -> Result<T, E>,

Helper for when you’d like to get a Vec<T> of all the rows returned by a query that takes named arguments. See also query_rows_and_then_cached.

fn query_rows_and_then_cached<T, E, P, F>( &self, sql: &str, params: P, mapper: F, ) -> Result<Vec<T>, E>

where Self: Sized, P: Params, E: From<Error>, F: FnMut(&Row<'_>) -> Result<T, E>,

Helper for when you’d like to get a Vec<T> of all the rows returned by a query that takes named arguments.

fn query_rows_into<Coll, T, E, P, F>( &self, sql: &str, params: P, mapper: F, ) -> Result<Coll, E>

where Self: Sized, E: From<Error>, F: FnMut(&Row<'_>) -> Result<T, E>, Coll: FromIterator<T>, P: Params,

Like query_rows_and_then_cachable, but works if you want a non-Vec as a result.

fn query_rows_into_cached<Coll, T, E, P, F>( &self, sql: &str, params: P, mapper: F, ) -> Result<Coll, E>

where Self: Sized, P: Params, E: From<Error>, F: FnMut(&Row<'_>) -> Result<T, E>, Coll: FromIterator<T>,

Same as query_rows_into, but caches the stmt if possible.

fn try_query_row<T, E, P, F>( &self, sql: &str, params: P, mapper: F, cache: bool, ) -> Result<Option<T>, E>

where Self: Sized, P: Params, E: From<Error>, F: FnOnce(&Row<'_>) -> Result<T, E>,

Like query_row_and_then_cacheable but returns None instead of erroring if no such row exists.

fn unchecked_transaction(&self) -> SqlResult<UncheckedTransaction<'_>>

Caveat: This won’t actually get used most of the time, and calls will usually invoke rusqlite’s method with the same name. See comment on UncheckedTransaction for details (generally you probably don’t need to care)

fn unchecked_transaction_imm(&self) -> SqlResult<UncheckedTransaction<'_>>

Begin unchecked_transaction with TransactionBehavior::Immediate. Use when the first operation will be a read operation, that further writes depend on for correctness.

fn get_db_size(&self) -> Result<u32, Error>

Get the DB size in bytes

impl Drop for UncheckedTransaction<'_>

fn drop(&mut self)

Executes the destructor for this type.

impl Deref for UncheckedTransaction<'_>

type Target = Connection

The resulting type after dereferencing.

fn deref(&self) -> &Connection

Dereferences the value.