dactyl-db is a single SQL-vendor-agnostic Rust persistence framework. One crate, one query(sql, params) API, one ambient env-var selector — talk to SQLite, Neon/Postgres, and (planned) Redis, MySQL, Cassandra from one import, instead of pulling a per-database library for each backend.
The same SQL string produces the same logical rows regardless of the active backend. Parameters are always bound, never interpolated, so dynamic values cannot become SQL.
- One import, many backends — SQLite and Neon ship today; Redis, MySQL, and Cassandra are planned behind the same
querysurface. - Ambient selection —
DATASTOREenv var picks the active backend at runtime. Noinit(), no per-call datastore argument, no global connection cache. - Safe parameter binding —
query(sql, &[params])binds typed values; SQL injection via parameter values is structurally impossible. - Atomic batches —
transaction(&[Statement])commits all-or-nothing on every backend (no nesting; caller-owned retry/idempotency; see contract below). - Caller-owned schema — dactyl never silently creates tables.
execute("create table ...")is the only way dactyl touches schema.
Add dactyl-db to your Cargo.toml:
[dependencies]
dactyl-db = { version = "0.1.7", features = ["sqlite", "neon"] }Set the active backend via env vars at runtime:
DATASTORE=sqlite DATASTORE_ROUTE=/path/to/store.db
# or
DATASTORE=neon DATASTORE_ROUTE=https://propodus.example DATASTORE_TOKEN=...
use dactyl_db::{self, query, execute, Parameter, Rows};
fn main() -> Result<(), dactyl_db::DactylError> {
// The caller owns the schema. dactyl never bootstraps tables on its own.
execute(
"create table if not exists todos (id integer primary key, title text not null, status text not null)",
&[],
)?;
execute(
"insert into todos (id, title, status) values ($1, $2, $3)",
&[
Parameter::Integer(1),
Parameter::Text("ship dactyl 0.1.7".into()),
Parameter::Text("open".into()),
],
)?;
let sql = query!("select id, title, status from todos");
for row in dactyl_db::query(&sql, &[])?.iter() {
// Strict typed projection (owned). Prefer try_get if you like Result style.
let id: i64 = row.try_get("id")?;
let title: String = row.get("title")?;
// Nullable columns use Option<T>; missing columns are ColumnNotFound.
let status: Option<String> = row.get("status")?;
println!("todo {id}: {title} [{status:?}]");
}
Ok(())
}Run it:
DATASTORE=sqlite DATASTORE_ROUTE=/tmp/dactyl-example.db \
cargo run --features sqlite --example readme_example
This is the stable contract for typed and NULL-safe extraction (dactyl #25, conformance #2; also DecapodLabs/decapod#1111):
| Concern | Semantics |
|---|---|
| Integer / real / bool / text | get_int, get_real, get_bool, get_str (owned) or strict get::<T> / try_get::<T> via serde |
| Portable bool | get_bool accepts JSON true/false or integer 0/1 (SQLite stores bools as integers) |
| JSON / text | get_json / get_json_ref return the raw cell. Text payloads stay strings until the caller parses them; Neon may surface structured JSON objects. |
| SQL NULL | get::<Option<T>> → None; non-Option getters → Conversion mentioning NULL; is_null / get_json surface null without converting |
| Missing column | DactylError::ColumnNotFound |
| Duplicate aliases | First match left-to-right. select a as x, b as x → get("x") is a. Use a positional index for later duplicates. |
| Owned vs borrowed | get / get_* return owned values that outlive the row. get_str_ref / get_json_ref borrow from &Row for the row lifetime. A Row outlives the adapter connection. |
| Conversion failure | DactylError::Conversion with the column key and a reason |
The active backend is chosen by ambient environment variables — no init() call, no per-call datastore argument, no process-wide connection cache. Each query / execute / transaction call constructs a fresh short-lived adapter and drops it on return, so workspace and session isolation is automatic and the public surface is Send + Sync without any lock.
| Environment Variable | Allowed Values / Format | Required | Description |
|---|---|---|---|
DATASTORE |
"sqlite" or "neon" |
yes | Selects the active backend. Any other value is a typed error. |
DATASTORE_ROUTE |
SQLite filepath OR Neon endpoint URL | yes | Connection route for the selected backend. |
DATASTORE_TOKEN |
Opaque string | no | Bearer token forwarded to Neon. Ignored by SQLite. |
| Function | Purpose |
|---|---|
query(sql, params) |
One entry point for any SQL (read or write). Returns Rows. |
execute(sql, params) |
DDL / migration / affected-row operations. Returns affected count. |
transaction(&[Statement]) |
Atomic batch; full rollback on any per-statement failure. |
query!("sql") macro |
Compile-time SQL-literal analysis. |
Parameter enumerates the typed binding set: Null, Bool, Integer, Real, Text. The adapter forwards the values verbatim — never as interpolated SQL.
Row provides get / try_get, lenient scalar getters, is_null, and borrowed get_str_ref / get_json_ref under the projection contract above.
Stable contract for multi-statement units of work (#24; prerequisite for DecapodLabs/decapod#1111 / #1120):
| Concern | Semantics |
|---|---|
| Atomicity | Any per-statement failure aborts the whole unit. SQLite uses a real transaction; Neon uses one POST /batch that the server accepts or rejects as a unit. Empty slice → Ok([]). |
| Nesting | Not supported. No SAVEPOINTs. Each call uses a fresh adapter; put every statement in one slice. |
| Retry | dactyl does not retry. Callers own retry policy. |
| Timeout | No public deadline. Neon uses reqwest defaults; SQLite is local. |
| Idempotency | Not idempotent. Replays may conflict or double-write. Design deterministic keys / upserts if retrying after ambiguous transport failures. |
| Proof | Conformance covers SQLite + Neon-mock failure injection and an event-plus-state fixture (state row + event row in one batch; mid-batch failure leaves neither). |
This project is licensed under the MIT License - see the LICENSE file for details.