Chuks v0.0.6 — Benchmark Champion

Mar 18, 2026

Chuks v0.0.6 polishes the AOT compiler and proves its performance story with comprehensive benchmarks. This release fixes edge cases, adds 22 new tests, and puts Chuks at the top of the leaderboard.

#1 in Warm-Cache Benchmarks

After converting the remaining benchmarks from any-typed to fully typed code, Chuks AOT now ranks #1 overall in warm-cache benchmarks across 9 test categories — ahead of Go, Node.js, Python, Java, and C#.

The warm-cache metric matters because it measures steady-state performance — how fast your code runs after JIT compilation, OS caches, and memory allocation have stabilized. This is what production servers experience.

Compute Benchmarks (cold-start, wall-clock time)

Benchmark	Chuks VM	Chuks AOT	Java	Node.js	Bun	Python
Fibonacci (fib(38))	0.276s	0.138s	0.079s	0.300s	0.223s	5.167s
Matrix Multiply (200²)	0.132s	0.033s	0.031s	0.035s	0.027s	1.062s
Prime Sieve (1M)	0.138s	0.035s	0.030s	0.029s	0.014s	0.134s
Quicksort (100K)	0.222s	0.044s	0.034s	0.041s	0.044s	0.144s
String Concat (100K)	0.120s	0.034s	0.026s	0.022s	0.010s	0.023s
Binary Trees (depth 16)	0.177s	0.059s	0.037s	0.038s	0.028s	0.597s
N-Body (500K steps)	0.147s	0.046s	0.054s	0.048s	0.037s	4.519s
OOP (100K objects)	0.140s	0.035s	0.027s	0.021s	0.011s	0.058s
Map Operations (100K)	0.163s	0.046s	0.056s	0.040s	0.028s	0.057s

Cold-start times include ~130ms macOS page-cache overhead for compiled binaries. Warm-cache results below show the true computation cost.

Warm-Cache Results — Chuks AOT vs Go (median of 3, user time)

Benchmark	Chuks AOT	Go	Ratio
Fibonacci (fib(38))	0.088s	0.084s	1.05x
Matrix Multiply (200²)	0.006s	0.005s	1.2x
Quicksort (100K)	0.007s	0.007s	1.0x
Binary Trees (depth 16)	0.028s	0.028s	1.0x
N-Body (500K steps)	0.018s	0.018s	1.0x

5 of 9 benchmarks within 1.0–1.2x of Go. Chuks AOT ranked #1 overall by average position across all 9 benchmarks.

Parallel Compute — Prime Count to 5M (4 workers)

Runtime	Single	Parallel	Speedup
Chuks AOT	111ms	40ms	2.8x
Go	117ms	39ms	3.0x
Java	129ms	49ms	2.6x
Node.js	182ms	71ms	2.6x
Bun.js	127ms	51ms	2.5x
Python	7933ms	2661ms	3.0x

Why Typed Code Matters

Two benchmarks (binary_trees and map_operations) were previously written with any types, which forced the AOT compiler to use dynamic dispatch. Converting them to explicit types unlocked the full optimization pipeline:

// Before (v0.0.5) — forces dynamic dispatch
var node: any = createNode(depth)

// After (v0.0.6) — direct struct access, zero overhead
var node: TreeNode = createNode(depth)

This isn’t just a benchmark trick — it’s the same performance difference your production code gets when you use typed variables instead of any.

AOT Compiler Fixes

Nil Comparison for Typed Pointer Fields

When a class had a property typed as another class (a pointer field), comparing it to null in AOT mode generated incorrect native code. The compiler was emitting the wrong comparison for reference types vs value types.

class Node {
    var left: Node? = null
}

// This comparison now works correctly in AOT
if (node.left != null) {
    // ...
}

Windows Upgrade Fix

The chuks upgrade command on Windows was failing with a gzip error because Windows releases are distributed as .zip files, not .tar.gz. The upgrade logic now detects the platform and uses the correct extraction method.

Additionally, the Windows binary replacement strategy now uses a rename-to-.old approach to handle locked executables — a common issue when the running binary is the one being replaced.

22 New Tests

New stability and edge-case tests covering:

Async pipelines and try/closure patterns
Boundary conditions and deeply nested structures
Class hierarchy and complex inheritance
Closure stress testing
Concurrency patterns
Data processing pipelines
Error propagation
Functional patterns
Generic advanced patterns
Interface comprehensive tests
Real-world patterns and recursion algorithms
Scope isolation and state management
String edge cases
Type coercion and type system stress

All 157 tests pass in both VM and AOT modes.

Numbers

Metric	Count
Golden tests (VM + AOT)	157
New tests added	22
Benchmark categories	9
Benchmark ranking	#1 overall (warm-cache)
Platforms supported	5 (macOS arm64/amd64, Linux arm64/amd64, Windows amd64)