[davidfowl]

Use ulong.TryFormat for Content-Length values >= 1000 in Kestrel

Faster HTTP/1.1 Content-Length header writing for values >= 1000

Description

BufferExtensions.WriteNumeric (used by HttpResponseHeaders.CopyToFast to write Content-Length) handles values < 1000 inline, but anything larger, which covers most real response bodies, went through WriteNumericMultiWrite. That path reads a [ThreadStatic] scratch array, runs a div/mod-by-10 loop, and then copies the digits into the output.

This change formats directly into the output span with ulong.TryFormat (UTF-8, invariant). Only when the remaining span is too small does it fall back to formatting into the existing [ThreadStatic] scratch array (unchanged from main) and then calling BufferWriter.Write. That fallback lives in its own non-inlined method, and it keeps the existing behavior of filling the tail of the current span. The change adds no unsafe code, stackalloc, Unsafe/MemoryMarshal or custom vectorization; it uses only the BCL TryFormat API. BufferWriter.cs is untouched.

No public API changes (no PublicAPI.*.txt edits) and no behavior change: the output bytes are identical to main for every value.

I also added BufferWriterWriteNumericBenchmark (and a BufferWriterWriteAsciiBenchmark companion) to Kestrel's Microbenchmarks.

Benchmark

BufferWriterWriteNumericBenchmark.WriteNumeric (BDN InProcess, 2 vCPU Linux VM, 3 interleaved base/PR rounds, 1000 ops per invoke; re-measured after removing stackalloc):

Number	main	PR	delta
5	1.10–1.12 ns	1.10–1.12 ns	=
421	2.22–2.51 ns	2.21–2.25 ns	= (noise)
4,096	9.61–9.65 ns	4.95–5.34 ns	-47%
1,048,576	12.44–12.70 ns	6.91–7.16 ns	-44%
123,456,789,012	17.16–17.25 ns	9.78–9.94 ns	-43%

No allocations before or after.

Tests

• Microsoft.AspNetCore.Server.Kestrel.Core.Tests: 9907 passed, 13 failed. All 13 failures are the AddressBinder/KestrelServerTests IPv6 tests, which fail the same way on main in this environment. The existing PipelineExtensionTests (WritesNumericToAscii, WritesNumericToAsciiFastPath, WritesNumericAcrossSpanBoundaries) pass. After the stackalloc removal, I re-ran PipelineExtensionTests|HttpResponseHeadersTests|ChunkWriterTests: 200 passed.

🤖 Posted by ASP.NET Core Perf 2 (AI Agent)

[Copilot]

🔵 Needs a closer look

The required frozen review bundle was not supplied, preventing validation of revision provenance, routed guidance completeness, and existing-feedback deduplication.

No open issues 🎉

What changed in this PR

This PR optimizes Kestrel’s HTTP/1.1 Content-Length formatting for values ≥ 1000 while retaining the existing inline path for smaller values.

Changes:

• Replaces thread-local scratch storage and manual digit conversion with direct UTF-8 ulong.TryFormat, using a stack-allocated fallback across span boundaries.
• Adds scoped span parameters so BufferWriter can accept stack-allocated input.
• Adds numeric and ASCII-writing microbenchmarks.

File	Description
src/​Shared/​ServerInfrastructure/​BufferWriter.cs	Marks span inputs as scoped to support stack-allocated buffers.
src/​Shared/​ServerInfrastructure/​BufferExtensions.cs	Uses direct numeric formatting with a separate cross-span fallback.
src/​Servers/​Kestrel/​perf/​Microbenchmarks/​BufferWriterExtensionsBenchmark.cs	Adds benchmarks for numeric and ASCII writes.

_{Review effort: Balanced}

[davidfowl]

/azp run aspnetcore-ci

The only CI failure is Components.WebView.StaticWebAssets.WebViewBuildBehaviorTests.Publish_ProjectReferenceToWebViewWithJsModuleRcl_SucceedsWithSingleModulesManifest on Windows Server x64. That test isn't related to this change, which only touches Kestrel's shared buffer helpers, so I'm rerunning it as a likely flake.

🤖 Posted by ASP.NET Core Perf 2 (AI Agent)

[azure-pipelines]

No pipelines are associated with this pull request.

[davidfowl]

/azp run

🤖 Posted by ASP.NET Core Perf 2 (AI Agent)

[azure-pipelines]

Command 'run

**🤖' is not supported by Azure Pipelines.



Supported commands

• help:
• Get descriptions, examples and documentation about supported commands
• Example: help "command_name"
• list:
• List all pipelines for this repository using a comment.
• Example: "list"
• run:
• Run all pipelines or specific pipelines for this repository using a comment. Use this command by itself to trigger all related pipelines, or specify specific pipelines to run.
• Example: "run" or "run pipeline_name, pipeline_name, pipeline_name"
• where:
• Report back the Azure DevOps orgs that are related to this repository and org
• Example: "where"

See additional documentation.

[davidfowl]

/azp run aspnetcore-components-e2e

The only failure is VirtualizationTest.VariableHeight_HtmlTable_DoesNotProduceNestedTrElements in the Mono client/WebAssembly leg. The same test passed in the CoreCLR and Mono server legs, and it also failed on unrelated PR #70108, so I'm rerunning it as a flake.

🤖 Posted by ASP.NET Core Perf 2 (AI Agent)

[azure-pipelines]

No pipelines are associated with this pull request.

[davidfowl]

CI status: I pushed an empty commit to retrigger CI. On that run, the Windows Server x64 platform tests that failed last time (WebViewBuildBehaviorTests) now pass.

The only failure on this run is aspnetcore-components-e2e → Blazor E2E tests on Linux (Mono client and WebAssembly), where 4 of 3,803 tests failed. The same leg passed on the first run against identical code, since the retrigger commit is empty. It's unrelated to this change, which only touches Kestrel's shared BufferExtensions/BufferWriter numeric formatting, so it looks flaky. Could a maintainer rerun that leg? I can't trigger reruns.

🤖 Posted by ASP.NET Core Perf 2 (AI Agent)

[davidfowl]

Pushed an empty commit to retrigger CI. /azp run isn't accepted from this account. The previous run's only failure was the Blazor E2E Mono client/WebAssembly leg (VirtualizationTest.VariableHeight_HtmlTable_DoesNotProduceNestedTrElements). It passed on the CoreCLR and Mono server legs, also fails on unrelated PR #70108, and a sibling VirtualizationTest is already quarantined (#69093). This PR only changes Kestrel's numeric header formatting.

🤖 Posted by ASP.NET Core Perf 2 (AI Agent)

[PureWeen]

/review