Harden C# resolution and ASP.NET graph extraction

[forhappy]

Summary

• recover malformed C# parser evidence into bounded partial facts instead of invalidating whole files
• resolve C# base classes and interfaces with namespace/import/project context while rejecting cross-language collisions
• type receiver calls through parameters, locals, fields, properties, generic constraints, interfaces, and inherited members
• extend ASP.NET Minimal API extraction for nested MapGroup prefixes, standard verbs, literal MapMethods, multiline lambdas, and named handlers
• canonicalize natural-language recall, bound persistent-store accelerator hydration, and rank exact predicate callables before generic subject types
• enforce the reviewed Kestrel request-processing and IIS middleware anchors on JSON and persistent-store backends
• reset pre-compatibility extraction, cache, publication, snapshot/store-index, query-index/ranker, overview, qualification, and semantic-diff identities to v1
• merge current main, preserving the new standalone source-navigation behavior and resolving the PR conflict

Why

The universal C# cutover exposed concrete gaps on the real dotnet/aspnetcore repository: zero-width recovery nodes caused 116 whole-file validation failures, hierarchy matching was not namespace-aware, typed receiver dispatch covered too few C# binding forms, Minimal API argument parsing was comma-sensitive, and generic role names could outrank reviewed ASP.NET implementation anchors. The first persistent-store discovery profile also hydrated up to 12,801 speculative accelerator candidates and exceeded the 30-second query deadline.

Compass does not yet promise compatibility for these internal pre-release identities, so carrying provisional version numbers forward only created unnecessary legacy surface. Existing pre-release artifacts must be discarded or rebuilt; disposable query indexes rebuild automatically.

Impact

C# graphs retain valid partial evidence from malformed or unsupported syntax, resolve hierarchy/member relationships with stronger language-safe evidence, and represent more modern ASP.NET endpoint registrations. Natural search removes question noise and canonicalizes inflections while preserving literal symbol-shaped typo resolution. Discovery caps speculative accelerator reads, recalls exact predicate names as one bounded ranked channel, and reports equal callable-name alternatives as ambiguity. Existing unresolved cases remain explicit. The change is native and deterministic and adds no Graphify runtime dependency.

Real-repository qualification

The real dotnet/aspnetcore qualification was produced with the same extraction logic before the identity-only reset to v1:

• 14,393 files extracted
• 239,676 nodes and 344,731 edges
• 0 error diagnostics and 0 invalid-range diagnostics, down from 116 whole-file validation failures
• 541.9 seconds total for the uncached JSON qualification run
• 3.93 GB maximum RSS
• published graph remains partial by contract: 1,896 nodes and 1,066 edges omitted by validation, with 9 identity collisions quarantined

The persistent-store build comparison used 3.69 GB peak RSS versus 4.93 GB for the original Compass JSON run, a 25.2% reduction. On its 239,376-node, 344,359-edge ASP.NET Core snapshot, the final persistent-store discovery runs complete within the 30-second contract:

• how are HTTP requests processed: 14.3 seconds, about 181 MB RSS; rank 1 is Microsoft.AspNetCore.Server.Kestrel.Core.Internal.Http.HttpProtocol::ProcessRequests at HttpProtocol.cs:646
• how is middleware invoked: 19.4 seconds, about 182 MB RSS; rank 1 is Microsoft.AspNetCore.Server.IISIntegration.IISMiddleware::Invoke at IISMiddleware.cs:101, with ambiguity retained

Validation

• cargo fmt --all -- --check
• cargo clippy --workspace --lib --bins --locked -- -D warnings
• cargo test --workspace --lib --bins --locked
• cargo test -p compass-query --locked
• cargo clippy -p compass-query --all-targets --all-features --locked -- -D warnings
• python3 -m unittest discover -s benchmarks/performance/tests -p test_*.py (205 tests)
• cargo test -p compass-languages --test csharp_universal_conformance --locked
• focused ASP.NET route and C# universal-resolution integration tests
• npm run typecheck:js
• npm run test:js (315 unit tests passed; one timing-sensitive canvas browser case passed on isolated rerun)
• node scripts/check_viewer_assets.mjs
• ./scripts/qualify_code_graph_v1.sh --fixtures-only
• sh scripts/check_product_boundary.sh

Fixture qualification passed all semantic, deterministic-byte, scale, and framework assertions.