Very very WIP in-process toy SQL database written in Rust in a similar vein to SQLite.
The next goal is to pass sqlites sqllogictest test suite. The next feature to be implemented is correlated subqueries.
A nsql-shell can be acquired with cargo r -r -- /path/to/db. It will be created if it doesn't exist.
nsql is the library crate which drives the helper crates.
nsql-parse is a tiny wrapper around the excellent sqlparser-rs library.
nsql-ir contains the intermediate representation definitions (think relational algebra).
nsql-catalog contains catalog definition and is used by the binder to resolve table references etc.
This is bootstrapped in a manner similar to postgres' system catalogs. E.g. nsql catalog metadata is stored in nsql tables.
Try SELECT * FROM nsql_catalog.nsql_table.
nsql-bind contains the binder which lowers the ast from parsing into a LogicalPlan defined by nsql-ir.
This includes typechecking and name resolution.
nsql-opt is effectively a noop layer at this point in time. It does some subquery flattening but not much else yet.
This takes as input a non-optimized logical plan and outputs an optimized logical plan.
nsql-execution takes a logical plan and (naively) transforms it into a physical plan.
This physical plan is converted into pipelines and then executed.
The executor is currently a (naive) push-based executor inspired by DuckDB's.
The storage layer is backed by a key-value store, LMDB and redb are supported.
Other key value can be implemented via the StorageEngine trait provided
that the storage engine's WriteTransaction doesn't require a mutable reference to use.