Contributing to TenCirPauli

TenCirPauli accepts focused changes that preserve the separation between the pure Rust core, the PyO3 boundary, the public TensorCircuit-facing Python API, and framework boundary code.

Development checks

Run the following before submitting a substantial change:

cargo fmt --check
cargo clippy --workspace --all-targets --all-features -- -D warnings
cargo test --workspace
black --check python tests benchmarks scripts examples
ruff check python tests benchmarks scripts examples
mypy
maturin develop --release
pytest

Install the local tooling with python -m pip install -e ".[quality,benchmark]". Rust formatting and linting use rustfmt and Clippy from the selected Rust toolchain. Python uses Black for deterministic formatting, Ruff for fast linting/import checks, and strict mypy for the public package; Pylint is intentionally not duplicated unless a concrete uncovered rule justifies it.

Run python scripts/check.py --fix before staging when formatting may be needed. The tracked pre-commit hook runs the same workflow in check mode, followed by Rust/Python tests and a full local benchmark record. Enable it once per clone with git config core.hooksPath .githooks; it never commits benchmark results.

New algebraic rules require property tests or differential tests against a small dense reference. Performance claims require a reproducible benchmark that separates construction, first execution, steady execution, memory, and numerical error.

Local performance history is managed separately from CI. Install the benchmark Python extra, then use python benchmarks/run.py record, python benchmarks/run.py list, and python benchmarks/run.py compare <baseline-label>. See benchmarks/README.md; benchmark results under .benchmarks/ are machine-specific and must not be committed.

Do not add machine-specific paths, Conda environment names, credentials, or developer workarounds to tracked files. Put local setup information in the ignored AGENTS.local.md.