Fingerprint tests

Fingerprint tests hash a trajectory of per-event simulation state (event types, packet counts, etc.) and compare the result against a value stored in the project's fingerprint_store JSON file. A tiny behavioral change produces a completely different hash, catching regressions that pass smoke/statistical tests.

Upstream reference: https://github.com/omnetpp/opp_repl/blob/main/doc/fingerprint_tests.md

Baseline store

SimulationProject parameter fingerprint_store (default "fingerprint.json") points to a JSON file containing (config, run_number, sim_time_limit) -> fingerprint entries. INET convention: tests/fingerprint/store.json.

First-time seeding

update_fingerprint_test_results(simulation_project=inet_project, sim_time_limit="1s")

Output includes INSERT lines for every new entry:

[04/42] Updating fingerprint . -c PureAlohaExperiment -r 3 for 1s INSERT 856a-c13d/tplx ... Multiple update fingerprint results: INSERT, summary: 42 INSERT (unexpected) in 0:00:01.567

Re-running the update when nothing has changed gives KEEP (baseline preserved):

Multiple update fingerprint results: KEEP, summary: 7 KEEP in 0:00:00.218

Running the tests

r = run_fingerprint_tests(simulation_project=inet_project, sim_time_limit="1s")

Output:

[02/42] Checking fingerprint . -c PureAlohaExperiment -r 1 for 1s PASS ... Multiple fingerprint test results: PASS, summary: 42 PASS in 0:00:01.129

Re-run only failures, or narrow to a region after intentional changes:

r.get_fail_results().rerun() update_fingerprint_test_results(simulation_project=inet_project, working_directory_filter="examples/ethernet", sim_time_limit="10s")

Command line

Typical CI workflow starting from an empty store:

1. No baseline yet -> every test SKIPs

opp_run_fingerprint_tests --load inet.opp -p inet -t 1s

2. Seed

opp_update_fingerprint_test_results --load inet.opp -p inet -t 1s

3. Tests PASS until behavior changes

opp_run_fingerprint_tests --load inet.opp -p inet -t 1s

Debugging a FAIL

A FAIL means the live simulation trajectory no longer matches the stored hash. Two useful follow-ups:

1. Pull the stdout / fingerprint trajectory from the failing

TaskResult (see opp-repl-tasks-and-results):

failed = r.get_fail_results().results[0] ft = failed.simulation_task_result.get_fingerprint_trajectory()

compare to a known-good run

1. Use compare_simulations() or

compare_simulations_between_commits() (see opp-repl-comparing-simulations) to locate the first divergent event.

Pitfalls

• Any change to simulation time limits or random seeds CHANGES the

fingerprint. Keep sim_time_limit stable across runs or regenerate the baseline.

• Fingerprint entries are keyed by (config, run, time limit).

Running the tests with a different sim_time_limit than was stored yields SKIP (no baseline for this key).

• Commit the updated fingerprint_store JSON alongside intentional

behavioral changes — otherwise CI will keep failing.

• Fingerprint tests are deterministic only if the underlying

simulation is deterministic. Non-deterministic features (e.g. real-time scheduler, certain exponential() RNG setups without fixed seeds) will produce PASS/FAIL flapping.

See also

• opp-repl-running-simulations — underlying run machinery.
• opp-repl-comparing-simulations — locate the divergence.
• opp-repl-tasks-and-results — .get_fingerprint_trajectory().
• opp-repl-statistical-tests — coarser, scalar-level regression.