Files
Aura/docs/plans/0001-core-streaming-substrate.md
T
Brummel 65967fb181 plan: cycle 0001 core streaming substrate
Five-task plan executing spec 0001: scalar set, Column/Window, KindMismatch,
AnyColumn, then a workspace build/test/clippy + surface-purity gate. Each task
adds its own module + re-export so the crate compiles green after every task;
14 unit tests pin the substrate contract incl. the wrong-kind-push guard.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-03 11:37:46 +02:00

680 lines
21 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Core Streaming Substrate — Implementation Plan
> **Parent spec:** `docs/specs/0001-core-streaming-substrate.md`
>
> **For agentic workers:** REQUIRED SUB-SKILL: use the `implement` skill to run
> this plan. Steps use `- [ ]` checkboxes for tracking.
**Goal:** Deliver the `aura-core` streaming substrate — the four scalar base
types, the bounded-lookback financial-indexed `Column`/`Window`, the type-erased
`AnyColumn` edge, and the `KindMismatch` guard — fully built, tested, and lint-clean.
**Architecture:** Four new private modules in `aura-core` (`scalar`, `column`,
`any`, `error`), each with its own inline `#[cfg(test)] mod tests`, re-exported
from `lib.rs`. No dependencies added. Interior-mutability-free: a sim is
single-threaded (C1), so columns are owned `&mut` on push, borrowed `&` on read.
**Tech Stack:** Rust (edition 2024, `aura-core` crate), `cargo build/test/clippy
--workspace`.
**Module ordering (by dependency):** `scalar` (standalone) → `column`
(generic, standalone) → `error` (uses `scalar::ScalarKind`) → `any` (uses all
three) → final workspace gate. Each task adds its file, its `mod`+`pub use` line
in `lib.rs`, and its tests, so the crate compiles green after every task.
**Greenfield note:** every type here is new; the tests pin each module's contract
and run green on first build. There is no pre-existing behaviour to RED against —
the must-fail guard (`wrong_kind_push_is_rejected`, Task 4) is the one test whose
*correct behaviour is rejection*.
**Files this plan creates or modifies:**
- Create: `crates/aura-core/src/scalar.rs``Timestamp`, `Scalar`, `ScalarKind`, `kind()`, four `From` impls.
- Create: `crates/aura-core/src/column.rs``Column<T>` ring + zero-copy `Window<'_, T>` with `Index`.
- Create: `crates/aura-core/src/error.rs``KindMismatch { expected, got }`.
- Create: `crates/aura-core/src/any.rs``AnyColumn` type-erased edge.
- Modify: `crates/aura-core/src/lib.rs:1-19` — module declarations + re-exports (doc prose updated).
- Test: inline `#[cfg(test)] mod tests` in each of the four module files.
---
## Task 1: The scalar set (`scalar.rs`)
**Files:**
- Create: `crates/aura-core/src/scalar.rs`
- Modify: `crates/aura-core/src/lib.rs`
- Test: inline in `crates/aura-core/src/scalar.rs`
- [ ] **Step 1: Create `crates/aura-core/src/scalar.rs` with the scalar set**
```rust
//! The closed four-type scalar set streamed on aura's hot path (C7):
//! `i64`, `f64`, `bool`, and `Timestamp` (epoch-ns UTC). The carrier `Scalar`
//! is a `Copy` POD enum — no type-erased payloads, no heap, no reference counting.
/// Canonical engine time: epoch-nanoseconds UTC. Newtype over `i64` (C7).
/// `Default` (epoch 0) lets `Column<Timestamp>` pre-size its ring.
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct Timestamp(pub i64);
/// The kind tag of a scalar / column, used for the edge-time type check (C7).
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
pub enum ScalarKind {
I64,
F64,
Bool,
Timestamp,
}
/// The four scalar base types, type-erased into one `Copy` carrier (C7).
