perf: Performance Optimization - Hot Path Improvements

[thomhurst]

Overview

This epic tracks performance optimizations targeting TUnit's hot paths in test registration, orchestration, and execution. The goal is holistic performance improvement across test discovery speed, execution throughput, and memory pressure.

Constraints

• No public API changes - All optimizations are internal implementation details
• AOT/Trimming compatible - Changes must work with Native AOT and IL trimming
• Incremental & measurable - Each change should be profiled before/after

Analysis Methodology

Static code analysis identified performance anti-patterns in hot paths:

• LINQ allocations (.Any(), .Where(), .OrderBy(), .Select().ToArray())
• Excessive object allocations per-test
• Lock contention limiting parallel throughput
• Defensive copies and intermediate collections

Sub-Issues

Priority	Issue	Scope	Status
P0	#4160 - LINQ Elimination in Hot Paths	8 files, ~25 locations	Open
P1	#4161 - Object Pooling for Frequent Allocations	5 files, ~10 locations	Open
P1	#4162 - Lock Contention Reduction	2 files, complex refactor	Open
P2	#4163 - Allocation Reduction via Span/ArrayPool	6 files, ~15 locations	Open

Profiling Approach

Using dotnet-trace + SpeedScope for profile-guided optimization:

# Capture profile
dotnet-trace collect --name TUnit.PerformanceBenchmarks -- \
    dotnet run --project TUnit.PerformanceBenchmarks -c Release

# Convert for SpeedScope
dotnet-trace convert trace.nettrace --format speedscope

Benchmark Suite

TUnit.PerformanceBenchmarks project with:

• Mixed realistic test patterns (60% simple, 30% data-driven, 10% lifecycle)
• Scale tiers: 100, 500, 1,000, 5,000, 10,000 tests
• Reproducible baselines for regression testing

Success Criteria

• Measurable improvement in test discovery time at 10k scale
• Reduced GC pressure during test execution
• No performance regression in any tier
• All changes pass existing test suite

[TimothyMakkison]

Hey, had a couple ideas of my own the other day. Hard to tell if any would actually be worth doing.

NVM I just saw the branch
Is it possible to run benchmarkdotnet or Rider/VSStudios profilers on TUnit? I'm not sure it would work perfectly as TUnit does a lot of one time initialisation. I tried running the dotnet-trace command but I don't have a TUnit.PerformanceBenchmarks project.