cuopt-developer

Building and testing cuOpt CUDA kernels requires activating conda, verifying CUDA compatibility, setting PARALLEL_LEVEL to prevent OOM, and running canonical build and test commands before submitting any CI contributions.

Contributing to cuOpt internals requires strict adherence to DCO signoff, minimal diffs, and safety rules: no CI or pre-commit bypass, no sudo, no installs outside dependencies.yaml, and no destructive commands during PR preparation.

Adding solver parameters or extending server endpoints in cuOpt requires modifying the correct internal files for your change type while following the end-to-end workflow for C++/CUDA, Python bindings, and server components without breaking conventions.

Your cuOpt build pipeline may fail if pre-flight checks are skipped, such as missing dataset availability, incorrect CUDA compatibility, inactive conda environments, or unset PARALLEL_LEVEL causing out-of-memory errors during the build process.

No, you cannot bypass CI checks or pre-commit hooks when debugging C++ development failures in cuOpt; the workflow enforces non-negotiable safety rules including no CI bypass, no sudo, and strict DCO signoff compliance.

The best way to prepare a cuOpt pull request is to confirm your component and change type, run required pre-flight validations, ensure minimal diffs, apply DCO signoff, and verify all build and test commands pass locally before submission.