Roadmap¶

djox grows in small steps, each useful on its own. Steps 0 to 8 are in 0.1.0.

Step	What
0	Benchmarks against Django's ORM and plain `sqlite3`
1	The Rust SQLite engine and `djox.Database`
2	djox as Django's SQLite `ENGINE`: one copy of SQLite in the process, Django's SQL functions registered
3	Django's QuerySet `a`-methods on djox, falling back to Django's own for the rest; queries inside a sync `atomic()` (like `TestCase`)
4	`djox.atomic()` and `djox.on_commit()`; methods that need a transaction (`aget_or_create()`, `abulk_create()`...) and `asave()` on djox's threads, with writers taking turns
5	Hooks: `before_query` and `after_query` around every statement
6	The fast lane: simple queries, sync and async, reads and writes, skip Django's compiler; Rust builds the results; streamed `aiterator()`
7	Postgres: `djox.Database.postgres()` and `djox.backends.postgresql`
8	Related objects: `select_related()` on the fast lane, `prefetch_related()` off the event loop
9	Speed (unreleased): `djox.asgi`, quick SQLite reads, `Q()`, `exclude()` and joins through foreign keys on the fast lane, `create()` and `asave()` too
10	The general fast lane (unreleased): any chain of calls is a recipe, so `annotate()`, `F()`, aggregates and joins through any relation take the fast lane

Next¶

Postgres: reading arrays, ranges and hstore from async code.
More of the fast lane: subqueries (Exists(), Subquery()), bulk_create(), the prefetching queries themselves.

Need something else? Open an issue.