Revival & Build (2025–2026)
Deep-dive page: how a 2011-vintage Java/OSGi project was brought back to a running, measurable system in 2025 and modernized for tooling in 2026. See also Architecture and the timeline.
Supersession note: the executable facts on this page come from the 2025 master thesis (Reviving LibreSocial, Imam 2025) and its companion setup guide (Running LibreSocial — Setup Guide, Imam 2025). The earlier draft (TH Bingen draft) is superseded by the final version. For the 2026 Gradle/Java-17 toolchain, the repository build files are now authoritative; the JDK-1.7/Eclipse-Indigo instructions below describe the legacy binary-first path that the 2025 evaluation used.
Why a "binary-first" revival
LibreSocial had not been compiled since ~2016. Its OSGi bundles, SWT/JFace GUI and FreePastry libraries pin an old toolchain, and an immediate upgrade of all dependencies at once would mix build breakage with behavior change. The thesis goal (F1–F4) was therefore: get the existing code running unmodified in its architecture, then measure a baseline — leaving modernization as a controlled next step.
The pinned legacy environment (2025 path)
| Component | Exact version used | Why it must match |
|---|---|---|
| JDK | 1.7.0_80 | bundles target Java 7; newer class files break SWT GUI start ("Unsupported major.minor version") |
| Crypto policy | Unlimited Strength JCE (local_policy.jar, US_export_policy.jar in jdk1.7.0_80/lib/security) |
ECC-based framework crypto needs full-strength keys |
| IDE | Eclipse for RCP/RAP Indigo (20110615-0604), compiler compliance Java 1.6/1.7, runtime JavaSE-1.7 | modern IDEs cannot import the legacy .project/.classpath PDE setup |
| GUI libs | the SWT/JFace versions shipped with the release | OSGi imports/exports must resolve without patching |
| Overlay libs | consistent FreePastry/PAST/SCRIBE jars | mixed versions cause runtime ClassNotFound inside bundles |
Git work should be done outside Indigo (Git CLI, VS Code, etc.): Indigo's Git integration is
unusable. Repository: gitlab.rlp.net/k.graffi/libresocial (2026 course: gitlab.cs.hs-rm.de/libresocial);
in 2025 the active branch was 2025_LibreSocial_Docker.
Reproducible evaluation bed
The thesis introduced the run-hygiene rules the 2026 project course now automates:
- Directory layout per run:
runs/<id>/withconfigs/(exact.propertiesused),metrics/(GLOBAL_*CSV time series from the monitoring),plots/,README.txt. - Plotting via gnuplot (
plot_all.sh), ~900 generated diagrams, 20 selected for the thesis. - Discipline rules: exactly one monitoring controller per experiment, warm-up ≥ 5 min, fixed random seeds, identical configs across scenarios, NTP synchronization.
- Health gates before measurement: stable node count, exceptions ≈ 0, sent ≈ received.
- Do not change plugins mid-run — a plugin change restarts bundles and invalidates the run.
Scenarios and baseline numbers (still the reference baseline in 2026)
| Scenario | Setup | Purpose |
|---|---|---|
| S1 baseline | N = 2 nodes, replication k = 1, 5 msg/s nominal, 600 s | sanity + steady state |
| S2 stability | 1,800 s | drift/leak check |
| S3 replication sweep | k sweep; k = 3 and k = 5 not run because k > N | replication cost |
Steady-state medians measured on modern hardware (single host/LAN):
- Memory per node: 846.17 MB (IQR 306.68 MB)
- Exceptions: 0 (NullPointerException and others eliminated by the fixes)
- Send/receive rate: 0.20 msg/s each (balanced)
- Pastry overlay traffic: 65.60 KB/s; hop count 1 (2-node overlay)
- Lookup latency: 11.50 ms; PAST store: 0.048 s
These numbers are a health baseline, not a performance claim: N = 2, LAN only, no churn, no WAN latency. Widening them is now open work (Docker cluster testing, test safety net).
What "modern systems" required in practice
About 120 compile/run issues were fixed during recovery — the full taxonomy with fix patterns is on the defect catalogue page. Roughly: toolchain/language-level mismatches, missing helpers/imports, OSGi wiring and resource loading, raw generics, and JDBC lifecycle handling. Architectural code (protocols, storage, overlay) was deliberately left intact.
2026 toolchain modernization (current state)
The 2026 project course moved the build forward while keeping the runtime semantics:
- Gradle + Bndtools build replaces manual Eclipse PDE work; Java 17 target (work items #104/#116; see developer-experience blog).
- FreePastry merged in as a maintained las2peer fork, so the overlay is no longer a dead binary dependency.
- Docker multi-node cluster brings the missing N > 2, multi-host dimension (work items #108, #131–133; blog post).
- Tester + automated system tests are being rebuilt on top of this bed (testing methodology, tester component).
Known open problems (as of 2026-11)
- S3 replication sweep incomplete (k ≥ 3 needs larger N — addressed by Docker cluster).
- Baseline measured on one host/LAN; WAN latency and churn unmeasured since 2011.
- Java 7 pin is only soft-superseded: the runtime bundles and GUI still expect the old SWT/JFace combination in parts; the Gradle path is the way forward.
- Security review of the storage path explicitly requested by the thesis (see security roadmap blog).
Sources
- Reviving LibreSocial: Error Fixes and a Baseline Evaluation on Modern Systems — F. Imam, master thesis, TH Bingen, 2025 — fixes, runbook, baseline metrics.
- Running LibreSocial: A Legacy Java Project Setup Guide — F. Imam, 2025 (final) — toolchain, JCE policy, troubleshooting.
- TH Bingen draft version, 2025 — superseded draft of the same guide.
- Repository build files and GitLab work items (2026) for the current Gradle/Java-17/Docker state.