eclipse-timekeeper

Test execution on Java 21

The maintainer requested that tests execute before continuing the remaining upgrade steps. This brings the persistence compatibility and test-runner work from steps 4/5 forward without changing the database format.

Changes

EclipseLink 2.7.16 and its ASM 9.8.0 bundle replace EclipseLink 2.7.3/ASM 6.2. The target pins the official release repository and exact bundle versions; the database manifest requires compatible versions. javax.persistence 2.2.1, H2 1.4.194 and the existing entity mappings remain unchanged.
Surefire 3.5.5 runs JUnit Platform with Jupiter and Vintage 5.13.4, preserving JUnit 4 tests and discovering both parameterized JUnit 5 report cases. Maven Compiler 3.14.1 compiles tests with release 21.
Each database test gets a unique in-memory database, closed with its connection pool afterwards. Persistence assertions clear both JPA caches before reading.
Report fixtures now populate task IDs and projects through the existing model. Previously undiscovered tests tried to persist tasks with null primary keys. Reports use closed UTF-8 writers and assert project, task and activity content. Generated reports are under target/test-reports and included in CI artifacts.
SWTBot’s runtime explicitly includes the SDK product and SLF4J provider. Startup polls the database’s latched ready state for at most 30 seconds, avoiding a missed-notification race. The forked UI process has a 180-second timeout. Both test runners fail if no tests are found.
UI setup uses the workbench shell rather than assuming an active OS window; workweek screenshots also use that shell, avoiding zero-sized layout containers. CI retains the Eclipse runtime log for failures not reflected in test counts.

Verification

Verified September 21, 2026 after the interrupted-activity recovery merge: macOS 27.0 / aarch64, Temurin 21.0.12.1, Maven 3.9.16. The build ran from a clean worktree, excluding local IDE metadata and uncommitted classpath edits. The normal Maven dependency cache was used; this was not an empty-cache verification.

mvn -B -ntp clean verify -Dtycho.localArtifacts=ignore

No skip flags were used. The clean build completed in approximately 31 seconds. The generated Surefire XML reports and p2 repository were inspected in their module target directories; these generated files are not committed.

Suite / check	Result
Database/report/packaging suites	64 passed
`TemplateTest`	2 bundled report templates rendered and content checked
`IntegrationTest` (SWTBot/JUnit 4)	6 passed
`WeekViewContentProviderTest` (SWTBot/JUnit 4)	4 passed
`IdleTimeDetectorTest` (SWTBot/JUnit 4)	1 passed locally and on Linux/X11, macOS and Windows CI
Six-project reactor and p2 repository/ZIP	Passed
Current-model fixture, restart, copy and SQL restore	Passed with Java 21

The automated acceptance coverage maps to the upgrade-plan item as follows:

Time tracking: IntegrationTest.testTaskActivationAndDeactivation verifies activity creation, closure, duration and workweek totals.
Manual editing: IntegrationTest.testEditTimeRange edits a workweek cell and verifies the resulting duration.
Labels and restart: the current-model fixture verifies two labels, three assignments, relationships and labelled totals after closing and reopening file-backed storage in StorageModesTest and FileStorageTest.
Report export: TemplateTest renders every bundled template and checks project, task and activity content; SWTBot also invokes both clipboard export paths from the workweek view.
Import/export: IntegrationTest.testExport checks current-schema CSV headers and performs a non-empty export/import round trip.

No active tests were disabled. The UI run also covers deleted-Mylyn-task history, preference storage and UI-thread refresh behavior.

Remaining limitations

The manual-edit test remains guarded on Linux because SWTBot can focus the workbench’s Find Actions editor instead of the workweek cell editor.
Template files are content-checked directly, while the SWTBot clipboard test verifies command/menu execution rather than reading the OS clipboard payload.
Installed-IDE acceptance outside macOS remains limited, and pure Wayland has no native idle detector. See platform acceptance. No personal database was used.

Sources

This site is open source. Improve this page.