NetGraph Development Workflow

For NetGraph contributors. For writing and running scenarios, see the netgraph-dsl skill.

Development Setup

make dev  # Creates venv, installs deps, sets up pre-commit

Requires Python 3.11+.

Code Iteration

Workflow

1. Make changes
2. make qt - fast feedback during development
3. make check - complete validation before commit
4. If API changed: make docs to regenerate API reference

Performance Benchmarking

For algorithm complexity analysis and regression testing:

make perf  # Run all benchmark profiles with analysis and plots

Output in dev/perfresults/ (JSON) and dev/perfplots/ (graphs).

Perf CLI

# List available benchmark profiles
./venv/bin/python -m dev.perf.main show profile

# Show profile details
./venv/bin/python -m dev.perf.main show profile spf_complexity_clos2tier

# Run specific profile
./venv/bin/python -m dev.perf.main run --profile spf_complexity_grid2d

# List topology types
./venv/bin/python -m dev.perf.main show topology

Available profiles: spfcomplexity, maxflowcomplexity_ (Clos and Grid2D topologies).

Running Commands

Use direct venv paths (avoids shell state issues):

./venv/bin/python -m pytest tests/dsl/
./venv/bin/ngraph inspect scenario.yaml

Dev Tools

Code Standards

Python Practices

• Use modern type hints on public APIs; avoid Any where practical
• Provide clear docstrings for public APIs (arguments, return values, important errors)
• Use the project's logging facilities instead of print() in library code
• Prefer algorithmic improvements over micro-optimizations

Testing

• Add or update tests for new behavior or bug fixes
• Keep tests readable and focused; use fixtures to avoid duplication
• When tests fail, fix the code rather than weakening tests (unless test is incorrect)

Completion Checklist

A change is complete only after:

1. make check passes (auto-fixes, schema, lint, type check, tests)
2. Relevant tests cover the new behavior

Pre-commit and CI

Pre-commit hooks run automatically on commit. To run manually:

make check  # Runs format, lint, type checks, tests

CI runs the same checks. If CI fails:

1. Pull latest changes
2. Run make check locally
3. Fix any failures
4. Push again

Troubleshooting

Venv Missing

make dev  # Recreates if missing

Import Errors After Pull

./venv/bin/pip install -e '.[dev]'

Pre-commit Failures

make format  # Auto-fix formatting
make check   # Verify all checks pass