Type Model

.github/workflows/check.yml

@@ -75,23 +75,6 @@ jobs:
   #         components: rustfmt
   #     - name: cargo-semver-checks
   #       uses: obi1kenobi/cargo-semver-checks-action@v2
-  doc:
-    # run docs generation on nightly rather than stable. This enables features like
-    # https://doc.rust-lang.org/beta/unstable-book/language-features/doc-cfg.html which allows an
-    # API be documented as only available in some specific platforms.
-    runs-on: ubuntu-latest
-    name: nightly / doc
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          submodules: true
-      - name: Install nightly
-        uses: dtolnay/rust-toolchain@nightly
-      - name: Install cargo-docs-rs
-        uses: dtolnay/install@cargo-docs-rs
-      - name: cargo docs-rs
-        # TODO: Once we figure out the crates, rename this.
-        run: cargo docs-rs -p optd-tmp
   hack:
     # cargo-hack checks combinations of feature flags to ensure that features are all additive
     # which is required for feature unification

Cargo.toml

@@ -1,3 +1,3 @@
 [workspace]
-members = ["optd-tmp"]
+members = ["optd-core"]
 resolver = "2"

optd-core/Cargo.toml

@@ -0,0 +1,10 @@
+[package]
+name = "optd-core"
+version = "0.1.0"
+edition = "2021"
+
+[dependencies]
+trait-variant = "0.1.2"
+
+# Pin more recent versions for `-Zminimal-versions`.
+proc-macro2 = "1.0.60" # For a missing feature (https://github.com/rust-lang/rust/issues/113152).

optd-core/src/expression.rs

@@ -0,0 +1,17 @@
+//! Types for logical and physical expressions in the optimizer.
+
+use crate::memo::GroupId;
+use crate::operator::relational::logical::LogicalOperator;
+use crate::operator::relational::physical::PhysicalOperator;
+
+/// A logical expression in the memo table.
+///
+/// References children using [`GroupId`]s for expression sharing
+/// and memoization.
+pub type LogicalExpression = LogicalOperator<GroupId, GroupId>;
+
+/// A physical expression in the memo table.
+///
+/// Like [`LogicalExpression`] but with specific implementation
+/// strategies.
+pub type PhysicalExpression = PhysicalOperator<GroupId, GroupId>;

optd-core/src/lib.rs

@@ -0,0 +1,5 @@
+pub mod expression;
+pub mod memo;
+pub mod operator;
+pub mod plan;
+pub mod rules;

optd-core/src/memo.rs

@@ -0,0 +1,65 @@
+//! Memo table implementation for query optimization.
+//!
+//! The memo table is a core data structure that stores expressions and their logical equivalences
+//! during query optimization. It serves two main purposes:
+//!
+//! - Avoiding redundant optimization by memoizing already explored expressions
+//! - Grouping logically equivalent expressions together to enable rule-based optimization
+//!
+//! # Structure
+//!
+//! - Each unique expression is assigned an expression ID (either [`LogicalExpressionId`],
+//!   [`PhysicalExpressionId`], or [`ScalarExpressionId`])
+//! - Logically equivalent expressions are grouped together under a [`GroupId`]
+//! - Logically equivalent scalar expressions are grouped toegether under a [`ScalarGroupId`]
+//!
+//! # Usage
+//!
+//! The memo table provides methods to:
+//! - Add new expressions and get their IDs
+//! - Add expressions to existing groups
+//! - Retrieve expressions in a group
+//! - Look up group membership of expressions
+//! - Create new groups for expressions
+
+use crate::expression::LogicalExpression;
+
+/// A unique identifier for a logical expression in the memo table.
+#[repr(transparent)]
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct LogicalExpressionId(u64);
+
+/// A unique identifier for a physical expression in the memo table.
+#[repr(transparent)]
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct PhysicalExpressionId(u64);
+
+/// A unique identifier for a scalar expression in the memo table.
+#[repr(transparent)]
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct ScalarExpressionId(u64);
+
+/// A unique identifier for a group of relational expressions in the memo table.
+#[repr(transparent)]
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct GroupId(u64);
+
+/// A unique identifier for a group of scalar expressions in the memo table.
+#[repr(transparent)]
+#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
+pub struct ScalarGroupId(u64);
+
+/// TODO(alexis) Add fields & link to storage layer.
+pub struct Memo;
+
+/// TODO(alexis) Stabilize API by first expanding the Python code.
+impl Memo {
+    /// TODO(alexis) Add docs.
+    pub async fn add_logical_expr_to_group(
+        &mut self,
+        _group_id: GroupId,
+        _logical_expr: LogicalExpression,
+    ) -> LogicalExpressionId {
+        todo!()
+    }
+}