#[derive(Clone, Copy, Debug, PartialEq)]
pub enum Scalar {
I64(i64),
F64(f64),
Bool(bool),
Ts(Timestamp),
}
impl Scalar {
/// The kind tag of this scalar.
pub fn kind(self) -> ScalarKind {
match self {
Scalar::I64(_) => ScalarKind::I64,
Scalar::F64(_) => ScalarKind::F64,
Scalar::Bool(_) => ScalarKind::Bool,
Scalar::Ts(_) => ScalarKind::Timestamp,
}
}
}
impl From<i64> for Scalar {
fn from(v: i64) -> Self {
Scalar::I64(v)
}
}
impl From<f64> for Scalar {
fn from(v: f64) -> Self {
Scalar::F64(v)
}
}
impl From<bool> for Scalar {
fn from(v: bool) -> Self {
Scalar::Bool(v)
}
}
impl From<Timestamp> for Scalar {
fn from(v: Timestamp) -> Self {
Scalar::Ts(v)
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn kind_matches_variant() {
assert_eq!(Scalar::I64(1).kind(), ScalarKind::I64);
assert_eq!(Scalar::F64(1.0).kind(), ScalarKind::F64);
assert_eq!(Scalar::Bool(true).kind(), ScalarKind::Bool);
assert_eq!(Scalar::Ts(Timestamp(1)).kind(), ScalarKind::Timestamp);
}
#[test]
fn from_widenings() {
assert_eq!(Scalar::from(7i64), Scalar::I64(7));
assert_eq!(Scalar::from(2.5f64), Scalar::F64(2.5));
assert_eq!(Scalar::from(true), Scalar::Bool(true));
assert_eq!(Scalar::from(Timestamp(9)), Scalar::Ts(Timestamp(9)));
}
#[test]
fn timestamp_orders_and_defaults() {
assert!(Timestamp(1) < Timestamp(2));
assert_eq!(Timestamp::default(), Timestamp(0));
}
#[test]
fn scalar_is_copy() {
let s = Scalar::F64(1.0);
let a = s;
let b = s; // Copy: `s` still usable after move-by-value
assert_eq!(a, b);
assert_eq!(s, a);
}
}
```
- [ ] **Step 2: Rewrite `lib.rs` doc and declare the `scalar` module only**
Each task adds only its own `mod` + `pub use` lines so the crate compiles green
after every task (the `mod` lines accumulate; by Task 4 all four are present).
Replace the entire current body of `crates/aura-core/src/lib.rs` (19 lines,
doc-only) with:
```rust
//! `aura-core` — the shared contract every aura component links against.
//!
//! This crate is the boundary forced by the cdylib hot-reload model: both the
//! engine and every hot-reloadable node cdylib depend on it, so it stays small
//! and stable.
//!
//! Delivered so far (cycle 0001 — the streaming substrate):
//!
//! - the four scalar base types streamed as SoA: [`Scalar`] / [`ScalarKind`] /
//! [`Timestamp`] (C7);
//! - [`Column`] — a fixed-capacity, bounded-lookback ring with financial
//! indexing (index 0 = newest) and a `run_count` freshness primitive (C5/C8),
//! read zero-copy through a [`Window`];
//! - [`AnyColumn`] — the type-erased edge over the four column kinds, with an
//! edge-time kind check ([`KindMismatch`]) (C7).
//!
//! Still to come (subsequent cycles): the `Node` trait and its `schema`/`eval`,
//! the evaluation context `Ctx`, the firing policies (A: fire-on-any-fresh +
//! hold; B: all-fresh barrier), and the deterministic sim loop.
mod scalar;
pub use scalar::{Scalar, ScalarKind, Timestamp};
```
The intra-doc links to `Column`/`Window`/`AnyColumn`/`KindMismatch` resolve once
Tasks 24 re-export them; they are not checked by `cargo build`/`test`/`clippy`
(only `cargo doc`), so they break no gate in the interim.
- [ ] **Step 3: Build and test the scalar module**
Run: `cargo test -p aura-core scalar::`
Expected: the build succeeds and the four `scalar::tests::*` tests report `ok`;
0 failed.
