eclipse-timekeeper

Installed-IDE acceptance

Verified September 21, 2026, after PR #196 merged (0cbf453). Historical migrations remain out of scope. All records and workspaces used here were synthetic; no personal Timekeeper database was opened.

Environment and artifacts

• Eclipse Platform Runtime 4.41, build R-4.41-202608281142, macOS Cocoa aarch64. This was a pristine Platform installation, not the Tycho test application or a preconfigured IDE package.
• Temurin Java 21, Maven 3.9.16, the repository’s dated 2026-09 target.
• Official runtime download. SHA-512 matched the release’s buildproperties.json: 5a35d58899a1400c8b0447c0e3a34b8124637c7c0da78326c2a1dac212ff9d61e0e80b307b9dd0c2a4ae1ae2cc9dacf526d7639267f812759b043e534c9737a1.
• Clean build: mvn -B -ntp clean verify -Dtycho.localArtifacts=ignore. All six reactor projects passed: 63 database/report/packaging tests and 9 UI/integration tests passed; current-schema CSV export is active again.
• Verified p2 ZIP: net.resheim.eclipse.timekeeper-site/target/net.resheim.eclipse.timekeeper-2.0.0.202609211009.zip. This is a verification artifact, not a published release.

Local evidence is under /private/tmp/timekeeper-installed.RaDTib (installation and runtime logs, three SQL snapshots, isolated applications and data). The clean source/build is /private/tmp/timekeeper-packaging.XhVDPr, with build log /private/tmp/timekeeper-packaging-final.log. These temporary files are not committed and may be removed by the operating system.

Packaging corrections

The first clean installation resolved bundles but omitted Mylyn’s Tasks search/index contributions. Opening Task List logged: No search provider was registered. Tasks search is not available. Timekeeper now requires the complete org.eclipse.mylyn.tasks.feature, version 4.12.0 with compatible matching. Bugzilla remains an optional separate connector.

The same installation contained both the host’s slf4j.simple and Timekeeper’s explicitly included org.eclipse.equinox.slf4j. Removing the latter from the shipping feature avoids installing a second SLF4J provider. Timekeeper retains the SLF4J API and uses the host’s logging provider. The test target’s logging configuration is separate from the shipping feature.

Two metadata regression tests check these requirements. They do not replace actual p2 resolution or runtime testing.

Installation and restart results

1. Installed the PR #196 baseline (2.0.0.202609210953) into a pristine runtime using p2 director and the dated SimRel repository. Installation succeeded.
2. Opened Workweek and Mylyn Task List; created a local task named Synthetic installed acceptance, activated it and deactivated it through Workweek. Inspected Timekeeper’s database and label preferences without changing settings. The database page has no historical recovery actions.
3. Closed Eclipse and exported a read-only SQL snapshot. Schema version 1 was READY with origin NEW. Task 1 had one completed activity, no current activity, and seven default labels existed. The activity ran from 2026-09-21 12:04:18.181108 to 2026-09-21 12:06:00.993865; H2 DATEDIFF('SECOND', ...) returned 102. The minute-formatted Workweek total was 0:01 on Monday September 21.
4. Built the corrected package and installed it into a second pristine runtime. Also replaced the original installed feature with 2.0.0.202609211009 in one p2 uninstall/install transaction. Both succeeded. Both installations contained Mylyn Tasks search/index bundles and only the host’s SLF4J provider, not org.eclipse.equinox.slf4j.
5. Restarted the updated installation with the original synthetic workspace and database. Workweek retained the task, completed activity and total. The missing-search and multiple-provider warnings were gone. A full SQL snapshot comparison found only TASK_URL changing from NULL to an empty string when the local Mylyn task was relinked. IDs, timestamps, current-activity state, labels and other stored data were unchanged; this is not an exact SQL match.
6. Opened the independently pristine corrected installation with a new empty Mylyn workspace against the same database, sequentially after closing the first installation. Workweek still displayed the stored task, activity and 0:01 total without the original Mylyn task list. Closed the IDE and exported SQL again: after-clean.sql exactly matched after-update.sql with cmp.

Every GUI launch explicitly supplied a temporary workspace, temporary Java user.home, Java 21 and -Dnet.resheim.eclipse.timekeeper.db.url=jdbc:h2:/private/tmp/timekeeper-installed.RaDTib/data/timekeeper. The database URL override acts before normal storage preference resolution; these original checks therefore did not validate preference-selected storage modes.

Storage preference acceptance

A follow-up used the corrected installed feature build 2.0.0.202609211112 from PR #202 with the same Eclipse Platform 4.41 and Java 21 runtime. The subsequent documentation-only PR #203 does not change the installed bundles. A clean build of its merged main passed 63 database/report/packaging tests and 10 UI/integration tests before the installed checks. No GUI launch in this follow-up supplied the database URL system property, so the persisted preference was the only storage selector.

The installed Database preference page selected and saved each of these modes, with a normal Eclipse shutdown and restart between changes:

• workspace-relative storage at <workspace>/.timekeeper/h2db;
• shared storage at ~/.timekeeper/h2db with AUTO_SERVER=TRUE and the configured fixed port;
• an explicit file URL in a separate temporary directory; and
• an explicit jdbc:h2:tcp://localhost:19092/timekeeper URL served by a separate local-only H2 process.

