[Draft] Use hybrid C# reference map for post-processing

[live1206]

Summary

Adds a hybrid reference-map replacement for C# generated-code post-processing.

The hybrid path replaces broad Roslyn reference-map construction with:

• provider metadata for generated-code references
• explicit provider dependencies for known generated body-only references
• a small Roslyn scan only for custom/shared code roots

No benchmark measurement/profiling code is included in the production generator path in this PR.

Why

The earlier experimental PR measured full generation and identified Roslyn reference-map construction inside GeneratedCodeWorkspace.PostProcessAsync() as the largest hotspot.

This PR keeps generated output parity with the Roslyn cleanup path while moving generated-code reachability to provider metadata.

Latest Benchmark Data

Latest data from benchmark PR #10885 after porting the current #10976 hybrid analyzer changes (including generated body invocation edges and base-preserved reachability).

Benchmark output root: /tmp/typespec-final-hybrid-bench-20260624-0943.

Full-generation BenchmarkDotNet, averaged across 3 local runs:

Mode	Avg Mean	Avg Allocated
Roslyn reference maps	895.8 ms	67.78 MB
Provider reference map	859.9 ms	57.75 MB

Approximate full-generation improvement:

Time:       ~4.0% faster
Allocation: ~14.8% less

Focused profile data, using 72 Roslyn-mode and 72 provider-mode invocations from the same runs (median):

Path	Median Time	Median Allocated
Roslyn reference-map construction	357.2 ms	23.70 MB
Provider map analysis + candidate consumption	263.2 ms	17.83 MB
Provider candidate consumption only	0.79 ms	0.06 MB

Approximate focused reference-map improvement:

Time:       ~26.3% faster
Allocation: ~24.8% less

Profile notes:

• Data comes from POSTPROCESSING_BENCHMARK_PROFILE_STEPS=true on the full-generation benchmark so both paths run against real TypeProvider output.
• Roslyn rows are PostProcessor.Internalize.BuildPublicReferenceMapAsync and PostProcessor.Remove.BuildAllReferenceMapAsync.
• Provider rows are Generation.ProviderReferenceMapShadowAnalysis, PostProcessor.Internalize.UseShadowCandidates, PostProcessor.Internalize.UseShadowPublicizeCandidates, PostProcessor.Remove.UseShadowCandidates, and PostProcessor.Remove.BuildShadowReferencedSet.
• Runtime: .NET 10.0.9, Ubuntu 26.04, AMD EPYC 7763.

Correctness Notes

The hybrid implementation preserves Roslyn cleanup behavior for generated output parity:

• model factory signatures and bodies do not keep otherwise-unused models alive
• MRW context/buildable attributes do not keep buildable-only models alive
• serialization providers are removable together with their owning model
• retained serialization providers report explicit helper dependencies such as ChangeTrackingDictionary and Optional
• collection-result providers report explicit body dependencies instead of relying on Roslyn body scanning
• client providers report explicit body dependencies for collection results, service method types, operation parameters, and operation response body/header types
• rest-client providers report explicit helper dependencies for generated collection parameter null checks, including ChangeTrackingList and ChangeTrackingDictionary
• generated body-only references are still handled for static helpers
• public discriminator subtypes stay public by matching Roslyn public-reference-map derived-class behavior
• current union/variant roots participate in internalization, matching Roslyn _typesToKeep behavior
• remove reachability includes current discriminator derived-model edges after root discovery
• previous generated files under src/Generated are not scanned; reachability is based on current providers/generated workspace plus custom/API roots, matching Roslyn's workspace inputs
• internal/non-public discriminator constructor/property references do not make discriminator enum types public

Azure SDK DPG Regen Note

Azure/azure-sdk-for-net#60128 completed a full DPG regen with no sdk/**/api/** changes.

The regen has a few extra generated internal model/serialization files. This is expected: the hybrid provider map keeps conservative client body dependencies from service metadata to avoid broad Roslyn body-reference scans and preserve correctness for body-only generated references. These files are internal implementation details and do not affect public API.

Validation

Local validation performed while stabilizing the PR included:

• dotnet build packages/http-client-csharp/generator/Microsoft.TypeSpec.Generator/src/Microsoft.TypeSpec.Generator.csproj -c Release
• focused generator tests covering post-processing/workspace/customization scenarios
• full generator test assembly: 1530/1530 passed
• regenerated and built previously failing generated projects/scenarios, including:
  • Sample-TypeSpec
  • Spector/http/authentication/api-key
  • Spector/http/parameters/collection-format
  • Spector/http/documentation
  • Spector/http/special-headers/repeatability
  • discriminator inheritance scenarios used by Spector tests

Implemented Generated Dependency Handling

This PR now avoids Roslyn body scanning for several generated cases:

• collection-result body dependencies are reported by CollectionResultDefinition
• client body dependencies are reported by ClientProvider
• rest-client helper dependencies are reported by RestClientProvider
• serialization provider helper dependencies are reported by serialization providers
• model factory is treated specially so unreachable model factory methods do not root models
• non-root MRW context/buildable attributes are excluded from model reachability

Custom/shared code references still use Roslyn because arbitrary user C# can reference generated types in ways providers cannot reliably describe.

Follow-Up Performance Opportunities

Potential next improvements, ordered by reward/risk:

Rank	Improvement	Reward	Risk	Notes
1	Precompute name lookup maps for AddMatchingName	Medium	Low	Avoid repeated full-node scans for helper/root matching.
2	Cache flattened provider lists and provider names	Medium	Low	Avoid repeated lazy provider/name materialization during analysis.
3	Conservative custom-code syntax prefilter	Medium	Medium/High	Can reduce custom Roslyn semantic work but must not miss arbitrary custom references.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@typespec/http-client-csharp@10976

commit: b8a5c3a

[github-actions]

No changes needing a change description found.

[azure-sdk-automation]

You can try these changes here

🛝 Playground	🌐 Website	🛝 VSCode Extension