cd3d1ca9edfe9dad9025604cb1d61b76a096cdce
5 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Brummel
|
cd3d1ca9ed |
refactor(aura-core): split Scalar into a tag-free Cell + ScalarKind
Motivation
----------
`Scalar` was a tagged enum (I64/F64/Bool/Ts), so every scalar value
physically carried its own kind tag. But the kind is already known from
the schema/port/column the value flows through (C7: the type is a
property of the column, not of the value — the hot path is already
columnar `Column<T>`, and `AnyColumn::get` *reconstructs* the tag from
the column on the way out). The per-value tag was therefore redundant
with the kind the surrounding context already holds.
That redundancy had three costs:
* It baked an implicit `match` (a branch) into every function that read
a Scalar payload — even where the caller statically knew the type.
The tag could never be exploited away.
* Size: a tagged enum is tag + payload = 16 bytes (f64/i64 alignment),
twice the 8 bytes the value needs. A `Column<Scalar>` would be double
the memory and half the cache utilisation.
* It is the shared root of several downstream papercuts we keep hitting
— the lossy f64 manifest field, the `unreachable!` panic on a
non-numeric param, the serde-tag question — all symptoms of "the type
is baked into the value".
Change
------
Introduce `Cell`: a type-erased 64-bit word (`struct Cell(u64)`) that is
not readable without external type context. It is constructed per base
type (`from_i64/from_f64/from_bool/from_ts`) and read only by naming the
type at the call site (`i64()/f64()/bool()/ts()`) — each a branch-free
bit-cast. The hot path resolves the kind once at the boundary (from the
schema) and then reads natively, with no per-value branch. `Cell` knows
nothing of `Scalar` or `ScalarKind`; the dependency is strictly one-way,
and it lives in its own `cell.rs` (more is planned on top of it).
`Scalar` becomes `struct { kind: ScalarKind, cell: Cell }` — the
self-describing form for the dynamic boundaries (builder binding,
serialization, rendering), built on top of `Cell`. Its `as_*` accessors
now `debug_assert` the kind and return the native value (free in
release); calling the wrong accessor is a caller bug, not a checked
`Option`. The variant constructors `Scalar::I64(..)` become associated
fns `Scalar::i64(..)`.
`PartialEq` is hand-written (not derived) to preserve the former enum's
value semantics: kinds must match, then native payloads compare, so f64
keeps IEEE-754 behaviour (`NaN != NaN`, `+0.0 == -0.0`) and a kind
mismatch is never equal even when the raw words coincide. A fixture
(`scalar_eq_is_value_not_bitwise`) pins exactly the cases where bit- and
value-equality diverge, so it can't silently regress. `Cell`'s own
`Eq`/`Hash` stay bitwise — correct for a raw word.
The change is behaviour-preserving: Scalar's observable behaviour is
identical to the pre-Cell enum (the value-equality fixture proves it);
only the internal representation changed. The ~440 call sites across the
workspace are a mechanical constructor rename plus ~12 destructuring
sites (match-arms / `let`-patterns) rewritten to `kind()` + `as_*`.
Verified: cargo build --workspace --all-targets, cargo clippy --workspace
--all-targets -- -D warnings, cargo test --workspace — all green.
|
||
|
Brummel
|
21c1621bd0 |
docs,engine: drop coined "compilat", use FlatGraph / "flat graph"
"Compilat" (German "Kompilat") was a coined noun for the product of the
bootstrap compilation — neither English nor a natural fit. The runtime
artifact already has a code identifier for exactly this thing: the
`FlatGraph` struct (harness.rs). Replace the coinage with that identifier:
- prose mentions -> "flat graph" (mirrors the type, reads plainly)
- definitional anchors -> `FlatGraph` (C11, C23, the running-graph line)
- `render_compilat` -> `render_flat_graph` (historical render symbol;
keeps the `render_blueprint` / `render_flat_graph`
source-vs-product pairing)
- `intra-compilat` -> `intra-graph`
- `from_compilat` (test) -> `from_flat`
"compilation" / "re-compilation" / "bootstrap-as-compilation" (the process,
ordinary English) are deliberately left untouched. Behaviour-preserving:
only comments, design ledger, specs/plans, one test-local variable and its
assert messages change. Full workspace test suite green; clippy clean.
