[Neftedollar]

Summary

• Support Orleans 10.4.2 without raising the 10.1.0 package floor or changing public F# APIs/wire bytes. Align all seven hand-written transport/observer copiers with the nullable Orleans contract.
• Upgrade the newest CI/Redis matrix and add a package-consumer smoke: the same floor-built nupkgs run on 10.4.2 without rebuilding Orleans.FSharp.
• Fix the Dashboard OpenTelemetry downgrade; update vulnerable build-only SourceLink and website dependencies without disabling audits.
• Strengthen tests for retry count/terminal exception, independent state-machine outcomes, JSON-vs-binary routing, key isolation, deactivation, persisted-vs-memory-only state, and actual TaskSeq item sequences. Direct registered-copier null tests cover all seven implementations.
• Native ResumeAsync controls capture the changed checkpoint boundary: the 10.4.2 memory provider replays the checkpoint event. Wrapper assertions require the exact native suffix; new subscriptions require the exact inclusive sequence.
• Update canonical/site docs and generated LLM corpus with 10.4 changes, upgrade caveats, TaskSeq status and explicitly deferred features.

Verification

• All mandatory CI checks, including Release Gate, passed for head 5489ea898c72af24686da8715d3b57756e59e3b1: run 41247956529. Publication is intentionally skipped on a PR.
• The CI package artifact contains exact version 5.0.2-ci.41247956529.1. Downloaded and revalidated its six packages, five symbol packages, and SHA-256 manifest after CI.
• Both solution matrix builds: zero warnings/errors.
• Unit suites: 2191 passed on each Orleans version.
• Full newest integration suite with real Redis: 605 passed, zero skipped.
• After the final stricter stream-filter review fix: all 18 newest streaming integration tests passed again.
• Floor integration suite with Redis: 604 passed; the one skip is the explicitly 10.3+ stateless-stream test.
• Six 5.0.2-rc.1 packages plus five symbol packages built/validated locally; generated template builds, six tests pass, and its live counter transcript passes on floor and 10.4.2 using the same package bytes.
• Packed analyzer runs in the matching CLI host and emits only the expected OF0001 diagnostic.
• Dashboard live API reports both functional actors on 10.4.2; the C# chat-room consumer runs successfully.
• Website build, link checks, docs mirror/API/archive checks and LLM corpus check pass; npm audit reports zero vulnerabilities. NuGet transitive audit reports no known vulnerable packages.
• Separate-context adversarial review found two assertion gaps; both were fixed. Controlled bad implementations/test seams were exercised for retry, state-machine, codec routing, and key collapse; the strengthened tests rejected them.

Boundaries / release

This prepares a 5.0.2 patch, but does not tag or publish it. Current stable documentation still identifies 5.0.1; update release status/install instructions when publishing.

TaskSeq remains 1.1.1 and SDK 10.0.201 remains pinned. The duplicate-final-item issue is not fixed by this change, and green characterization tests do not declare those wrappers safe. The F# compiler backport has not shipped in stable SDK 10.0.403.

EarliestAvailable/Latest F# subscription helpers are a separate feature; DurableGrain and Durable Jobs remain deferred prereleases. Rolling tests exercise application N/N+1 and rollback, not a cluster mixing Orleans 10.1/10.4. The floor-built package smoke covers a live counter, not every provider/transport combination. This test audit is bounded, not a claim of exhaustive mutation coverage.

The checkpoint behavior change was traced to Orleans PR #10936, present in 10.4.0: explicit start tokens remain inclusive during handshake renegotiation. It is distinct from the unresolved TaskSeq final-item duplication.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T22:23:40.895631Z	643cc64	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.