Useful contributions include application workloads, stronger failure models, independent replay backends, calibrated timing parameters, and Zettai topology coverage. Small, reproducible examples are particularly valuable.
Use Python 3.11 or newer; these checks use only the standard library:
python3 -m unittest discover -s targets/splash-type2 -p 'test_*.py' -v
python3 -m unittest discover -s targets/splash-bsp -p 'test_*.py' -v
python3 -m unittest discover -s targets/splash-endpoint -p 'test_*.py' -vThe GitHub workflow runs these checks. It does not run the external QEMU build or claim to reproduce a simulator campaign. Full integration instructions are in the endpoint guide.
- State the hypothesis, work allocation, baseline, and modeled resources.
- Retain fixed inputs, all declared configurations, and explicit failure outcomes.
- Use a fresh artifact directory. Preserve failures and do not overwrite a campaign.
- Compare values and resource schedules to independent oracles. For bridge changes, include a disabled-forwarding control and verify unaffected branches.
- Report virtual time, host wall time, and hardware measurements separately.
Large raw campaigns stay outside Git. Include source, compact summaries, provenance, and reproduction commands in a pull request. Do not replace measured data with estimated numbers without labeling the estimate.
Open an issue with the commit, QEMU binary/source revision when applicable, exact command, expected behavior, and smallest failing configuration. Include the first failed assertion and relevant log excerpts. Remove credentials and unrelated private paths before sharing logs. For a new workload, explain its data dependencies and what the replay boundary must cover.
The Rust PTX/RTL stack is separate from the Python/QEMU experiments. Contributions there should identify the crate and required vendored frontend/RTL inputs, and run the relevant Cargo checks. See the source and existing design notes before assuming that a Python-only checkout includes every Rust build dependency.