mirror of
https://github.com/DeusData/codebase-memory-mcp.git
synced 2026-11-05 08:23:53 +08:00
Andrew Hundt
e6126468fa
Move the full implementation from run-benchmark-campaign.py to run-benchmark-experiments.py and reduce the old filename to a backwards-compatible shim that loads the new module by path and re-exports every public name, so direct imports (tests/test_benchmark_campaign.py, scripts/autotune.py, scripts/summarize-benchmark-results.py) and existing invocations keep working with byte-identical behavior. - Add --experiment-root and --allow-temporary-experiment-root, keeping --campaign-root and --allow-temporary-campaign-root as accepted aliases on both entry points. - Add --candidate-ref LABEL=REF override and an upstream/main -> origin/main -> main fallback chain so --quick/--full keep working after the api-consolidation branch is merged and its pinned refs no longer resolve; explicitly requested refs remain fail-closed. - Say experiment instead of campaign in --help text, runtime error text, and generated-report boilerplate (summarize-benchmark-results.py); keep the campaign_version manifest key, the campaigns composition key, and the .worktrees/benchmark-campaign/ directory so every retained runset stays valid. - Move docs/BENCHMARK_CAMPAIGN.md content to docs/BENCHMARK_EXPERIMENTS.md and leave a pointer stub so existing links keep resolving. - Add tests/test_benchmark_experiments_shim.py (25 subtests) proving the shim re-exports the implementation and both CLIs accept old and new flags; update the temporary-root error-text assertion. Verified: ruff clean on all touched files; 171 passed, 1 skipped across the four benchmark test files; --help output identical between the two entry points except the program name. Signed-off-by: Andrew Hundt <ATHundt@gmail.com>