Skip to content

COMPARE.md

Latest commit

 

History

History
55 lines (44 loc) · 3.38 KB

COMPARE.md

File metadata and controls

55 lines (44 loc) · 3.38 KB
Raw

Rynix vs peers (honest)

This document is product positioning, not a scoreboard. For the definitive “who is ahead?” audit, see VERDICT.md. Methodology and Suite5 tables: END_PEER_GAP.md.

vs End (IrMaho/End)

Both target AI-first / Zero-GC / systems backends. Both compile with a Rust driver (endc / rynixc) and lean on C for native runtime.

Area End (typical positioning) Rynix (shipping today)
README / packaging Strong (logo, badges, 12-bench matrix) Suite5 12 × 5–6 langs + bakeoff (checksum CI)
Benchmark matrix suite12 heavy sims (SDF, HFT, SHA, …) Suite5 integer microkernels (different algorithms)
End in same harness End is a first-class suite12 lang end slot in Suite5 when endc + .end ports exist
AI CLI graph / impact / slice / eval / patch / arch Same + MCP (19 tools) + verify/precheck/context/security/scope/deps/dna
Agent contracts 50-feature Intent→Verify pipeline Wave 1 toolchain evidence (ADR-0009); not End syntax clone
Editor VS Code chrome; no vscode-languageclient VS Code + LanguageClient + LSP + CodeLens
Frameworks EndHyper, EndForge, UI canvas Deferred (ADR-0007)
Concurrency Threads / channels / OpenMP (per End docs) Colorless fibers + PARKED + io_uring (Linux) / IOCP (Windows)
Packages Registry staging (per End docs); PubGrub theater Path deps + local filesystem index (dir-scan or sparse; ADR-0010; no CDN)
Memory story 4-tier regions + leases + borrow Escape → stack/region/heap + injected free
Backend C11 shipping + LLVM alpha Textual LLVM (ADR-0005); C11 deferred (ADR-0008)
Correctness gates checksum per bench (suite12 caveats) CI: C ↔ Rynix all 12 + LLVM↔interp diff
TLS / crypto TLS = plaintext + label; mixed crypto claims Real TLS + NIST KATs
Release Signed binaries (claimed) GitHub Release + SHA256SUMS + optional GPG (release.yml)

Verdict (2026-08-25, peer @cf5bef3): Rynix leads shipping systems + agent toolchain. End leads spectacle (domain wallpaper, C11/UI narrative, suite12 marketing) that we refuse to fake. Phase 12 closed product UX (manifest, HTTP loop, structs, std, Suite12 .ryx, LSP). See VERDICT.md and LEAD_AHEAD.md §8.

Why the compiler is Rust (not Zig/Go)

  • End’s compiler is also Rust.
  • Zig/Go/C/Rust/End in benchmark folders are peer implementations of workloads, not the implementation language of the compiler.
  • Rynix runtime is C (rt/).

Run all peers locally:

python benchmarks/suite5/run_suite5.py --langs c,rust,go,zig,rynix,end --summary
python benchmarks/suite5/analyze_results.py

What we refuse to copy blindly

  • Marketing tables without reproducible harnesses.
  • Marking features complete without in-tree tests.
  • Comparing Suite5 ms to End suite12 ms as if they were the same programs.
  • Inflating RPS or binary-size claims beyond gates in CI/tests.
  • Claiming Phase 12 product waves (structs, std wrappers, LSP) are microbench wins.