---
## Task 2: The column and its window (`column.rs`)
**Files:**
- Create: `crates/aura-core/src/column.rs`
- Test: inline in `crates/aura-core/src/column.rs`
- [ ] **Step 1: Create `crates/aura-core/src/column.rs`**
```rust
//! `Column<T>` — a fixed-capacity, bounded-lookback SoA ring with financial
//! indexing (index 0 = newest), and its zero-copy read view `Window` (C5/C8).
/// A fixed-capacity bounded-lookback column. Pre-sized at construction
/// (C8: no realloc in the hot loop); index 0 = newest; carries a monotonic
/// `run_count` of total pushes — the freshness primitive (C5).
pub struct Column<T> {
buf: Box<[T]>,
head: usize, // next write slot
len: usize, // valid elements, saturates at capacity
run_count: u64, // monotonic total pushes (C5)
}
impl<T: Copy + Default> Column<T> {
/// Create a column sized to `lookback` (must be >= 1). The engine sizes this
/// once at wiring; it is never reallocated (C8).
pub fn with_capacity(lookback: usize) -> Self {
assert!(lookback >= 1, "Column lookback must be >= 1");
Column {
buf: vec![T::default(); lookback].into_boxed_slice(),
head: 0,
len: 0,
run_count: 0,
}
}
/// Append the newest value. O(1), allocation-free. On overflow the oldest
/// value is overwritten (true ring). Bumps `run_count`.
pub fn push(&mut self, v: T) {
let cap = self.buf.len();
self.buf[self.head] = v;
self.head = (self.head + 1) % cap;
if self.len < cap {
self.len += 1;
}
self.run_count += 1;
}
/// Financial index: 0 = newest, `len()-1` = oldest. `None` if `k >= len`
/// (C8: not-yet-warmed-up / out of range). Zero-copy.
pub fn get(&self, k: usize) -> Option<T> {
if k >= self.len {
return None;
}
let cap = self.buf.len();
Some(self.buf[(self.head + cap - 1 - k) % cap])
}
pub fn len(&self) -> usize {
self.len
}
pub fn is_empty(&self) -> bool {
self.len == 0
}
pub fn capacity(&self) -> usize {
self.buf.len()
}
/// The monotonic push counter — the freshness primitive (C5). Never moved
/// by a read.
pub fn run_count(&self) -> u64 {
self.run_count
}
/// A zero-copy read view handed to a reading node (C8).
pub fn window(&self) -> Window<'_, T> {
Window { col: self }
}
}
/// Read-only, zero-copy view into a [`Column`] — what a reading node sees (C8).
/// Carries no storage, only a borrow.
pub struct Window<'a, T> {
col: &'a Column<T>,
}
impl<'a, T: Copy + Default> Window<'a, T> {
/// Financial index, 0 = newest; `None` if cold / out of range.
pub fn get(&self, k: usize) -> Option<T> {
self.col.get(k)
}
pub fn len(&self) -> usize {
self.col.len()
}
pub fn is_empty(&self) -> bool {
self.col.is_empty()
}
pub fn run_count(&self) -> u64 {
self.col.run_count()
}
}
/// `window[k]` — newest-at-0 indexed access; panics on out-of-range like a
/// slice. Nodes that may be cold use [`Window::get`] instead.
