Avoid array-pool traffic for short encoded StringPool inputs

[Garume]

Related to #699.

Maintainer feedback requested: I could not find recorded maintainer approval on the issue. Please confirm whether the short-input gains and the documented empty/long-input tradeoffs are acceptable before this is considered ready to merge.

For #699, use a bounded 64-character stack buffer when Encoding.GetMaxCharCount fits, avoiding the array-pool rent/return pair for short encoded inputs. Larger inputs retain the existing pooled path. The pointer overload of Encoding.GetChars, its capacity argument, and the actual decoded length remain unchanged. A fixed nonempty buffer also preserves a non-null destination for custom encodings reporting zero characters.

This is a performance tradeoff, not a universal throughput improvement or a warmed-allocation reduction. Both versions allocate 0 B for warmed cache hits. On Windows x64/.NET 10.0.1, selected short-input cases improved, while empty input and some pooled cases regressed. The table below includes every measured scenario. ARM64, other runtimes, cache misses, and contention have not been benchmarked.

Validation:

• All 661 HighPerformance tests pass on .NET 10.
• All 39 StringPool tests pass on .NET Framework 4.7.2.
• Library builds for netstandard2.0, netstandard2.1, and net8.0.
• Added coverage for both sides of the threshold, Unicode, UTF-8 fallback/throwing behavior, empty/null behavior, interned identity, and custom encoding capacity/pointer behavior.
• An ArrayPool EventListener audit observes 1,000 rent/return pairs before and zero after for 1,000 warmed eligible calls. Large inputs retain 1,000 pairs.

Benchmarks compare compiled baseline and candidate libraries; the baseline assembly was renamed to allow aliases in the same harness. These are single-machine screening measurements, with one benchmark process per case, three warmups, and eight measured iterations. Maintainer feedback on the short-input/empty-input tradeoff is welcome.

BenchmarkDotNet v0.16.0-nightly.20260801.595, Windows 11 (10.0.26200.9457/25H2/2025Update/HudsonValley2)
AMD Ryzen AI 9 HX 370 w/ Radeon 890M 2.00GHz, 1 CPU, 24 logical and 12 physical cores
Memory: 63.12 GB Total, 36.52 GB Available
.NET SDK 10.0.101
  [Host]     : .NET 10.0.1 (10.0.1, 10.0.125.57005), X64 RyuJIT x86-64-v4
  Job-RTBDTB : .NET 10.0.1 (10.0.1, 10.0.125.57005), X64 RyuJIT x86-64-v4

IterationCount=8  IterationTime=150ms  WarmupCount=3  

Method	Scenario	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Before	Empty	1.474 ns	0.0833 ns	0.0297 ns	1.00	0.00	-	NA
After	Empty	2.424 ns	0.0388 ns	0.0203 ns	1.65	0.03	-	NA
Before	Unicode	36.817 ns	0.2646 ns	0.1384 ns	1.00	0.00	-	NA
After	Unicode	34.164 ns	0.1499 ns	0.0784 ns	0.93	0.00	-	NA
Before	Utf16_16	28.776 ns	0.2278 ns	0.1191 ns	1.00	0.00	-	NA
After	Utf16_16	24.211 ns	0.0988 ns	0.0517 ns	0.84	0.00	-	NA
Before	Utf8_1024	115.664 ns	1.5392 ns	0.6834 ns	1.00	0.00	-	NA
After	Utf8_1024	118.840 ns	0.4153 ns	0.2172 ns	1.03	0.01	-	NA
Before	Utf8_126	63.372 ns	0.4188 ns	0.1860 ns	1.00	0.00	-	NA
After	Utf8_126	66.063 ns	0.5296 ns	0.2770 ns	1.04	0.01	-	NA
Before	Utf8_127	64.699 ns	0.2023 ns	0.0898 ns	1.00	0.00	-	NA
After	Utf8_127	65.135 ns	0.1958 ns	0.0869 ns	1.01	0.00	-	NA
Before	Utf8_128	46.265 ns	0.1916 ns	0.0851 ns	1.00	0.00	-	NA
After	Utf8_128	43.146 ns	0.2842 ns	0.1262 ns	0.93	0.00	-	NA
Before	Utf8_62	44.268 ns	0.3175 ns	0.1661 ns	1.00	0.00	-	NA
After	Utf8_62	40.494 ns	0.2192 ns	0.1146 ns	0.91	0.00	-	NA
Before	Utf8_63	47.320 ns	0.2697 ns	0.1197 ns	1.00	0.00	-	NA
After	Utf8_63	40.136 ns	0.2929 ns	0.1301 ns	0.85	0.00	-	NA
Before	Utf8_64	39.877 ns	0.2050 ns	0.1072 ns	1.00	0.00	-	NA
After	Utf8_64	43.532 ns	0.2453 ns	0.1283 ns	1.09	0.00	-	NA
Before	Utf8_8	27.135 ns	0.2062 ns	0.1079 ns	1.00	0.00	-	NA
After	Utf8_8	26.944 ns	0.1497 ns	0.0783 ns	0.99	0.00	-	NA

Developed with AI assistance; changes and validation were reviewed before submission.

PR Checklist

• Created a feature branch in my fork.
• Based on the latest main (b135626dd54d33b8f05f2ff31591592c004aa848).
• Contains no merge commits.
• Tested with the repository's .NET SDK 10.0.101.
• Added tests for the changed behavior and compatibility cases.
• Contains no public API or intentional breaking behavior changes.
• Existing file/style conventions followed; no new production APIs or source files.