optd-core/src/operator/mod.rs

@@ -0,0 +1,5 @@
+//! This module contains type definitions related to query plan operators, both relational (logical
+//! / physical) and scalar.
+
+pub mod relational;
+pub mod scalar;

optd-core/src/operator/relational/logical/filter.rs

@@ -0,0 +1,8 @@
+/// Logical filter operator that selects rows matching a condition.
+///
+/// Takes input relation (`Relation`) and filters rows using a boolean predicate (`Scalar`).
+#[derive(Clone)]
+pub struct Filter<Relation, Scalar> {
+    pub child: Relation,
+    pub predicate: Scalar,
+}

optd-core/src/operator/relational/logical/join.rs

@@ -0,0 +1,11 @@
+/// Logical join operator that combines rows from two relations.
+///
+/// Takes left and right relations (`Relation`) and joins their rows using a join condition
+/// (`Scalar`).
+#[derive(Clone)]
+pub struct Join<Relation, Scalar> {
+    pub join_type: String,
+    pub left: Relation,
+    pub right: Relation,
+    pub condition: Scalar,
+}

optd-core/src/operator/relational/logical/mod.rs

@@ -0,0 +1,31 @@
+//! Type definitions of logical operators in `optd`.
+
+pub mod filter;
+pub mod join;
+pub mod project;
+pub mod scan;
+
+use filter::Filter;
+use join::Join;
+use project::Project;
+use scan::Scan;
+
+/// Each variant of `LogicalOperator` represents a specific kind of logical operator.
+///
+/// This type is generic over two types:
+/// - `Relation`: Specifies whether the children relations are other logical operators or a group id.
+/// - `Scalar`: Specifies whether the children scalars are other scalar operators or a group id.
+///
+/// This makes it possible to reuse the `LogicalOperator` type in [`LogicalPlan`],
+/// [`PartialLogicalPlan`], and [`LogicalExpression`].
+///
+/// [`LogicalPlan`]: crate::plan::logical_plan::LogicalPlan
+/// [`PartialLogicalPlan`]: crate::plan::partial_logical_plan::PartialLogicalPlan
+/// [`LogicalExpression`]: crate::expression::LogicalExpression
+#[derive(Clone)]
+pub enum LogicalOperator<Relation, Scalar> {
+    Scan(Scan<Scalar>),
+    Filter(Filter<Relation, Scalar>),
+    Project(Project<Relation, Scalar>),
+    Join(Join<Relation, Scalar>),
+}

optd-core/src/operator/relational/logical/project.rs

@@ -0,0 +1,9 @@
+/// Logical project operator that specifies output columns.
+///
+/// Takes input relation (`Relation`) and defines output columns/expressions
+/// (`Scalar`).
+#[derive(Clone)]
+pub struct Project<Relation, Scalar> {
+    pub child: Relation,
+    pub fields: Vec<Scalar>,
+}

optd-core/src/operator/relational/logical/scan.rs

@@ -0,0 +1,9 @@
+/// Logical scan operator that reads from a base table.
+///
+/// Reads from table (`String`) and optionally filters rows using a pushdown predicate
+/// (`Scalar`).
+#[derive(Clone)]
+pub struct Scan<Scalar> {
+    pub table_name: String, // TODO(alexis): Mocked for now.
+    pub predicate: Option<Scalar>,
+}

optd-core/src/operator/relational/mod.rs

@@ -0,0 +1,4 @@
+//! Type definitions for relational (logical and physical) operators.
+
+pub mod logical;
+pub mod physical;

optd-core/src/operator/relational/physical/filter/filter.rs

@@ -0,0 +1,10 @@
+/// Physical filter operator that applies a boolean predicate to filter input rows.
+///
+/// Takes a child operator (`Relation`) providing input rows and a predicate expression
+/// (`Scalar`) that evaluates to true/false. Only rows where predicate is true
+/// are emitted.
+#[derive(Clone)]
+pub struct Filter<Relation, Scalar> {
+    pub child: Relation,
+    pub predicate: Scalar,
+}

optd-core/src/operator/relational/physical/filter/mod.rs

@@ -0,0 +1,2 @@
+#[allow(clippy::module_inception)]
+pub mod filter;

optd-core/src/operator/relational/physical/join/hash_join.rs

@@ -0,0 +1,14 @@
+/// Hash-based join operator that matches rows based on equality conditions.
+///
+/// Takes left and right input relations (`Relation`) and joins their rows using
+/// a join condition (`Scalar`). Builds hash table from build side (right)
+/// and probes with rows from probe side (left).
+#[derive(Clone)]
+pub struct HashJoin<Relation, Scalar> {
+    pub join_type: String,
+    /// Left relation that probes hash table.
+    pub probe_side: Relation,
+    /// Right relation used to build hash table.
+    pub build_side: Relation,
+    pub condition: Scalar,
+}