impl<'a, T: Copy + Default> core::ops::Index<usize> for Window<'a, T> {
type Output = T;
fn index(&self, k: usize) -> &T {
let col = self.col;
assert!(k < col.len, "Window index {k} out of range (len {})", col.len);
let cap = col.buf.len();
&col.buf[(col.head + cap - 1 - k) % cap]
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn newest_at_zero_oldest_at_len_minus_one() {
let mut c = Column::<i64>::with_capacity(4);
c.push(10);
c.push(20);
c.push(30);
assert_eq!(c.len(), 3);
assert_eq!(c.get(0), Some(30)); // newest
assert_eq!(c.get(1), Some(20));
assert_eq!(c.get(2), Some(10)); // oldest
assert_eq!(c.get(3), None); // beyond len
}
#[test]
fn wraparound_keeps_newest_capacity_values() {
let mut c = Column::<i64>::with_capacity(3);
for v in [1, 2, 3, 4, 5] {
c.push(v);
}
assert_eq!(c.len(), 3); // saturated
assert_eq!(c.capacity(), 3);
assert_eq!(c.get(0), Some(5)); // newest
assert_eq!(c.get(1), Some(4));
assert_eq!(c.get(2), Some(3)); // oldest retained
assert_eq!(c.get(3), None); // 1 and 2 were dropped
}
#[test]
fn run_count_counts_pushes_not_len() {
let mut c = Column::<i64>::with_capacity(2);
assert_eq!(c.run_count(), 0);
for v in [1, 2, 3, 4, 5] {
c.push(v);
}
assert_eq!(c.run_count(), 5); // counts pushes, past the ring wrap
assert_eq!(c.len(), 2); // while len stays at capacity
let _ = c.get(0);
assert_eq!(c.run_count(), 5); // a read never moves it
}
#[test]
fn window_is_a_view_matching_the_column() {
let mut c = Column::<f64>::with_capacity(4);
c.push(1.0);
c.push(2.0);
let cap_before = c.capacity();
let w = c.window();
assert_eq!(w.len(), 2);
assert_eq!(w.get(0), Some(2.0));
assert_eq!(w[1], 1.0); // Index, newest-at-0
assert_eq!(w.run_count(), 2);
assert_eq!(c.capacity(), cap_before); // taking a window did not realloc
}
#[test]
fn empty_column_reads_none() {
let c = Column::<bool>::with_capacity(2);
assert!(c.is_empty());
assert_eq!(c.get(0), None);
assert_eq!(c.run_count(), 0);
}
}
```
- [ ] **Step 2: Declare and re-export the `column` module in `lib.rs`**
Add to `crates/aura-core/src/lib.rs`, after the `mod scalar;` line and after the
`pub use scalar::...;` line respectively:
```rust
mod column;
```
```rust
pub use column::{Column, Window};
```
- [ ] **Step 3: Test the column module**
Run: `cargo test -p aura-core column::`
Expected: the five `column::tests::*` tests report `ok`; 0 failed.
---
## Task 3: The guard error (`error.rs`)
**Files:**
- Create: `crates/aura-core/src/error.rs`
- Test: inline in `crates/aura-core/src/error.rs`
- [ ] **Step 1: Create `crates/aura-core/src/error.rs`**
```rust
//! `KindMismatch` — the wiring-time guard error for a wrong-kind edge push (C7).
use crate::scalar::ScalarKind;
/// Returned by [`crate::AnyColumn::push`] when a `Scalar`'s kind does not match
/// the column's kind; the column is left untouched.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub struct KindMismatch {
/// The kind the column holds.
pub expected: ScalarKind,
/// The kind of the value that was offered.
pub got: ScalarKind,
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn kind_mismatch_is_a_plain_value() {
let e = KindMismatch { expected: ScalarKind::I64, got: ScalarKind::F64 };
assert_eq!(e, KindMismatch { expected: ScalarKind::I64, got: ScalarKind::F64 });
assert_ne!(e, KindMismatch { expected: ScalarKind::F64, got: ScalarKind::I64 });
}
}
```
- [ ] **Step 2: Declare and re-export the `error` module in `lib.rs`**
Add to `crates/aura-core/src/lib.rs`:
```rust
mod error;
```
```rust
pub use error::KindMismatch;
```
- [ ] **Step 3: Test the error module**
Run: `cargo test -p aura-core error::`
Expected: `error::tests::kind_mismatch_is_a_plain_value` reports `ok`.
