Skip to content

spec-todo.md

Latest commit

 

History

History
102 lines (86 loc) · 6.29 KB

spec-todo.md

File metadata and controls

102 lines (86 loc) · 6.29 KB
Raw

RSScript language-spec TODO

Tracks the spec surface (RSScript_v0.7_Spec.md) beyond the executable MVP.

The executable MVP (§3.1) is implemented and tested at VM↔compiled parity. The open items below are the committed in-scope roadmap (spec §20.2). Everything else is either already shipped or a recorded non-goal.

Open items — in scope, to implement (spec §20.2)

Four committed, design-compatible directions. All are large; each extends the model without reversing a review-first tenet. Ordered by readiness/value.

  • Two-tier execution substrate: interpreter + AOT (§20.1-C, §20.2-4) — done. rss run --vm and the reg-VM test/benchmark harness give a fast no-rustc path, while rss run --release remains the Rust-lowering AOT authority. Both tiers observe identical semantics/diagnostics (the existing VM↔compiled parity invariant), so the fast tier is trustworthy. Single files run through the source VM; package directories through the package VM with native host bindings loaded. Measured: ~54 ms (VM) vs ~14 s (AOT) for the same program.

  • Scoped views / slices (zero-copy borrowed regions) (§3.2, §20.1-I, §20.2-1) — done. The borrowed-view types already existed and ran at parity (String.view/StringView, Bytes.view/BytesView, Buffer.view/BufferView, lowered to Rust slices). Added the spec's dedicated view name = expr form: a parser-level desugar turns view v = e; <rest> into with e as v { <rest> }, so the view's scope ends at the enclosing block and it inherits every with-lease rule — cannot escape (return), be retained, cross await, or enter a managed graph — with no new borrow analysis and zero changes below the parser (so VM↔compiled parity is automatic). Verified: parity_scoped_view_binding (incl. a view-of-a-view) and the scoped-view-escape.rss fail fixture (RS0702). The with Buffer.view(...) as v { … } form also already works.

  • Capability objects / explicit dynamic dispatch (§3.2, §20.1-G, §20.2-2) — done. The compiled backend already lowered Capability<Protocol> to a closed-world enum-of-impls with match dispatch (no dyn); this completes the feature: (1) the spec's capability Protocol keyword form now parses as sugar for Capability<Protocol> (parser desugar), so who: read capability Greeter and local x: capability Greeter work; (2) fixed a real VM parity bug — the reg-VM had no runtime dynamic dispatch, so any Protocol.method(self: x, …) call (capability objects and generic bounds) returned Unit instead of dispatching. Added a CallDynamic instruction + Hir::protocol_method_targets that selects the concrete impl by the receiver's runtime struct type, mirroring the compiled enum dispatch. Verified at parity (parity_capability_dynamic_dispatch, two impls). The review map already flags dynamic dispatch (has_dynamic_protocol_dispatch).

  • [~] Cross-isolate message API (zero-copy transfer) (§20.1-B, §20.2-3) — message contract landed; multi-isolate execution still future. The reviewable core — the message payload contract — is implemented: Channel.message<T> creates a channel whose payload T is statically verified cross-isolate-transferable (is_cross_isolate_transferable: a self-contained value with no managed handle — v1 allows Copy scalars, String, Bytes), rejecting managed/container payloads at check time (RS0036). It reuses the bounded-channel runtime, so it runs today at VM↔compiled parity (parity_message_channel_roundtrip) and is the forward-compatible transport for real isolates: nothing crossing it ever shares mutable state. This realizes the "message is the right choice for a solo isolate" decision — message semantics + take move-in, no reference-capability machinery. Remaining (future): (a) broaden the payload contract to data-only structs/sums and owned containers via deep-snapshot; (b) a structured isolate-spawn so two isolates actually run with disjoint heaps. True multi-thread/multi-heap isolates are a separate multi-quarter refactor. Full plan in cross-isolate-design.md.

Removed — non-goals (not deferred; deleted from the roadmap)

These contradict core RSScript principles and were removed from the spec rather than kept as future work (spec §20.2 "Removed, not deferred"; §21 non-goals).

  • Unstructured spawn / public task handles — breaks the single-isolate, structured-concurrency model; task_group is the sanctioned form. The compiler continues to reject spawn with a stable diagnostic (RS0015/RS0101).
  • Rust-style open enums / open sum extension — breaks the sealed-sum guarantees that make match-exhaustiveness and review-diffs sound. Sealed sum is the model.

Shipped

  • await in expression position (§14.6.2) — await-hoisting pass; VM↔compiled parity. Short-circuit &&/|| RHS stays RS0411 by design.
  • Structured-fix tooling (§20.1-D) — FixEdit payload + rss fix [--write] [--json]; editor integration belongs in rss-lsp.
  • FFI / native-ABI adapter contracts (§20.1-N) — compact [adapter.<Namespace>] whole-boundary binding (pkg §9.5), single expansion point.
  • Stream<T> + await for (§20.1-H) — in the executable MVP; parity-tested.
  • Sum-type hardening (§20.1-E) — v0.7 baseline complete (payload variants, generics, exhaustiveness, review/diff metadata).
  • Registry-level review-risk badges (§20.1-F) — badges on the package review + registry index, derived from risk + capability summary.
  • Module visibility (§20.1-M) — pub enforced at the package boundary; non-pub is package-private by design (§14.8).

Optional post-v0.7 enhancements (non-blocking)

Small/medium add-ons to shipped features; pick up if a real driver demands it.

  • Fix edits for the remaining machine-applicable fixes (extend rss fix).
  • LSP-protocol daemon (stdio) — use rss-lsp rather than expanding rss.
  • pub use re-export — re-export a dependency's item as a package's own.
  • Sum named payload fields — Variant { field: T } alongside positional.
  • Stream combinators / user-defined async generators — .map/.filter, custom stream sources.