Radix 0.53.0

Install this version#

Pinned download for Radix 0.53.0. For the current release, use Install instead.

curl -fsSL -o radix.tgz \
  https://github.com/faberlang/releases/releases/download/radix-v0.53.0/radix-v0.53.0-aarch64-apple-darwin.tar.gz
tar -xzf radix.tgz
sudo mv radix-v0.53.0-aarch64-apple-darwin/radix /usr/local/bin/
radix --version

Release notes#

Minor release spanning 26 non-merge commits (v0.52.0..v0.53.0). This release moves the exempla corpus and e2e harness into their own crate; retools the Rust, TypeScript, and Go e2e runners for shared target dirs, batch builds, and parallel subprocess tiers (cutting full-corpus time from ~10 min to ~30 s); and introduces the modulus namespace construct as Stage 2 of the native stdlib campaign.

Scale#

Major tracks#

Exempla crate migration#

The exempla corpus (220+ .fab files, expectations, support files) and the entire exempla_e2e harness (Rust, TS, Go, MIR, LLVM, WASM, script, smoke, roundtrip tiers) move from crates/radix/src/ into a dedicated crates/exempla/ crate (1c1c767e4). This decouples the exempla dependency tree from the compiler crate — corpus additions no longer trigger a radix recompile, and the harness gets its own Cargo.toml for targeted dependency management.

Post-move visibility normalization collapses 28 now-orphaned pub(super) items to pub(crate) (0252cecdb). The migration is documented through goal/plan/ ledger phases (60e4ca0db, ec8f8c492) and closed out in the ledger (3cd5f05ce).

E2E speed work#

A coordinated campaign to pull the full 220-exemplum e2e suite from ~10 min down to a responsive developer cycle. Three structural improvements compound:

Shared `CARGO_TARGET_DIR` (b1984809d): Every Rust exemplum previously wrote its own fresh target/ directory, causing the full faber + norma + tokio dependency tree to recompile ~220 times. A single shared target dir reuses the compiled tree across all exempla. Clippy --fix (a duplicate-dep compile) is dropped from the per-exemplum loop — it belongs to the rust-canonical RC-003 tier. Result: ~10 min → 137 s with streaming per-exemplum progress.

Batch workspace build (ad71894b9): The remaining 137 s was dominated by Cargo's per-invocation spawn + fingerprint overhead (~400 ms × 220 = ~95 s). Restructured into three phases: (1) Faber-compile + write each exemplum as a workspace member, (2) a single cargo build --keep-going at the workspace root, (3) run + verify each binary. Status: 220/220 in 45.8 s (was 137 s).

Parallel binary runner (99708fbde): The run phase still launched 218 binaries serially (~175 ms spawn tax each). A new scripta/run-exempla-parallel script runs the built binaries concurrently via xargs -P, collapsing spawn tax across cores. Status: ~0.7 s run phase on an 18-core host (was 38 s).

Tier-level parallelization for TS and Go:

• TypeScript (e31a3c79b): Restructured into serial frontend + parallel typecheck/runtime tiers via std::thread::scope. 48–63 s (was 163 s) — ~2.6–3.4× speedup. Identical pass/fail: 147/220 runnable, 20/220 behavior-checked.
• Go (c4c786754): Same two-phase model; go vet dropped from the loop (pure cost, never affected pass/fail). Unique .go files per exemplum for parallel safety. 27.97 s (was 64.45 s) — ~2.3× speedup.

Compiler performance docs (0bf23ed5a): A new examples/bench_compile.rs benchmarks the full pipeline (frontend ~16 MB/s linear, Rust codegen, scena scripting path). Documented in README.md with backend/linking scaling analysis (c580cadc8): dep model, per-package target-dir reuse, the what-recompiles matrix, and cold-vs-warm costs.

Modulus namespace (Stage 2)#

The modulus keyword introduces a body-bearing namespace/module construct, separating the overloaded pactum role into two distinct constructs (b77f20186):

• modulus (new): namespace/module façade with body-bearing methods by default. First customer: norma:chorda.
• pactum (unchanged): interface contract, signature-only, implet target.
• genus (unchanged): instantiable object with ego and instance state.

Prerequisite: the untyped frontmatter metadata key modulus is renamed to group to free the keyword for permanent language surface (84cac383d).

The full parse/lower/codegen path lands in 9892f3722: lexer token, parser decl, AST node, HIR lower (LoweredProgram.modules, AnalyzedUnit.modules), Rust codegen dispatch, and five targeted tests. Chorda surface docs (2b0b11111) lock the stdlib API surface. Stage 2 is frozen at 9892f372 (d9d9c58f8).

Other changes#

• Factory docs — tensor ledger closeout: Tensor-types ledger is closed with per-stage SHAs and verification evidence (f522dbe26, a9775fa03, 68d4ee289). Native-stdlib delivery spec, plan, and ledger added (eec5d31bc).
• Factory docs — e2e speed documentation: Goal, phased plan, measured results, and cross-harness comparison recorded in the factory ledger (86e9f05ce, f9e36d8c2, 0688a48ed, 888ddbe74).
• Stage 2 pause: Native-stdlib delivery is paused at Stage 2 closeout (706ec8b6c).

---

All releases · Install the current release