Open Manual

Status and execution routes

Nytrix is a working compiler with native, LLVM, and browser routes; it is still under active development. Parsing and semantic checks are shared across those routes, so source generally behaves consistently, while code generation and host APIs keep route-specific limits. For a project, check the features, ABI signatures, and host services it needs on the target it will use.

Prerequisites: none; check the compiler revision and target before using this status information.

Where Nytrix fits

Small tools, native experiments, and compiler research are good fits today: C interop, compiler inspection, and executable checks are available. APIs, package support, and target coverage are still evolving, so pin a commit for reproducible builds and expect to adapt to compiler changes.

The bounded proof, ownership, and effect analyses are useful local checks, not general formal verification. A program that runs through LLVM may need a different source shape or flag set for the owned native backend or a browser.

Execution routes

GoalCommandBoundary
Run a file during developmentny file.nyDefault route; LLVM JIT on supported builds.
Build and run an executableny --run=aot file.nyAOT compilation followed by execution.
Save an executableny --output=app file.nyAOT compilation without execution; needs a compatible target to run.
Require the owned native pathny --native-only file.nyNYIR and native JIT; unsupported shapes fail without LLVM fallback.
Emit WebAssemblyny --native-backend=wasm --emit=wasm=app.wasm file.nyWasm target with explicit imports and its supported runtime surface.
Test browser behavior./make web-testBrowser host contract and browser-specific fixtures.
Inspect interpreted NYRny --nyr-run file.nyDiagnostic VM, not native or LLVM execution.

--output implies compile-only. Combine --native-only with --run=aot or --output when checking an AOT artifact rather than native JIT execution. Command defaults and available backends depend on the build; ny --help reports its options. See Native for target and ABI rules, REPL for interactive execution, and Browser for browser capabilities.

Feature boundaries

AreaCurrent boundaryReference
Memory safety checks--heap=safe traps use-after-free, out-of-bounds, and double free on the program's own malloc/zalloc/raw-heap blocks (guard pages, quarantine) and reports leaks. There is no static use-after-free analysis and no user-program ASan/UBSan/MSan/TSan lane yet; ownership analysis stays opt-in and local.Safety, Ownership
Types, traits, and overloadsShared semantic checks precede backend selection; lowering still has target- and shape-specific limits.Types, Traits
Refinements and proofsA bounded calculus; false and unknown proof obligations are not evidence.Types, Proofs
Dependent types and indexed dataValue-indexed Fin, dependent parameters and returns with call-site substitution, Sigma pairs, GADT evidence, first-class Type values, type-level functions, and runtime-index checks are implemented within the spec's boundaries.Dependent types, Types
Preconditions and terminationrequires, decreases, and @total have distinct checks; a measure alone is not a termination proof.Function contracts
Trait lawsChecked in shared trait validation with the proof solver and compile-time Boolean evaluation.Law blocks
EffectsDeclared masks and bounded callback-row propagation; not arbitrary callback provenance or resource authorization.Effects, Effect rows
OwnershipOpt-in borrow, move, and allowance checks; aggregate and alias tracking is partial. Cleanup policy is separate.Ownership, Heap policy
Compile-time generationBounded evaluation and syntax insertion; generated declarations pass through normal checks.Comptime
C interopDeclaration import and aggregate ABI support depend on frontend, target, and signature.Native
UI, audio, and networkingHost capabilities differ between native and browser routes.UI, Networking, Browser

Evaluate a project

  1. List its required targets, ABI signatures, host services, and language features.
  2. Read their contracts in the specification, including unsupported cases.
  3. Compile and run a small representative program through the route you will ship.
  4. Keep checks for the boundaries the application relies on, including expected failures and cleanup.

Testing explains focused fixtures. A fixture covers its declared inputs and execution routes, not every member of a feature family. Changelog results are historical evidence, not a test of the current checkout.

Previously observed smoke findings

A manual smoke on 2026-10-22 (0.10.0+git.60f30d2f0220, default route, library caches disabled) found four source-shape failures. Later fixes cover the dictionary setter reassignment, top-level is_nil folding, atomic tagged-word updates, and deferred mutation of captured locals. The defer and loop-capture fixtures now check LIFO mutation, integer and floating-point slots, panic unwinding, null callbacks, and registration-time snapshots across native JIT, native-only, LLVM MCJIT/ORC, and the diagnostic VM.

These are historical observations, not current route limitations. See the changelog for the defer fix and its focused test command.

Next: