setup.cfg

[metadata]
url = https://github.com/NVIDIA/NVFlare
description = Federated Learning Application Runtime Environment
long_description = file:README.md
long_description_content_type = text/markdown; charset=UTF-8
license = Apache-2.0
license_files =
    LICENSE
classifiers =
    Programming Language :: Python :: 3.10
    Programming Language :: Python :: 3.11
    Programming Language :: Python :: 3.12
    Programming Language :: Python :: 3.13
    Programming Language :: Python :: 3.14
    License :: OSI Approved :: Apache Software License
    Operating System :: POSIX :: Linux

[options]
zip_safe = True
python_requires = >= 3.10
install_requires =
    certifi>=2026.10.19
    uv>=0.13.6,<0.14.0
    cryptography>=45.0.0
    grpcio>=1.62.1
    numpy
    protobuf>=4.24.4
    psutil>=5.9.1
    PyYAML>=6.0
    requests>=2.28.0
    msgpack>=1.0.3
    docker>=6.0
    aiohttp
    pydantic>=2.0
    PyJWT>=2.0,<3.0
    safetensors
    packaging

[options.extras_require]
dashboard =
    Flask==3.1.3
    Flask-JWT-Extended==4.6.0
    Flask-SQLAlchemy==3.1.1
    gunicorn>=22.0.0
edge-http =
    Flask==3.1.3
HE =
    # tenseal currently has no Python 3.14 wheel, so HE extra is unavailable on 3.14.
    tenseal==0.3.15; python_version < "3.13"
    tenseal==0.3.16; python_version >= "3.13" and python_version < "3.14"
PSI =
    openmined.psi==2.0.6; python_version < "3.14"
PT =
    torch
    torchvision
SKLEARN =
    scikit-learn
    pandas>=1.5.1
TRACKING =
    mlflow
    wandb
    tensorboard
MONITORING =
    datadog
K8S =
    kubernetes!=36.0.0
app_opt =
    %(HE)s
    %(PSI)s
    %(PT)s
    %(SKLEARN)s
    %(TRACKING)s
    %(MONITORING)s
    %(K8S)s
    packaging
    pytorch_lightning
    xgboost<3.4
    bitsandbytes
app_opt_mac =
    %(PT)s
    %(SKLEARN)s
    %(TRACKING)s
    %(K8S)s
    packaging
doc =
    sphinx>=5
    sphinx_rtd_theme
    recommonmark
    sphinx-llm>=0.4.1
    sphinx-copybutton
    sphinxcontrib-jquery
all =
    %(app_opt)s
    %(dashboard)s
    %(edge-http)s
# Documentation needs runtime imports for autodoc, but no test tooling.
doc_build =
    %(doc)s
    %(dashboard)s
    %(edge-http)s
    %(HE)s
    %(PSI)s
    %(PT)s
    %(SKLEARN)s
    %(TRACKING)s
    %(MONITORING)s
    %(K8S)s
    packaging
    pytorch_lightning
    xgboost-cpu<3.4; sys_platform == "linux" and platform_machine == "x86_64"
    xgboost<3.4; sys_platform != "linux" or platform_machine != "x86_64"
    kagglehub!=1.0.1
    kagglesdk<0.1.31; python_version >= "3.11"
    fastdigest>=0.4.0,<0.6; python_version < "3.14"
all_mac =
    %(app_opt_mac)s
    %(dashboard)s
    %(edge-http)s
test_support =
    isort==5.13.2
    flake8==7.1.1
    black==25.9.0
    click==8.1.7; python_version < "3.14"
    click==8.2.1; python_version >= "3.14"
    # pytest 9.2 changes node IDs and breaks loadgroup in pytest-xdist 3.6.1.
    pytest<9.2
    pytest-xdist==3.6.1
    pytest-cov==5.0.0
    pytest-timeout==2.4.0
    pandas>=1.5.1
    nbformat
    nbmake
    kagglehub!=1.0.1
    # kagglesdk 0.1.31 and 0.1.32 import a missing competitions.legacy package.
    kagglesdk<0.1.31; python_version >= "3.11"

test =
    %(all)s
    %(test_support)s
    fastdigest>=0.4.0,<0.6; python_version < "3.14"

test_mac =
    %(all_mac)s
    %(test_support)s

dev =
    %(doc)s
    %(test)s
dev_mac =
    %(doc)s
    %(test_mac)s
# CPU-only variant of `app_opt`: omits torch (install separately from
# https://download.pytorch.org/whl/cpu first) and bitsandbytes (GPU-only).
# torchvision is included so the extra is self-sufficient for tests; install
# the CPU torch (and a matching torchvision) from the CPU index beforehand to
# avoid pip pulling the GPU torchvision wheel from PyPI as a dependency.
app_opt_cpu =
    %(HE)s
    %(PSI)s
    %(SKLEARN)s
    %(TRACKING)s
    %(MONITORING)s
    %(K8S)s
    packaging
    pytorch_lightning
    torchvision
    xgboost<3.4
all_cpu =
    %(app_opt_cpu)s
    %(dashboard)s
    %(edge-http)s
test_cpu =
    %(all_cpu)s
    %(test_support)s
    fastdigest>=0.4.0,<0.6; python_version < "3.14"
dev_cpu =
    %(doc)s
    %(test_cpu)s

# SkillEvaluator remains CI-only because PyPI rejects direct URL dependencies in
# published package metadata. The premerge workflow installs it in isolation.

[options.entry_points]
console_scripts =
    nvflare=nvflare.cli:main

[flake8]
select = B,C,E,F,N,P,T4,W,B9
max_line_length = 120
# E501 is not flexible enough, we're using B950 instead
# N812 lowercase 'torch.nn.functional' imported as non lowercase 'F'
ignore =
    E203,
    E302,
    E303,
    E305,
    E402,
    E501,
    E701,
    E721,
    E722,
    E741,
    F821,
    F841,
    F999,
    N812,
    W291,
    W503,
    W504
per_file_ignores = __init__.py: F401
exclude = *.pyi,.git,.eggs,nvflare/_version.py,versioneer.py,venv,.venv,_version.py,*grpc.py,*_pb2.py

[pydocstyle]
convention = google
add_ignore = D100,D101,D102,D103,D104,D105,D106
# match-dir=([^\.].*)|(.*(?!protos).*)
match-dir=((?!protos).)*

[versioneer]
VCS = git
style = pep440
versionfile_source = nvflare/_version.py
versionfile_build = nvflare/_version.py
tag_prefix =