---
## Task 4: The type-erased edge (`any.rs`)
**Files:**
- Create: `crates/aura-core/src/any.rs`
- Test: inline in `crates/aura-core/src/any.rs` (incl. the must-fail guard)
- [ ] **Step 1: Create `crates/aura-core/src/any.rs`**
```rust
//! `AnyColumn` — the type-erased edge: the four concrete `Column<T>` behind one
//! enum. The kind check is paid at the edge (wiring time); the hot loop grabs
//! the concrete column once (`as_*_mut`) and pushes monomorphically (C7).
use crate::column::Column;
use crate::error::KindMismatch;
use crate::scalar::{Scalar, ScalarKind, Timestamp};
/// A type-erased edge over the four scalar columns (C7).
pub enum AnyColumn {
I64(Column<i64>),
F64(Column<f64>),
Bool(Column<bool>),
Ts(Column<Timestamp>),
}
impl AnyColumn {
/// Create a type-erased column of `kind`, sized to `lookback` (C8).
pub fn with_capacity(kind: ScalarKind, lookback: usize) -> Self {
match kind {
ScalarKind::I64 => AnyColumn::I64(Column::with_capacity(lookback)),
ScalarKind::F64 => AnyColumn::F64(Column::with_capacity(lookback)),
ScalarKind::Bool => AnyColumn::Bool(Column::with_capacity(lookback)),
ScalarKind::Timestamp => AnyColumn::Ts(Column::with_capacity(lookback)),
}
}
pub fn kind(&self) -> ScalarKind {
match self {
AnyColumn::I64(_) => ScalarKind::I64,
AnyColumn::F64(_) => ScalarKind::F64,
AnyColumn::Bool(_) => ScalarKind::Bool,
AnyColumn::Ts(_) => ScalarKind::Timestamp,
}
}
pub fn len(&self) -> usize {
match self {
AnyColumn::I64(c) => c.len(),
AnyColumn::F64(c) => c.len(),
AnyColumn::Bool(c) => c.len(),
AnyColumn::Ts(c) => c.len(),
}
}
pub fn is_empty(&self) -> bool {
self.len() == 0
}
pub fn run_count(&self) -> u64 {
match self {
AnyColumn::I64(c) => c.run_count(),
AnyColumn::F64(c) => c.run_count(),
AnyColumn::Bool(c) => c.run_count(),
AnyColumn::Ts(c) => c.run_count(),
}
}
/// Edge-time typed push. `Err(KindMismatch)` if `v`'s kind != this column's
/// kind; the column is left untouched (C7 guard). The hot path bypasses this
/// by taking the concrete `Column<T>` once via `as_*_mut`.
pub fn push(&mut self, v: Scalar) -> Result<(), KindMismatch> {
match (self, v) {
(AnyColumn::I64(c), Scalar::I64(x)) => {
c.push(x);
Ok(())
}
(AnyColumn::F64(c), Scalar::F64(x)) => {
c.push(x);
Ok(())
}
(AnyColumn::Bool(c), Scalar::Bool(x)) => {
c.push(x);
Ok(())
}
(AnyColumn::Ts(c), Scalar::Ts(x)) => {
c.push(x);
Ok(())
}
(this, v) => Err(KindMismatch { expected: this.kind(), got: v.kind() }),
}
}
/// Read one type-erased value (0 = newest). `None` if cold / out of range.
pub fn get(&self, k: usize) -> Option<Scalar> {
match self {
AnyColumn::I64(c) => c.get(k).map(Scalar::I64),
AnyColumn::F64(c) => c.get(k).map(Scalar::F64),
AnyColumn::Bool(c) => c.get(k).map(Scalar::Bool),
AnyColumn::Ts(c) => c.get(k).map(Scalar::Ts),
}
}
/// Concrete-column accessor for the monomorphic hot path; `Some` only for
/// the matching kind.
