Skip to content

benchmarks.md

Latest commit

 

History

History
267 lines (183 loc) · 7.34 KB

benchmarks.md

File metadata and controls

267 lines (183 loc) · 7.34 KB

FreakyKit.Forge — Benchmark Results

Full benchmark results comparing Forge against hand-written code and popular mapping libraries. Source code: benchmarks/FreakyKit.Forge.Benchmarks

.NET 8

Environment
Runtime .NET 8.0.11 (Arm64 RyuJIT armv8.0-a)
Machine Apple M4 Pro, 14 cores, macOS Tahoe 26.3
Benchmark tool BenchmarkDotNet v0.15.8
Warmup / Iterations 10 warmup, 50 iterations

Competitors

Library Version
AutoMapper 16.1.1
Mapperly 4.3.1
Mapster 7.4.0
Facet 5.8.2

Simple Mapping (4 properties)

Method Mean Ratio Rank Allocated
Hand-written 6.37 ns 1.00x 1 40 B
Forge 6.46 ns 1.01x 1 40 B
Mapperly 6.47 ns 1.02x 1 40 B
Facet 12.60 ns 1.98x 2 104 B
Mapster 12.68 ns 1.99x 2 40 B
AutoMapper 30.06 ns 4.72x 3 40 B

Medium Mapping (10 properties)

Method Mean Ratio Rank Allocated
Forge 12.43 ns 0.86x 1 96 B
Mapperly 12.71 ns 0.88x 1 96 B
Hand-written 14.49 ns 1.00x 2 96 B
Mapster 18.52 ns 1.28x 3 96 B
Facet 20.57 ns 1.42x 4 160 B
AutoMapper 37.95 ns 2.62x 5 96 B

Nested Object Mapping

Method Mean Ratio Rank Allocated
Forge 21.92 ns 0.93x 1 136 B
Hand-written 23.57 ns 1.00x 2 136 B
Mapperly 24.47 ns 1.04x 2 136 B
Mapster 29.77 ns 1.26x 3 136 B
Facet 38.62 ns 1.64x 4 328 B
AutoMapper 46.37 ns 1.97x 5 136 B

Property Flattening

Method Mean Ratio Rank Allocated
Forge 10.97 ns 0.94x 1 56 B
Hand-written 11.72 ns 1.00x 2 56 B
Mapperly 12.16 ns 1.04x 2 56 B
Mapster 18.70 ns 1.60x 3 56 B
Facet* 38.12 ns 3.25x 4 320 B
AutoMapper 38.35 ns 3.27x 4 56 B

*Facet maps nested objects rather than flattening — source types cannot be annotated with [Flatten].

Deep Object Graph (scalars + 2 nested objects + collections)

Method Mean Ratio Rank Allocated
Hand-written 204.5 ns 1.00x 1 1.86 KB
Forge 208.5 ns 1.02x 1 1.86 KB
Mapster 245.2 ns 1.20x 2 1.79 KB
Mapperly 260.8 ns 1.28x 3 1.83 KB
AutoMapper 326.5 ns 1.60x 4 2.13 KB
Facet 1,641.1 ns 8.03x 5 8.51 KB

Collection Mapping (1,000 items)

Method Mean Ratio Rank Allocated
Hand-written 5,261 ns 1.00x 1 64,232 B
Forge 5,270 ns 1.00x 1 64,232 B
AutoMapper 7,648 ns 1.45x 2 72,704 B
Mapperly 7,696 ns 1.46x 2 64,200 B
Mapster 7,991 ns 1.52x 3 64,160 B
Facet 58,252 ns 11.07x 4 314,216 B

Update Mapping (void, modify existing object)

Mapperly and Facet excluded — neither supports void in-place update. Timings use InvocationCount=1 (high variance expected).

Method Mean Rank
Hand-written ~25 ns 1
Forge ~28 ns 1
Mapster ~175 ns 2
AutoMapper ~534 ns 3

Throughput (10,000 objects)

Method Mean Ratio Rank Allocated
Forge 152.0 μs 0.98x 1 1,016 KB
Hand-written 155.2 μs 1.00x 1 1,016 KB
Mapperly 171.1 μs 1.10x 2 1,016 KB
Mapster 209.0 μs 1.35x 3 1,016 KB
Facet 240.8 μs 1.55x 4 1,641 KB
AutoMapper 414.0 μs 2.67x 5 1,016 KB

Real-World: E-Commerce Order (enums + nested customer + line items + addresses)

Method Mean Ratio Rank Allocated
Forge 161.5 ns 1.00x 1 1.13 KB
Hand-written 161.9 ns 1.00x 1 1.13 KB
Mapperly 165.6 ns 1.02x 1 1.09 KB
Mapster 166.7 ns 1.03x 1 1.05 KB
AutoMapper 208.0 ns 1.28x 2 1.13 KB
Facet 544.7 ns 3.36x 3 2.99 KB

Real-World: Nullable Database Entity (16 nullable columns)

Fully populated (all values present):

Method Mean Ratio Rank Allocated
Forge 11.45 ns 0.98x 1 168 B
Mapperly 11.57 ns 0.99x 1 168 B
Hand-written 11.71 ns 1.00x 1 168 B
Mapster 17.51 ns 1.50x 2 168 B
Facet 18.99 ns 1.62x 3 232 B
AutoMapper 36.17 ns 3.09x 4 168 B

Sparse (many nulls — new/incomplete accounts):

Method Mean Ratio Rank Allocated
Hand-written 11.50 ns 1.00x 1 168 B
Forge 11.66 ns 1.01x 1 168 B
Mapperly 12.11 ns 1.05x 1 168 B
Mapster 17.37 ns 1.51x 2 168 B
Facet 17.98 ns 1.56x 2 232 B
AutoMapper 36.84 ns 3.20x 3 168 B

Key Takeaways (.NET 8)

  • Forge matches hand-written code — consistently within 1–2% across all scenarios, including real-world models with enums, nullables, and nested graphs
  • Zero allocation overhead — identical memory footprint to hand-written mappers
  • 2.5–4.7x faster than AutoMapper — no reflection overhead at runtime
  • Faster than Mapster — especially in medium, nested, flattening, and collection scenarios
  • Competitive with Mapperly — trades leads across scenarios; Forge wins on medium mappings, nested graphs, flattening, and e-commerce real-world
  • Forge leads on throughput — fastest across 10,000-object batches at 152 μs vs hand-written 155 μs
  • Facet — competitive on flat/simple mappings but allocates significantly more and struggles on deep graphs and large collections

.NET 10

Benchmarks for .NET 10 have not been run yet. When available, results will be added here in the same format as the .NET 8 section above.

To run them: update TargetFramework in benchmarks/FreakyKit.Forge.Benchmarks/FreakyKit.Forge.Benchmarks.csproj to net10.0, then run:

dotnet run -c Release -- -f '*'

and populate the tables below using the same structure as the .NET 8 section.

Environment

Simple Mapping (4 properties)

Medium Mapping (10 properties)

Nested Object Mapping

Property Flattening

Deep Object Graph

Collection Mapping (1,000 items)

Update Mapping

Throughput (10,000 objects)

Real-World: E-Commerce Order

Real-World: Nullable Database Entity

Key Takeaways (.NET 10)