|
||
|
Brummel
|
227d004c9d |
feat(aura-engine): reject unwired or double-wired input ports
A graph that leaves an interior input slot unconnected, or wires one slot from more than one producer, is now a compile-time error instead of a silently-wrong run. Until now an unwired interior slot was accepted and bootstrapped to an empty column (harness.rs:137-148), so a forgotten connection ran a deterministic but wrong backtest; two producers into one slot — ill-formed, since a slot holds one column — was likewise uncompiled-against. A single name-free check, `check_ports_connected`, is added to `validate_wiring` after the existing index/kind checks and before the nested-composite recursion. It counts coverage of each (interior-node, slot) uniformly across interior edges and role targets, and rejects zero coverage (new `CompileError::UnconnectedPort`) or >1 (new `CompileError::DoubleWiredPort`). Because `validate_wiring` is called once from `compile_with_params` and recurses, both the raw `Composite::new` path and the `GraphBuilder::build()` path inherit it at every nesting level with no builder-side code (mirrors `check_param_namespace_injective`, the param-side sibling). A composite's own open input roles (source: None) are coverage providers — the wired-by-enclosing boundary, the root already guarded by UnboundRootRole — not consumers. Index-based and name-free: nothing reaches the compilat, so C23 holds. Supporting fix: the check computes `signature()` on every interior node before the recursion validates that node's interior, so `derive_signature` and `interior_slot_kind` are made bounds-total (a placeholder kind for an out-of-range OutField/target, never a panic; the real fault is still reported by validate_wiring's guarded OutputPortOutOfRange/BadInteriorIndex). This latent panic was exposed by the new check, not introduced by it. Test maintenance (blast radius enumerated empirically, not hand-traced): four existing tests relied incidentally on under-wiring being accepted — three negative tests get a covering root role so their intended deep fault still surfaces via the recursion, and one param-order nesting test is fully wired (param order unchanged). Six new tests: five raw-surface (unwired, double-wire via edge+role and edge+edge, nested, open-role boundary) and one builder-surface forgotten-leg (proving the GraphBuilder inherits the check). Narrowed scope of #65: the original "promote names to load-bearing wiring keys / close the exposure-price swap structurally" goal was dropped — #64 already moved authoring to handles + visible port names, so the residual same-kind swap is a valid-but-wrong name choice no structural check catches, and a structural fix would push domain semantics into the domain-free engine (C10/C7). This ships the name-free half that stands on its own: wiring completeness, which catches forgotten and doubled connections. Verified: cargo test --workspace 231 passed / 0 failed; cargo clippy --workspace --all-targets -- -D warnings clean. closes #65 |
||
|
Brummel
|
b6174cff80 |
refactor(aura-engine): rename NodeHandle::in_/out to input/output
The trailing underscore on in_() was a keyword-escape scar (`in` is reserved).
Rename the NodeHandle port/field accessors to input()/output(), which also
aligns the authoring surface with the schema's own nomenclature
(NodeSchema.inputs/output, "input port"/"output field"): node.input("lhs")
now reads in the same vocabulary the schema uses, instead of a second one.
Mechanical, behaviour-preserving: the two method definitions plus every call
site in builder.rs and the aura-cli sample blueprints. Full workspace green;
clippy --all-targets -D warnings clean.
|
||
|
Brummel
|
a6a314bb98 |
feat(aura-engine): name-based blueprint wiring via GraphBuilder
Add an additive, fluent GraphBuilder authoring surface that wires a blueprint's topology by typed node handles and port/field names instead of raw positional indices. Node references are Copy NodeHandle values (a node reference cannot be mistyped); only port/field names are strings, resolved against each node's cached NodeSchema by exactly-one-match at a single fallible build() -> Result<Composite, BuildError> — the Binder posture, one level up. build() lowers to the unchanged Composite::new, so the compilat stays index-wired (C23): names resolve at the authoring boundary and never reach FlatGraph. A new From<Composite> for BlueprintNode lets add() accept nested composites. Verified: builder-authored sma_cross is structurally equal to the hand-wired fixture, and the full builder-authored harness lowers to a byte-identical FlatGraph (equal edges + sources). The five BuildError variants (Unknown/Ambiguous In/Out, BadHandle) are covered, and the SimBroker exposure/price legs are now addressable by name (the #21 legibility win). The structural close of the exposure/price swap (#21) is deliberately out of scope — this builder makes the swap legible, not impossible — tracked as #65. Two plan-test corrections applied during implementation (both in-scope, no behaviour change): the harness parity test uses compile_with_params with a topology-invariant param vector on both sides (the harness declares three params, so a no-param compile() would trip ParamArity), and the error asserts use .err()/Some(...) rather than unwrap_err() (Composite is not Debug). aura-engine 123 tests green (+9); full workspace green; clippy --all-targets -D warnings clean. Existing index-form Composite::new sites and tests untouched (coexistence). closes #64 |