Workweek opened normally after every restart. Console output identified the expected JDBC URL in each run. Each new database contained schema version 1 in READY/NEW state and seven default labels. Switching storage left the other database files intact; modification times changed only when their corresponding mode was active. The TCP database was also queried through its server after Eclipse shut down, then the temporary server was stopped cleanly.

This verifies the installed storage selector, embedded workspace storage, shared mixed mode and a local TCP server. It does not certify an externally managed remote server, authentication, encryption or network-failure behavior. Evidence remains under /private/tmp/timekeeper-installed.RaDTib; it is temporary and contains only synthetic databases.

Concurrent installed instances

A later follow-up ran two copies of the installed application with separate workspaces against one shared AUTO_SERVER database. Both read the same task; the second committed a new activity while the first remained connected, and the first client’s later shutdown did not overwrite it. The second client also saved successfully after the original server-owner process shut down.

The test exposed an important support boundary: a running client does not invalidate its EclipseLink or Workweek cache when another process commits, and the shared current activity has no client owner or lease. Restart reloads the new records. A deliberate port 9090 conflict produced a visible database- unavailable status and an unchanged before/after SQL export. See the concurrent installed-instance report for the full procedure, evidence and limitations.

Interrupted activities

The installed acceptance matrix now also covers abrupt termination and startup reconciliation. A durable open activity retained the same ID and start time across restart, SIGKILL and another restart while its Mylyn task remained active. When the same durable database state was opened with that Mylyn task inactive, startup closed the activity at its persisted last-active tick and committed both the end time and cleared current reference.

This run found and fixed two recovery defects (an uncommitted update and a 30-minute probe end that could be in the future) plus a null idle-status race. The corrected installed build was 2.0.0.202609211243; the full build passed 64 database/report/packaging and 10 UI/integration tests. See the interrupted-activity report for exact identifiers, timestamps, reproduction boundaries and the immediate post-activation durability limitation.

Repeating the checks safely

Use disposable copies of the runtime and separate temporary workspace/home/data directories. Never point this procedure at an unsupported or personal database. Build from clean sources, then use the p2 director with these arguments, substituting absolute temporary paths:

<app>/Contents/MacOS/eclipse
  -nosplash -consoleLog -vm <jdk21>/bin/java
  -data <temporary-director-workspace>
  -application org.eclipse.equinox.p2.director
  -repository file:<absolute-built-repository>,https://download.eclipse.org/releases/2026-09/202609091000/
  -installIU net.resheim.eclipse.timekeeper.feature.group/<built-version>
  -vmargs -Duser.home=<temporary-home> -Declipse.p2.mirrors=false
  -Dnet.resheim.eclipse.timekeeper.db.url=jdbc:h2:mem:provisioning

For the update transaction, additionally provide -uninstallIU net.resheim.eclipse.timekeeper.feature.group/<installed-version>. For the GUI launch omit the director application/repository/install arguments, use a separate GUI workspace and replace the in-memory URL with an explicit temporary file database URL. Close Eclipse before taking SQL snapshots using H2’s org.h2.tools.Script, with IFEXISTS=TRUE;ACCESS_MODE_DATA=r on the URL and NOPASSWORDS NOSETTINGS export options. Preserve the first snapshot for comparison.

Remaining issues and limits

• Opening Task List in the updated real installation still logs an upstream org.eclipse.mylyn.commons.ui theme error: NumberFormatException: For input string: "or" under radix 16, from E4ThemeColor.getRGBFromCssString, through GradientColors and TaskListToolTip. Inspection of the selected Mylyn source shows the parser treats an unrecognized symbolic CSS color as hex. It logs the failure and returns null. Workweek remains usable, but clean dependency-error acceptance is not complete. No upstream fix has been included here.
• EclipseLink still warns about a password encrypted with a deprecated algorithm with the synthetic database’s sa/empty-password configuration.
• Test-harness Cocoa/menu/theme messages were not reproduced as those same errors in the installed IDE. Passing tests alone does not settle runtime logs.
• This verifies embedded storage selected by the original test override, normal shutdown, sequential restart/update, all installed storage preferences and a local-only TCP server. Durable interrupted activities and abrupt process restart are characterized separately; the immediate post-activation storage window, filesystem/power failure, live cache synchronization and active- activity ownership are not certified. Externally managed remote servers, remote connectors and OS idle detection remain unverified. Manual editing and CSV import/export are covered by the follow-up UI harness, not by installed- runtime interaction.
• macOS Apple Silicon was exercised here. Linux CI is separate evidence; Windows and other desktop/runtime combinations are not certified by this run.

The follow-up clean build re-enabled current-schema CSV export coverage and added a round-trip import assertion: all 10 UI/integration tests passed, including manual activity time-range editing on macOS aarch64. The editor test remains guarded on Linux because of its known SWTBot focus issue. The CSV format is not a database backup or migration mechanism.

The follow-up lifecycle test also found and fixed a timestamp defect in Task.endActivity(LocalDateTime): the idle-time path supplied the last active timestamp, but the method discarded it and used the current clock time. The regression test now verifies that the supplied timestamp is persisted exactly.

The clean install/update portions of step 6 are verified. The upgrade is not ready for publication while the remaining runtime and platform checks are open.

This site is open source. Improve this page.