AppSurface Search
Section How-to Guides
Guide

Issue 867 execution progress

Source of truth

View source Edit this page

Last updated

Plan: approved design. Verification requirements: test plan.

Scope and branch

  • Branch: codex/make-it-so-867-evidence-readiness; base: main.
  • Approved design captured in a55bc385; initial integration of origin/main at 5c87b13c through 2a2e5f0e. Final integration and verification are recorded below.
  • One neutral BCL-only package; typed fixture and Evidence.Aspire adapters reuse their existing owners. Bootstrap relocation, whole-host cleanup changes, publication and downstream adoption remain outside this change.
  • Pinned Skoolit equivalents are acceptance fixtures, not actual adoption or author feedback. No external messages are authorized or sent.

Verification ledger

Check Result
Neutral deterministic suite 77/77 passed. The completed solution collection measured 202/205 lines (98.53%) and 90/95 branches (94.74%) for the neutral package.
Typed fixture suite 45/45 focused tests passed, including ownership, original deadline, cancellation, null build, primary-error preservation and reentry.
Evidence adapter and public wrappers 32 adapter tests and 2 wrapper tests passed. The focused collector run including the host boundary passed 35/35; adapter coverage was 246/247 lines and 71/74 branches.
Real Aspire resource proof All 4 example cases passed in the second solution run, including default/custom probe clients and the EvidenceHost health → HTTP → producer → manifest → cleanup path.
Packed neutral and typed programs Standard functional QA passed at clean e91ab9e8, package 0.1.0-issue867.20261016042507.80091. The neutral package has no dependencies; the typed program verified first-party archive/cache hashes and two real Aspire/DCP startup equivalents. The committed immutable Git-export replay also passed at clean bda92c18, package 0.1.0-issue867.20261016050750.86571, preserving both real Aspire startup equivalents.
Two pinned gate equivalents Passed both startup equivalents against pinned Skoolit a589f164ef3346d4ecf8bf39b8a73ec71ea1cb1e. No downstream adoption or browser/business claim.
Package catalog and artifacts At e91ab9e8, catalog verify and gate passed (61 entries; 3,634 source paths), and verify-packages passed all 50 artifacts for 0.0.0-ci.issue867.
Formatting and diff whitespace C# whitespace formatting passed. Final proof-script syntax and merge-base diff whitespace checks pass; plan metadata hard-break whitespace was removed.
Enhance review and QA Initial two-cycle review ended clean before its checkpoint. The second pass added failure-path tests and found a package-proof source-attribution race, repaired with an immutable Git export. Focused native rereview completed; its custom artifact-directory hypothesis was disproved in disposable Git repositories because Git ignores the newly created empty directories. The full third solution run passed at 59aa9491. Standard functional SDK QA passed; no browser UI is introduced. A subsequent ship review found the shared-application registration issue recorded below; its repair requires fresh verification.
./scripts/coverage-solution.sh Attempt 2 completed all 56 projects at frozen e91ab9e8: aggregate 95.37% lines / 89.49% branches, but exit 1 due to three RazorWire integration failures. Two release-page baselines still contain preview.13 content; the approved branch renders preview.14 plus new release navigation. Five prepared release-only candidates passed both visual tests privately with update mode off; the user approved applying these exact five candidates (D1, choice B). A dialog navigation timeout passed its isolated rerun with the original timeout unchanged. The exact third full run at clean 59aa9491 passed all 56 projects: 17,517 passed, zero failed, four existing platform skips. Aggregate coverage was 95.37% lines / 89.50% branches; patch coverage was 97.55% / 93.67%. The later lifecycle repair below requires a fresh full run.
Coverage policy Unchanged hard gate: aggregate 95% lines / 85% branches; patch 95% lines / 85% branches against origin/main, Codecov mode. A passing measurement alone does not make a failed test run pass.
Draft PR Pending final checks and ship workflow. The screenshot disposition is approved.

Solution restore completed successfully; native macOS assets were restored after regenerating the Linux example lock.

Review findings addressed during implementation

  • Preserve the original caller token through each startup phase, rather than returning the linked work token.
  • Release owned probe clients before a successful startup return. Check cancellation again after response/client disposal.
  • Invoke lifecycle cleanup callbacks outside locks and join one shared completion.
  • Preserve primary readiness errors if client/builder cleanup also fails; retain only a fixed safe secondary-cleanup category.
  • Carry the new dependency into the existing packed-testing closure and both macOS/Linux example lock files.

Measurement limits

Human onboarding time, cold/warm restore observations, reliability improvement and latency improvement are unmeasured. Controlled handlers establish retry, cancellation and ownership behavior; only real-resource runs establish native startup/allocation/HTTP integration. No package has been published.

Final verification repairs

  • The package proof now records exact nuspec package/source identities and checks restored first-party raw SHA-512 against both the local archive and isolated cache. Evidence.Aspire must retain exactly Contracts and Readiness as first-party dependencies, the supported Aspire.Hosting range, and no Aspire.Hosting.Testing edge; central transitive third-party pins remain permitted. An initial over-strict assertion was reproduced before correction.
  • The packaged readiness README uses one canonical chooser/release-hub link pair inside its generated guidance region, as required by the artifact gate.
  • Added fixture checks for a null build result and caller exceptions whose diagnostic storage is unavailable. The typed fixture now carries the same safe allowlisted cleanup type as the Evidence adapter; all 45 focused fixture tests passed.
  • Adapter coverage work targets thrown phase cancellation, cancellation/deadline during client disposal, safe secondary metadata failure, nested reentry, and combined stop-callback/lease cleanup failure. Unreachable catch wrappers around a private result-only drain completion and owned cancellation-source disposal were removed.
  • The proof harness refuses dirty inputs, exports the captured commit with git archive, and builds every producer and consumer fixture from that export with explicit source identity. Retained failure paths include the source export and generated consumers. Physical temporary-directory paths avoid macOS /var alias inconsistencies in transitive project restore.
  • A private rehearsal of the immutable-export harness passed all eight package builds and both consumers. The final committed public-script QA replay passed at bda92c18; the third full solution verification later passed at 59aa9491. Final ship verification, documentation audit and draft PR remain pending after the lifecycle repair below. No failing or interrupted run is counted as passing evidence.

Ship lifecycle repair

The final native review found that independently disposing one adapter could stop the shared Aspire application while another adapter still had an admitted wait. Health and HTTP adapters now retain separate registrations with the existing application owner. Each HTTP adapter still drains its own waits and clients; only the final registration release starts application cleanup. Concurrent or repeated application disposers join the same result, and cleanup callback reentry avoids self-await. Direct owner disposal still requires the caller to finish readiness work first. This does not change the separate whole-host cleanup scope of #859.

Focused lifetime and public composition tests, the unchanged full coverage script, package artifact validation, packed consumer proof, and documentation audit must verify this final repair before the draft PR is created. Earlier passing runs above remain historical evidence for their recorded inputs.