Benchmark harness and reproduction layout

Build the full-history baseline checkout with -p:AssemblyName=CommunityToolkit.HighPerformance.Baseline, then build this PR's checkout normally. The projects below use aliases to compare both compiled libraries. The measurements above used baseline b135626dd54d33b8f05f2ff31591592c004aa848 and candidate 879500f891a2dbd8b526938dc7d434394ec6bd3a.

Directory layout:

communitytoolkit-perf-first/   # library checkout
  src/CommunityToolkit.HighPerformance/bin/Release/net8.0/CommunityToolkit.HighPerformance.dll
toolkit-first-evidence/
  baseline/CommunityToolkit.HighPerformance.Baseline.dll
  benchmark/                 # Program.cs and Benchmark.csproj below

From the common parent directory, first restore and build the baseline checkout:

git -C communitytoolkit-perf-first checkout b135626dd54d33b8f05f2ff31591592c004aa848
dotnet restore communitytoolkit-perf-first/src/CommunityToolkit.HighPerformance/CommunityToolkit.HighPerformance.csproj
dotnet build communitytoolkit-perf-first/src/CommunityToolkit.HighPerformance/CommunityToolkit.HighPerformance.csproj -c Release -f net8.0 --no-restore -p:AssemblyName=CommunityToolkit.HighPerformance.Baseline -o toolkit-first-evidence/baseline
# Fetch this PR's branch first if using an upstream-only clone.
git -C communitytoolkit-perf-first checkout 879500f891a2dbd8b526938dc7d434394ec6bd3a
dotnet build communitytoolkit-perf-first/src/CommunityToolkit.HighPerformance/CommunityToolkit.HighPerformance.csproj -c Release -f net8.0 --no-restore

Place the following files in toolkit-first-evidence/benchmark. The pinned prerelease BenchmarkDotNet package is available from the public https://pkgs.dev.azure.com/dnceng/public/_packaging/benchmark-dotnet-prerelease/nuget/v3/index.json feed; include nuget.org for its dependencies. Run dotnet run -c Release -- --filter '*' --artifacts results from that directory. Use an isolated NuGet.Config if unrelated machine-wide package sources interfere.

Benchmark.csproj:

<Project Sdk="Microsoft.NET.Sdk"><PropertyGroup><OutputType>Exe</OutputType><TargetFramework>net10.0</TargetFramework><ImplicitUsings>enable</ImplicitUsings><Nullable>enable</Nullable></PropertyGroup><ItemGroup><PackageReference Include="BenchmarkDotNet" Version="0.16.0-nightly.20260801.595"/><Reference Include="CommunityToolkit.HighPerformance.Baseline"><HintPath>../baseline/CommunityToolkit.HighPerformance.Baseline.dll</HintPath><Aliases>baseline</Aliases></Reference><Reference Include="CommunityToolkit.HighPerformance"><HintPath>../../communitytoolkit-perf-first/src/CommunityToolkit.HighPerformance/bin/Release/net8.0/CommunityToolkit.HighPerformance.dll</HintPath><Aliases>candidate</Aliases></Reference></ItemGroup></Project>

Program.cs:

extern alias baseline;
extern alias candidate;
using System.Text;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Configs;
using BenchmarkDotNet.Exporters.Json;
using BenchmarkDotNet.Jobs;
using BenchmarkDotNet.Running;
using Perfolizer.Horology;
using OldPool = baseline::CommunityToolkit.HighPerformance.Buffers.StringPool;
using NewPool = candidate::CommunityToolkit.HighPerformance.Buffers.StringPool;

var config = ManualConfig.Create(DefaultConfig.Instance)
    .AddJob(Job.Default.WithWarmupCount(3).WithIterationCount(8).WithIterationTime(TimeInterval.FromMilliseconds(150)))
    .AddExporter(JsonExporter.Full);
var reports = BenchmarkSwitcher.FromAssembly(typeof(Program).Assembly).Run(args, config).ToArray();
return reports.Length == 0 || reports.Any(s => s.Reports.Any(r => !r.Success)) ? 1 : 0;

[MemoryDiagnoser]
public class StringPoolBenchmarks
{
    [Params("Empty", "Utf8_8", "Utf8_62", "Utf8_63", "Utf8_64", "Utf8_126", "Utf8_127", "Utf8_128", "Utf8_1024", "Utf16_16", "Unicode")]
    public string Scenario { get; set; } = "";
    private byte[] input = null!;
    private Encoding encoding = null!;
    private OldPool oldPool = null!;
    private NewPool newPool = null!;

    [GlobalSetup]
    public void Setup()
    {
        encoding = Scenario.StartsWith("Utf16") ? Encoding.Unicode : Encoding.UTF8;
        string text = Scenario == "Empty" ? "" : Scenario == "Unicode" ? "日本語😀の文字列" : new string('x', int.Parse(Scenario.Split('_')[1]));
        input = encoding.GetBytes(text);
        oldPool = new OldPool(); newPool = new NewPool();
        oldPool.Add(text); newPool.Add(text);
        if (oldPool.GetOrAdd(input, encoding) != text || newPool.GetOrAdd(input, encoding) != text) throw new Exception("Decode mismatch");
    }

    [Benchmark(Baseline = true)] public string Before() => oldPool.GetOrAdd(input, encoding);
    [Benchmark] public string After() => newPool.GetOrAdd(input, encoding);
}

[Garume]

@dotnet-policy-service agree