pub fn as_i64_mut(&mut self) -> Option<&mut Column<i64>> {
match self {
AnyColumn::I64(c) => Some(c),
_ => None,
}
}
pub fn as_f64_mut(&mut self) -> Option<&mut Column<f64>> {
match self {
AnyColumn::F64(c) => Some(c),
_ => None,
}
}
pub fn as_bool_mut(&mut self) -> Option<&mut Column<bool>> {
match self {
AnyColumn::Bool(c) => Some(c),
_ => None,
}
}
pub fn as_ts_mut(&mut self) -> Option<&mut Column<Timestamp>> {
match self {
AnyColumn::Ts(c) => Some(c),
_ => None,
}
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn same_kind_push_round_trips() {
let mut e = AnyColumn::with_capacity(ScalarKind::F64, 4);
e.push(Scalar::F64(1.5)).unwrap();
e.push(Scalar::F64(2.5)).unwrap();
assert_eq!(e.kind(), ScalarKind::F64);
assert_eq!(e.len(), 2);
assert_eq!(e.run_count(), 2);
assert_eq!(e.get(0), Some(Scalar::F64(2.5))); // newest
assert_eq!(e.get(1), Some(Scalar::F64(1.5)));
assert_eq!(e.get(2), None);
}
#[test]
fn wrong_kind_push_is_rejected() {
let mut edge = AnyColumn::with_capacity(ScalarKind::I64, 8);
let err = edge.push(Scalar::F64(1.5)).unwrap_err(); // f64 into an i64 edge
assert_eq!(err, KindMismatch { expected: ScalarKind::I64, got: ScalarKind::F64 });
assert_eq!(edge.len(), 0); // nothing was stored
assert_eq!(edge.run_count(), 0); // and the counter did not move
}
#[test]
fn concrete_accessor_matches_only_its_kind() {
let mut e = AnyColumn::with_capacity(ScalarKind::F64, 2);
assert!(e.as_f64_mut().is_some());
assert!(e.as_i64_mut().is_none());
assert!(e.as_bool_mut().is_none());
assert!(e.as_ts_mut().is_none());
// hot-path push through the concrete column, monomorphic:
e.as_f64_mut().unwrap().push(3.0);
assert_eq!(e.get(0), Some(Scalar::F64(3.0)));
assert_eq!(e.run_count(), 1);
}
#[test]
fn timestamp_edge_round_trips() {
let mut e = AnyColumn::with_capacity(ScalarKind::Timestamp, 2);
e.push(Scalar::Ts(Timestamp(1_700_000_000_000_000_000))).unwrap();
assert_eq!(e.get(0), Some(Scalar::Ts(Timestamp(1_700_000_000_000_000_000))));
}
}
```
- [ ] **Step 2: Declare and re-export the `any` module in `lib.rs`**
Add to `crates/aura-core/src/lib.rs` (this completes the four modules):
```rust
mod any;
```
```rust
pub use any::AnyColumn;
```
- [ ] **Step 3: Test the any module**
Run: `cargo test -p aura-core any::`
Expected: the four `any::tests::*` tests report `ok`, including
`wrong_kind_push_is_rejected`.
---
## Task 5: Workspace gate (build + full test + lint)
**Files:** none (verification only).
- [ ] **Step 1: Full workspace build**
Run: `cargo build --workspace`
Expected: `Finished` with 0 errors, 0 warnings.
- [ ] **Step 2: Full workspace test with a count assertion**
Run: `cargo test --workspace`
Expected: all tests pass; the `aura-core` unit-test binary reports
`test result: ok. 14 passed; 0 failed` (4 scalar + 5 column + 1 error + 4 any).
If the count is not 14, a test was lost or misnamed — investigate before
proceeding.
- [ ] **Step 3: Lint clean under `-D warnings`**
Run: `cargo clippy --workspace --all-targets -- -D warnings`
Expected: `Finished` with no clippy diagnostics (exit 0).
- [ ] **Step 4: Surface-purity check — no `RefCell`/`Rc`/`dyn Any` leaked**
Run: `! grep -rnE 'RefCell|Rc<|dyn Any' crates/aura-core/src`
Expected: exit 0 (no matches). The substrate must carry none of these on its
surface (C1/C7). A match is a regression to fix before the cycle closes.