[AGiorgetti]

Summary

Fixes #88.

• Give diagnostic instrumentation a dedicated ordinary-type activator that follows native DI constructor preference and parameter-type-set ambiguity rules.
• Ignore ActivatorUtilitiesConstructor on ordinary type registrations, accept equivalent/permuted parameter sets, and reject shorter non-subset constructor ambiguity before activation.
• Preserve contextual key attributes, optional defaults, null factory results, generic-constraint checks, constructor exception types, lifetimes and disposal ownership.
• Leave explicit non-empty DependsOn maps and decorator constructor selection on their existing paths.
• Add 144 regression cases and correct an older keyed test that explicitly expected diagnostic mode to choose a different constructor.
• Update README consumer guidance and the top changelog fixes section.

Test-first verification

• Reviewed develop baseline: 335d3266ba5756af8230fd9a1ff64ce98375ba0c; 776 baseline tests passed on .NET 10.
• Initial regression pass reproduced 69 failures on each matching .NET 8/9/10 runtime before the implementation change.
• Extended focused baseline controls: 71 failed / 71 passed on each .NET 8/9/10 runtime. An independent review then added a constrained-generic candidate case: 2 failed / 2 passed before its targeted fix.
• Final Release solution build passed, including the netstandard2.0 library and net472 test assembly. Only the two pre-existing net472 package-support warnings remain.
• Final full suites passed: 920/920 on each of net8.0, net9.0 and net10.0, using DI 10.0.0 and matching installed runtimes (8.0.31, 9.0.20, 10.0.12).
• Native hot-cache regression passed on each modern target: 100 interleaved enumerations after observed native compilation.
• Independent review finding was reproduced and fixed; no remaining blocker in this targeted change.
• Windows CI, including executing net472 and its hot-cache regression, is pending at creation and will be checked on this exact head.

Scope

Cycle handling is tracked separately in #89. This change does not introduce recursive graph validation, optimize activation, change workflows, or publish a release. The PR remains a draft for review.

Prepared with OpenAI Codex.

[AGiorgetti]

Test-first checkpoint for dae92c8bfd46525ba2ed0659930a2dddbdaaf916:

• The regression suite reproduced the diagnostic constructor-selection failures before implementation on .NET 8, 9 and 10.
• The dedicated diagnostic activation path now passes all 920 tests on each matching modern runtime. Full Release build succeeds; the two existing net472 dependency-support warnings remain.
• Native hot-cache controls pass on all three modern targets. Independent review also caught and verified the fix for constrained generic bindings in unselected constructor candidates.
• The GitHub tree SHA matches the tested local tree. Windows CI will provide the net472 execution check; I am monitoring this exact PR head before calling it ready.

The PR remains a draft. Cycle handling (#89), optimization work and release publishing are outside this change.

Prepared with OpenAI Codex.

[AGiorgetti]

Exact-head verification complete for dae92c8bfd46525ba2ed0659930a2dddbdaaf916.

Windows CI run 37440533272 succeeded. The job logs confirm 920/920 tests passed on each of net472, net8.0, net9.0 and net10.0, followed by four successful native hot-cache regressions (100 interleaved enumerations each after observed native compilation). Pack and Publish were skipped.

After fetching the published head locally, the Release build, all three modern 920-test suites, and their hot-cache controls also passed again. The only build warnings are the two existing net472 dependency-support warnings.

GitHub reports no merge conflicts. There are no submitted reviews or unresolved review threads at this checkpoint. This PR remains a draft for review; issue #88 has not been closed, and no subsequent issue work has started.

Prepared with OpenAI Codex.