optd-core/src/operator/relational/physical/join/merge_join.rs

@@ -0,0 +1,13 @@
+/// Merge join operator that matches rows based on equality conditions.
+///
+/// Takes sorted left and right relations (`Relation`) and joins their rows using
+/// a join condition (`Scalar`). Both inputs must be sorted on join keys.
+#[derive(Clone)]
+pub struct MergeJoin<Relation, Scalar> {
+    pub join_type: String,
+    /// Left sorted relation.
+    pub left_sorted: Relation,
+    /// Right sorted relation.
+    pub right_sorted: Relation,
+    pub condition: Scalar,
+}

optd-core/src/operator/relational/physical/join/mod.rs

@@ -0,0 +1,3 @@
+pub mod hash_join;
+pub mod merge_join;
+pub mod nested_loop_join;

optd-core/src/operator/relational/physical/join/nested_loop_join.rs

@@ -0,0 +1,13 @@
+/// Nested-loop join operator that matches rows based on a predicate.
+///
+/// Takes outer and inner relations (`Relation`) and joins their rows using
+/// a join condition (`Scalar`). Scans inner relation for each outer row.
+#[derive(Clone)]
+pub struct NestedLoopJoin<Relation, Scalar> {
+    pub join_type: String,
+    /// Outer relation.
+    pub outer: Relation,
+    /// Inner relation scanned for each outer row.
+    pub inner: Relation,
+    pub condition: Scalar,
+}

optd-core/src/operator/relational/physical/mod.rs

@@ -0,0 +1,33 @@
+//! Type definitions of physical operators in optd.
+
+pub mod filter;
+pub mod join;
+pub mod project;
+pub mod scan;
+
+use filter::filter::Filter;
+use join::{hash_join::HashJoin, merge_join::MergeJoin, nested_loop_join::NestedLoopJoin};
+use project::project::Project;
+use scan::table_scan::TableScan;
+
+/// Each variant of `PhysicalOperator` represents a specific kind of physical operator.
+///
+/// This type is generic over two types:
+/// - `Relation`: Specifies whether the children relations are other physical operators or a group
+///   id.
+/// - `Scalar`: Specifies whether the children scalars are other scalar operators or a group id.
+///
+/// This makes it possible to reuse the `PhysicalOperator` type in [`PhysicalPlan`]
+/// and [`PhysicalExpression`].
+///
+/// [`PhysicalPlan`]: crate::plan::physical_plan::PhysicalPlan
+/// [`PhysicalExpression`]: crate::expression::PhysicalExpression
+#[derive(Clone)]
+pub enum PhysicalOperator<Relation, Scalar> {
+    TableScan(TableScan<Scalar>),
+    Filter(Filter<Relation, Scalar>),
+    Project(Project<Relation, Scalar>),
+    HashJoin(HashJoin<Relation, Scalar>),
+    NestedLoopJoin(NestedLoopJoin<Relation, Scalar>),
+    SortMergeJoin(MergeJoin<Relation, Scalar>),
+}

optd-core/src/operator/relational/physical/project/mod.rs

@@ -0,0 +1,2 @@
+#[allow(clippy::module_inception)]
+pub mod project;

optd-core/src/operator/relational/physical/project/project.rs

@@ -0,0 +1,9 @@
+/// Column projection operator that transforms input rows.
+///
+/// Takes input relation (`Relation`) and projects columns/expressions (`Scalar`)
+/// to produce output rows with selected/computed fields.
+#[derive(Clone)]
+pub struct Project<Relation, Scalar> {
+    pub child: Relation,
+    pub fields: Vec<Scalar>,
+}

optd-core/src/operator/relational/physical/scan/mod.rs

@@ -0,0 +1 @@
+pub mod table_scan;

optd-core/src/operator/relational/physical/scan/table_scan.rs

@@ -0,0 +1,9 @@
+/// Table scan operator that reads rows from a base table
+///
+/// Reads from table (`String`) and optionally filters rows using
+/// a pushdown predicate (`Scalar`).
+#[derive(Clone)]
+pub struct TableScan<Scalar> {
+    pub table_name: String, // TODO(alexis): Mocked for now.
+    pub predicate: Option<Scalar>,
+}

optd-core/src/operator/scalar/add.rs

@@ -0,0 +1,6 @@
+/// Addition expression for scalar values.
+#[derive(Clone)]
+pub struct Add<Scalar> {
+    pub left: Scalar,
+    pub right: Scalar,
+}