Turbosql
An easy local data persistence layer, backed by SQLite.
- Schema auto-defined by your Rust
struct
s - Automatic schema migrations
- Super-simple basic
INSERT
/SELECT
/UPDATE
/DELETE
operations - Use complex SQL if that's your jam
- Validates all SQL (including user-supplied) at compile time
Usage
use ;
See integration_test.rs
or trevyn/turbo for more usage examples!
Under the Hood
Turbosql generates a SQLite schema and prepared queries for each struct:
use ;
↓ auto-generates and validates the schema
INTEGER PRIMARY KEY,
name TEXT,
age INTEGER,
image_jpg BLOB,
) STRICT
INSERT INTO person (rowid, name, age, image_jpg) VALUES (?, ?, ?, ?)
SELECT rowid, name, age, image_jpg FROM person
(
rowid
Queries with SQL predicates are also assembled and validated at compile time. Note that SQL types vs Rust types for parameter bindings are not currently checked at compile time.
let people = select!;
↓
SELECT rowid, name, age, image_jpg FROM person WHERE age > ?
Automatic Schema Migrations
At compile time, the #[derive(Turbosql)]
macro runs and creates a migrations.toml
file in your project root that describes the database schema.
Each time you change a struct
declaration and the macro is re-run (e.g. by cargo
or rust-analyzer
), migration SQL statements are generated that update the database schema. These new statements are recorded in migrations.toml
, and are automatically embedded in your binary.
↓ auto-generates migrations.toml
= [
'CREATE TABLE person(rowid INTEGER PRIMARY KEY) STRICT',
'ALTER TABLE person ADD COLUMN name TEXT',
]
= '''
CREATE TABLE person (
rowid INTEGER PRIMARY KEY,
name TEXT
) STRICT
'''
When your schema changes, any new version of your binary will automatically migrate any older database file to the current schema by applying the appropriate migrations in sequence.
This migration process is a one-way ratchet: Old versions of the binary run on a database file with a newer schema will detect a schema mismatch and will be blocked from operating on the futuristically-schema'd database file.
Unused or reverted migrations that are created during development can be manually removed from migrations.toml
before being released, but any database files that have already applied these deleted migrations will error and must be rebuilt. Proceed with care. When in doubt, refrain from manually editing migrations.toml
, and everything should work fine.
- Just declare and freely append fields to your
struct
s. - Check out the
migrations.toml
file that is generated in your project root to see what's happening. - If you run into any weird compiler errors, try just re-compiling first; depending on the order the proc macros run, sometimes it just needs a little push to get in sync after a schema change.
- Schema migrations are one-way, append-only. This is similar to the approach taken by leafac/sqlite-migration for the Node.js ecosystem; see that project for a discussion of the advantages!
- On launch, versions of your binary built with a newer schema will automatically apply the appropriate migrations to an older database.
- If you're feeling adventurous, you can add your own schema migration entries to the bottom of the list. (For creating indexes, etc.)
- You can hand-write complex migrations as well, see turbo/migrations.toml for some examples.
- Please open a GitHub issue with any questions or suggestions!
Where's my data?
The SQLite database file is created in the directory returned by directories_next::ProjectDirs::data_dir()
+ your executable's filename stem, which resolves to something like:
$XDG_DATA_HOME
/{exe_name}
or $HOME
/.local/share/{exe_name}
/home/alice/.local/share/fooapp/fooapp.sqlite
$HOME
/Library/Application Support/{exe_name}
/Users/Alice/Library/Application Support/org.fooapp.fooapp/fooapp.sqlite
{FOLDERID_LocalAppData}
\{exe_name}
\data C:\Users\Alice\AppData\Local\fooapp\fooapp\data\fooapp.sqlite
Transactions and async
SQLite, and indeed many filesystems in general, only provide blocking (synchronous) APIs. The correct approach when using blocking APIs in a Rust async
ecosystem is to use your executor's facility for running a closure on a thread pool in which blocking is expected. For example:
async
(Note that spawn_blocking
returns a JoinHandle
that must itself be unwrapped, hence the need for ??
near the end of these examples.)
Under the hood, Turbosql uses persistent thread_local
database connections, so a continuous sequence of database calls from the same thread are guaranteed to use the same exclusive database connection. Thus, async
transactions can be performed as such:
use ;
async
Turbosql sets a SQLite busy_timeout
of 3 seconds, so any table lock contention is automatically re-tried up to that duration, after which the command that was unable to acquire a lock will return with an error.
For further discussion of Turbosql's approach to async
and transactions, see https://github.com/trevyn/turbosql/issues/4. Ideas for improvements to the ergonomics of the solution are very welcome.
-wal
and -shm
files
SQLite is an extremely reliable database engine, but it helps to understand how it interfaces with the filesystem. The main .sqlite
file contains the bulk of the database. During database writes, SQLite also creates .sqlite-wal
and .sqlite-shm
files. If the host process is terminated without flushing writes, you may end up with these three files when you expected to have a single file. This is always fine; on next launch, SQLite knows how to resolve any interrupted writes and make sense of the world. However, if the -wal
and/or -shm
files are present, they must be considered essential to database integrity. Deleting them may result in a corrupted database. See https://sqlite.org/tempfiles.html.
Example Query Forms
Check integration_test.rs
for more examples of what works and is tested in CI.
let result = select!?;
Returns one value cast to specified type, returns Error
if no rows available.
let result = select!?;
SELECT
keyword is always optional when using select!
; it's added automatically as needed.Parameter binding is straightforward.
let result = select!?;
Returns Vec
containing another type. If no rows, returns empty Vec
.
let result = select!?;
Returns Ok(None)
if no rows, Error(_)
on error.
let result = select!?;
Column list and table name are optional if type is a #[derive(Turbosql)]
struct.
let result = select!?;
You can use other struct types as well; column names must match the struct and you must specify the source table in the SQL.Implement Default
to avoid specifying unused column names.(And, of course, you can put it all in a Vec
or Option
as well.)
let result = select!?;
Sometimes everything is optional; this example will retrieve all Person
rows.
"turbosql" or "Turbosql"?
Your choice, but you definitely do not want to capitalize any of the other letters in the name! ;)