eclipse-timekeeper

Real web / installed Eclipse acceptance

Issue #221/#219; stacked on #255. The test starts Chromium from the installed Mylyn/account-B acceptance JVM and uses the same checksum-verified server as the actual p2-installed Workweek. Neither client is substituted by a fake UI.

Verified local candidate — 2026-10-13

Source: ebb17c8f340ba3d58e56111353c8a152d257d184. ZIP SHA-256: 86eaa2c2d7a8f45ded36e8e4760e63e9a1fde29283ff14caf5e848dbe15330aa. Temurin 21.0.12.1 / macOS aarch64, Node 26.9.0, Playwright 1.64.0, Chromium 156.0.8078.4. CI uses Node 24.

• Server reactor: 61 tests passed without failures/errors/skips.
• Selected installed reactor: 108 main tests plus seven secondary tests, all without failures/errors/skips. All nine installed reports pass. These counts overlap with other reactors and are not summed as unique tests.
• A real browser creates a project, task, label and one-nanosecond interval through form controls. The same UUIDs/labels/exact instants reach Workweek automatically, and its cached task report retains the one nanosecond.
• The browser changes the owner’s time zone to Europe/Oslo; the installed calendar follows it automatically, then returns to the original zone.
• A browser edit preserves an open native draft and makes Save reject it. Cancel/reopen permits an explicitly reviewed native edit. A still-open browser draft then survives that native edit and rejects its stale Save. Neither draft is silently upgraded by refresh, and server versions advance once.
• Workweek activation appears in the browser and browser Stop ends that same UUID. Browser Start appears in the actual installed active snapshot with its assigned label; Workweek Deactivate ends that interval. No duplicate intervals are recorded; both client directions observe the stop.
• The browser’s rendered daily and weekly totals match exact backend queries and the installed report cache for identical owner/zone/range. This checks native cached reports; it does not claim native totals render nanoseconds.
• Browser deletion removes intervals, task, project, label and Mylyn projection; later server/local/return/restart phases remain green. Owner A’s recorded state stays unchanged. Original local storage stays closed and unchanged in remote phases.
• Five phase.web.eclipse.* markers pass alongside previous owner/switch markers. All 20 recorded browser/server/proxy PIDs have exited.

The first run passed records, drafts, tracking and reports but failed cleanup: a Delete control was repeatedly detached while the activity table refreshed. The helper now waits for the acknowledged row-count reduction; it also closes the stdin protocol explicitly so normal exit does not require termination. The actual installed snapshot check before native Deactivate was strengthened. The entire nine-phase run then passed; no assertion was removed.

The peer uses synthetic credentials in its environment. The protocol sends only synthetic names/times/actions; credentials are not put in command lines or retained artifacts. Browser diagnostics and optional failure screenshots exclude traces/videos and private registry/keyring files. CI installs the locked dependencies before the installed runtime; missing dependencies fail the test rather than skipping the browser requirement.

Remaining release gates

CI for this follow-up must pass and its artifacts must be inspected. Parent #255 passed all eight Build jobs and Server containers verification; downloaded evidence proves eight browser scenarios and its nine preceding installed phases. This new combined scenario is additional evidence.

Remaining native label/report/CSV preference and failure/uncertain-operation acceptance, actual server-backup restoration with live clients, official Mylyn Error Log #243, human review/merge, final tagged distributions and public GitHub p2 installation remain required before Timekeeper 2.0 publication.

This site is open source. Improve this page.