Skip to content

Repository navigation

issue-17-progress.md

Latest commit

 

History

History
139 lines (119 loc) · 8.35 KB

issue-17-progress.md

File metadata and controls

139 lines (119 loc) · 8.35 KB
Raw

Issue #17 — optional simulation progress

Review-time evidence, 2026-10-17. Proposed application build 0.1.0.8 from accepted main 2fcfc900d99b59380b4822fc242bfb7389ed2478 (0.1.0.7). Owner separately authorized the explicit schema-4 opt-in described in the design. No user library was migrated during testing.

Implementation and boundaries

  • Structured run.progress(current, total, stage=...) works for editor recordings and joined CLI/portal runs. Known totals display a bar; missing totals remain unknown. Final status stays independent of the last report, including 100%.
  • Explicit step and percent adapters parse bounded, whole plain-text lines, including CR updates, without changing retained stdout/stderr. Default is off; structured reports take precedence. No argument-derived totals, regex execution or ETA inference.
  • Only explicit confirmed enablement upgrades schemas 1–3 to 4. Active runs block migration. Startup does not migrate. Older builds reject opted-in libraries.
  • A separate latest-state table and atomic checkpoint keep telemetry out of scientific events, measurements and immutable finalized execution records. Normal writes coalesce to at most one per second per run; final flush is exempt. Recovery, concurrent writers, backup/restore and run/screenshot deletion retain the existing scientific-data boundaries.

Automated checks

Linux, dedicated development Python, disposable directories under /tmp:

PYTHONPATH="$PWD/src" /home/Nozba/Codex/SimShelf/environments/SimShelf/bin/python -B -m pytest -q -p no:cacheprovider \
  --basetemp=/tmp/simshelf17-final-verified \
  tests/test_progress.py tests/test_recording.py tests/test_capture_backup.py \
  tests/test_removal.py tests/test_screenshots.py tests/test_portal.py \
  tests/test_shutdown.py tests/test_application_version.py

125 passed in 27.37s, one existing Starlette/httpx deprecation warning. Coverage includes explicit consent and rollback, schemas 1/2/3, invalid reports, resets, infrequent updates, 10,000 coalesced reports, concurrent threads/processes, checkpoint interruption/replay, delayed reports, failure/interruption, raw CR logs, structured priority, CLI/portal joining and reconnect, capture/metrics/annotations, active backups, corrupt restored telemetry rejection and deletion cleanup.

TypeScript tsc --noEmit and node frontend/build.mjs passed (the package scripts' underlying commands, invoked with the runtime's absolute Node path). The three tests in frontend/progress.test.mjs passed after esbuild bundling for Node: unknown values, accessible escaped known values/status/source, invalid totals and early rounding. No dependency or lockfile changes. Full third-party licenses are byte-identical; generated app.js.LEGAL.txt only reorders the existing React/JSX notice blocks.

Source and installed browser checks

Chromium via the local browser, own synthetic model and disposable libraries:

  • Legacy-library settings: compatibility/recovery disclosure, Cancel initial focus, consent required, cancellation without enablement, explicit enablement, and no repeated migration prompt once schema 4 is active.
  • Actual portal text run: observed live 43%, 53%, 58%, retained across reload, then completed at 100%. Raw log interpretation is identified as such.
  • Actual editor runs: active unknown state without percentage; structured stage Solve and live 4%, subsequently interrupted with final 22% retained.
  • Dark desktop, Light 390px, and installed Dark 320px layouts: progress remains readable without document horizontal overflow (305px document at 320px).
  • Installed restored library: interrupted 22% report, stage/source/time and application 0.1.0.8 verified in Dark narrow and Light desktop. Final browser warning/error log was empty. Theme/viewport restored and own tab closed.

Local screenshots (temporary, not distributed package artifacts): /tmp/simshelf17-source-dark-live.png, /tmp/simshelf17-source-unknown-live.png, /tmp/simshelf17-source-structured-live.png, /tmp/simshelf17-installed-dark-320.png, and /tmp/simshelf17-installed-light-final.png.

Packaging and compatibility

Built wheel and sdist from an external disposable source copy, then a wheel from the extracted sdist. All metadata reports 0.1.0.8; all bundled static bytes match between both wheels. Installed the wheel into an external disposable environment (system-site-packages enabled; this was not a fully isolated dependency audit). Python, CLI, workspace API, OpenAPI and static routes passed with Git/Node absent from PATH; API startup checks denied outbound socket connections after dependency installation. Browser used the installed wheel with the same minimal PATH.

Verified backup/restore before enabling the source fixture. Backed up four runs after enablement and restored with the installed wheel: two completed and two interrupted, all latest progress equal. Archived accepted-main source was verified as actual imported version 0.1.0.7: it rejects schema-4 startup and restore without creating the restore destination. No historical backup or source/output changed.

Fresh fetch before handoff found accepted main unchanged; the application-version gate passed against origin/main.

Failures, corrections and limits

  • Restricted-sandbox HTTP TestClient stalled; interrupted and reran the focused checks outside that restriction. Final suite passed.
  • Initial test assumptions were corrected for existing Stop HTTP 202 and stale concurrent report suppression. A corrupt-backup fixture initially edited only a WAL sidecar; switched fixture to DELETE journal mode before archiving, then confirmed corrupt telemetry rejection. Unknown-progress frontend test caught an unnecessary 100% in explanatory copy; removed it and reran successfully.
  • Final review corrected the no-report persistence-error heading on finished runs from Running to Progress unavailable, with a frontend regression assertion. Typecheck, frontend tests and bundled/package assets were rebuilt afterward. This shell had no npm on PATH, so final frontend checks used Node directly.
  • An initial old-main archive command used the wrong working directory. Its apparent pass was discarded; reran with explicit repository path and verified imported module/version before accepting compatibility evidence.
  • Two disposable UI server sessions ended with SIGTERM during the validation period. A Light screenshot then showed Failed to fetch. Cause was not proven; restarted installed UI after tests and verified the successful final screenshot and empty browser error log. The failed image is not used as passing evidence.
  • This verifies Linux/Chromium only, not other platforms, browsers or screen readers. Bounded synthetic workloads are not week-long performance validation. Writes can wait on the workspace lock/fsync; do not report per particle. Abrupt process death can lose the pending second. Telemetry stores latest state, not a full history; stdout/stderr ordering and arbitrary terminal formats are not inferred. External reference-only scientific outputs are still not backups.

Owner verification

  1. Use a disposable or verified-restored library. Stop all recorders, including older builds. Back up and restore-check before selecting Start simulation → Progress reporting → Enable progress reporting for this library. Verify the incompatibility warning and consent requirement; Cancel should change nothing.
  2. Follow the progress guide with a harmless model. Report current/total and a stage; observe a live bar, reload, reset the stage/current, then fail or stop the run. Verify last progress and final status stay distinct.
  3. Run without reports: expect Running and no percentage. Select the step adapter for CR-delimited step 10/100 output; check retained logs, then confirm default/off ignores the same output. Do not use a production scientific run for the first check.
  4. Inspect Dark/Light at desktop and narrow widths, keyboard access to settings, and source labeling. Reopen settings: schema-4 consent should not repeat.
  5. Back up and restore the test library into a new destination; compare records, measurements, annotations, copied files and latest progress. Verify the old build rejects this opted-in library. Recheck the accepted-main version gate before merge. Review only; no release/publication is authorized by this PR.