Issue: #220. This extends the remote Eclipse client and synchronization acceptance. It does not clear the remaining native preference/editor workflows or the 2.0 release.
The optional Mylyn fixture now retains one installed Eclipse workspace through seven separate JVMs. The first three still verify remote activation, restart, account switching, failures and peer synchronization. The next four follow the actual backend preference page through:
Both servers are independently launched from the checksum-verified packaged Quarkus distribution, with distinct loopback ports and database files. All data, credentials and the workspace belong to the disposable acceptance fixture. Every runtime starts both servers so their retained data can be checked while the other backend is selected. The native-only fixture retains its two phases and does not launch the second server.
Applying a mode or endpoint change must show the restart notice while preserving the running backend identity, task snapshot and actual embedded connection object. Server URL/token controls enable only in server mode; the token remains encrypted and its input uses the password style. After restarting, the actual installed client must use the mode and URL saved by the preference page. The bootstrap initializes backend preferences only in the first phase and then honors their persisted values.
The new server and local database must initially have no tasks copied from the previous backend. Native Workweek controls create projects/tasks with the same visible names in each backend. Actual Mylyn activation/deactivation must record time against that backend’s own task UUID and owner. Foreign saved projections must reject resolution and remain excluded from the selected backend.
Each backend’s projection receives a random test marker, saved through Mylyn’s official externalization operation. Returning to server A and to local storage must reconnect the original handle and marker, resource UUIDs and exact recorded intervals. Recreating an unmarked projection cannot satisfy those checks. Owned metadata revisions/state and interval state are hashed in the credential-free receipt, then compared across subsequent launches. This checks that switching does not transfer records or alter the inactive backends.
The local fixture’s initial checksum stays unchanged throughout the remote-only phases. The deliberate embedded phase then opens that database and creates local records. The next remote phase accepts the resulting checksum only after the embedded acceptance marker has passed, records it separately, and requires the app connection to be null with that file unchanged again. Local record and interval assertions prove persistence on the final embedded restart.
Verified locally on October 13, 2026, with Temurin 21.0.12.1 / macOS aarch64:
108 main tests plus seven separate restart/account/switch tests passed,
with no failures, errors or skips. This includes all nine installed phase
reports: two native-only and seven optional-Mylyn phases. All previous receipt
markers and all five backend-switch markers passed. Both server fixtures record
seven distinct process IDs. Server ZIP SHA-256:
739e932b5196620eaa9175243593564b03edf109b37d63970b6e77de1f33b9eb.
The selected reactor runs shared domain, HTTP and database tests; counts across separate reactors overlap and are not summed. The local task-state checkpoint is captured after activation/stop, because JPA updates the task version through its activity relationship. Restart must preserve that acknowledged final state.
The new reports are retained in
surefire-reports-server, surefire-reports-local, surefire-reports-return
and surefire-reports-local-restart. CI uploads these alongside the existing
reports and credential-free receipts/logs. Keyrings, password files and token
registries remain excluded.
The parent synchronization commit f3fbf3c passed all seven Build jobs in
run 38313548997
and Server containers verification in
run 38313549045.
Its downloaded installed evidence confirms all five phases and the four
peer-synchronization markers. These are parent-commit results; CI must verify
the new switching commit separately. That verification subsequently passed:
all seven Build jobs for 81e205a in
run 38316380367
and Server containers verification in
run 38316380282.
Downloaded reports confirm all nine installed phases and five backend-switch
markers, with fixture secrets excluded from the artifact.
Native label/report/CSV preference flows and other form/uncertain-operation failure paths remain in #220. #221 web workflows, bidirectional web/Eclipse synchronization in #219, the separate Mylyn Error Log gate #243, final tagged artifacts and public GitHub p2 installation remain required before 2.0.