[davidfowl]

Reduce allocations in HttpMethodMatcherPolicy.GetEdges

Reduce allocations and time in HttpMethodMatcherPolicy.GetEdges during DFA build

Description

• GetEdges runs for many nodes while the DFA is built. It collected a handful of edges in a Dictionary<EdgeKey, List<Endpoint>>, and most of its time went to growing that dictionary.
• It now uses an insertion-ordered list with a linear search and the same key equality, so the edge order is unchanged. EdgeKey.Equals is case-insensitive but GetHashCode is ordinal. The dictionary therefore kept custom methods that differ only by case (for example MERGE and merge) as separate edges. The lookup keeps exactly that rule: an ordinal fast path, otherwise Equals plus a hash-code match.
• It reuses a static array for the any-method (*) sentinel.
• The lookup uses only safe code (a plain List<T> indexer loop), with no CollectionsMarshal, Unsafe or custom vectorization.
• Adds a GetEdges benchmark to HttpMethodMatcherPolicyBenchmark.
• Adds GetEdges_CustomMethodsDifferingInCase_CreateSeparateEdgesInInsertionOrder, which pins the existing edge order and case handling. It passes on both main and this branch.

There are no public API or behavior changes.

Benchmarks

Release builds of Microsoft.AspNetCore.Routing.Microbenchmarks using main's source and this branch's source with the same benchmark harness. Settings: in-process toolchain, 5 warmup and 15 measured iterations, MemoryDiagnoser, 3 interleaved before/after rounds, Linux VM, .NET 12 alpha. Values are the mean of the 3 rounds.

Benchmark	Before	After	Alloc before	Alloc after
HttpMethodMatcherPolicyBenchmark.GetEdges (DestinationCount=3)	628 ns	351 ns	1,288 B	864 B
HttpMethodMatcherPolicyBenchmark.GetEdges (DestinationCount=5)	839 ns	557 ns	1,560 B	1,352 B
HttpMethodMatcherPolicyBenchmark.GetEdges (DestinationCount=11)	1,801 ns	1,456 ns	3,280 B	2,816 B
MatcherBuilderGithubBenchmark.Dfa	601 µs	570 µs	672,945 B	644,182 B

Each round showed the same direction for GetEdges. The full GitHub-routes DFA build allocates 4.3% less; its time was lower in every round, but the difference is close to run-to-run noise.

Tests

• Microsoft.AspNetCore.Routing.Tests: 3506/3506 passed, including the new test.

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

[Copilot]

🟢 Approval recommended

No active review issues are reported for the routing edge-allocation optimization.

No open issues 🎉

What changed in this PR

Reduces allocations in HttpMethodMatcherPolicy.GetEdges during routing DFA construction while preserving existing edge ordering and key-matching behavior.

Changes:

• Replaces dictionary-based edge collection with an insertion-ordered list and linear lookup.
• Reuses a static array for the any-method sentinel.
• Adds a focused benchmark and a regression test for custom HTTP methods differing only by case.

File	Description
src/​Http/​Routing/​test/​UnitTests/​Matching/​HttpMethodMatcherPolicyTest.cs	Tests separate edges and insertion order for case-differing custom methods.
src/​Http/​Routing/​src/​Matching/​HttpMethodMatcherPolicy.cs	Uses list-based edge lookup and reuses the any-method array.
src/​Http/​Routing/​perf/​Microbenchmarks/​Matching/​HttpMethodMatcherPolicyBenchmark.cs	Adds a GetEdges benchmark with method-specific, CORS, and unrestricted endpoints.

_{Review effort: Balanced}

[davidfowl]

/azp run aspnetcore-components-e2e

Rerunning: the only failure is ServerRoutingTest.NavigationLock_CanBlockNavigation_ThenCancel, which Build Analysis shows with a ~1.6% failure rate. It covers Blazor client-side navigation locking, not endpoint HTTP-method matching.

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

[azure-pipelines]

No pipelines are associated with this pull request.

[davidfowl]

CI status: aspnetcore-ci (builds, Helix and platform tests) is green. The only remaining failure is aspnetcore-components-e2e (Mono server) on ServerRoutingTest.NavigationLock_CanBlockExternalNavigation. Build Analysis lists a ~25.6% failure rate for that test, and it doesn't depend on endpoint HTTP-method matching. The earlier runs each failed on a different Blazor UI test, and this pipeline also failed on 4 of the last 5 main commits. Routing tests pass.

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

[PureWeen]

/review