hatch-cython

hatch-cython

CI/CD
Package
Meta

---

Table of Contents

• Usage
• Configuration Options
• Notes
  • Templating (Tempita)
• License

Usage

The build hook name is cython.

• pyproject.toml

# [tool.hatch.build.hooks.cython]
# or
# (hatch.toml)
# [build.targets.wheel.hooks.cython]
# or
[tool.hatch.build.targets.wheel.hooks.cython]
dependencies = ["hatch-cython"]

# [tool.hatch.build.hooks.cython.options]
# or
# (hatch.toml)
# [build.targets.wheel.hooks.cython.options]
# or
[tool.hatch.build.targets.wheel.hooks.cython.options]
# include .h or .cpp directories
includes = []
# include numpy headers
include_numpy = false
include_pyarrow = false

# include_{custom}
include_somelib = {
    # must be included in build-dependencies
    pkg = "somelib",
    # somelib.gets_include() -> str
    include = "gets_include",
    # somelib.gets_libraries() -> list[str]
    libraries = "gets_libraries",
    # somelib.gets_library_dirs() -> list[str]
    library_dirs = "gets_library_dirs",
    # somelib.some_setup_op() before build
    required_call = "some_setup_op"
}

compile_args = [
  # single string
  "-v",
  # by platform
  { platforms = ["linux", "darwin"], arg = "-Wcpp" },
  # by platform & arch
  { platforms = "darwin", arch = "x86_64", arg = "-arch x86_64" },
  { platforms = ["darwin"], arch = "arm64", arg = "-arch arm64" },
  # with pep508 markers
  { platforms = ["darwin"], arch = "x86_64", arg = "-I/usr/local/opt/llvm/include", depends_path = true, marker = "python_version <= '3.10'"  },
]

directives = { boundscheck = false, nonecheck = false, language_level = 3, binding = true }

compile_kwargs = { }

• hatch.toml

# [build.hooks.cython]
# or
[build.targets.wheel.hooks.cython]
dependencies = ["hatch-cython"]

# [build.hooks.cython.options]
# or
[build.targets.wheel.hooks.cython.options]
directives = { boundscheck = false, nonecheck = false, language_level = 3, binding = true }
compile_args = [
    "-O3",
]
includes = []
include_numpy = false
# equivalent to include_pyarrow = true
include_somelib = { pkg = "pyarrow", include="get_include", libraries="get_libraries", library_dirs="get_library_dirs", required_call="create_library_symlinks" }
define_macros = [
    # ["ABC"] -> ["ABC", "FOO"] | ["ABC", "DEF"]
    ["NPY_NO_DEPRECATED_API", "NPY_1_7_API_VERSION"],
]

Configuration Options

Field	Type
src	str | None   directory within src dir or .  which aliases the package being built. e.g. package_a -> src/package_a_lib   src = "package_a"
directives	directives to cython (see compiler-directives)
compile_args	str or { platforms = ["*"] | "*", arg = str }. see extensions for what args may be relevant
extra_link_args	str or { platforms = ["*"] | "*", arg = str }. see extensions for what args may be relevant
env	{ env = "VAR1", arg = "VALUE", platforms = ["*"], arch = ["*"] }  if flag is one of:  - ARFLAGS  - LDSHARED   - LDFLAGS  - CPPFLAGS   - CFLAGS   - CCSHARED the current env vars will be merged with the value (provided platform & arch applies), separated by a space. This can be enabled by adding { env = "MYVAR" ... , merges = true } to the definition.
includes	list str
include_{package}	{ pkg = str, include = str, libraries = str| None, library_dirs = str | None , required_call = str | None }  where all fields, but pkg, are attributes of pkg in the type of callable() -> list[str] | str | list[str] | str. pkg is a module, or loadable module object, which may be imported through import x.y.z.
include_numpy | include_pyarrow | include_pythran	bool 3rd party named imports. must have the respective opt in dependencies
parallel	bool = False  if parallel, add openmp headers important: if using macos, you need the homebrew llvm vs apple's llvm in order to pass -fopenmp to clang compiler
compiler	compiler used at build-time. if msvc (Microsoft Visual Studio), /openmp is used as argument to compile instead of -fopenmp  when parallel = true. default = false
compile_py	whether to include .py files when building cython exts. note, this can be enabled & you can do per file / matched file ignores as below. default = true
define_macros	list of list str (of len 1 or 2). len 1 == [KEY] == #define KEY FOO . len 2 == [KEY, VALUE] == #define KEY VALUE. see extensions
** kwargs	keyword = value pair arguments to pass to the extension module when building. see extensions

Files

[build.targets.wheel.hooks.cython.options.files]
exclude = [
    # anything matching no_compile is ignored by cython
    "*/no_compile/*",
    # note - anything "*" is escaped to "([^\s]*)" (non whitespace).
    # if you need an actual * for python regex, use as below:
    # this excludes all pyd or pytempl extensions
    "([^.]\\*).(pyd$|pytempl$)",
    # only windows
    { matches = "*/windows", platforms = ["linux", "darwin", "freebsd"] },
    # only darwin
    { matches = "*/darwin", platforms = ["linux", "freebsd", "windows"] },
    # only linux
    { matches = "*/linux", platforms = ["darwin", "freebsd", "windows"] },
    # only freebsd
    { matches = "*/freebsd", platforms = ["linux", "darwin", "windows"] }
]
aliases = {"abclib._filewithoutsuffix" = "abclib.importalias"}

sdist

Sdist archives may be generated normally. hatch must be defined as the build-system build-backend in your pyproject.toml. As such, hatch will automatically install hatch-cython, and perform the specified e.g. platform-specific adjustments to the compile-time arguments. This allows the full build-process to be respected, and generated following specifications of the developer.Note: If you specify hatch-cython to run outside of simply wheel-step processes, the extension module is skipped. As such, the .c & .cpp, as well as templated files, may be generated and stored in the sdist should you wish. However, there is currently little purpose to this, as the extension will likely have differed compile arguments.

Templating

Cython tempita is supported for any files suffixed with .in, where the extension output is:

• .pyx.in
• .pyd.in
• .pyi.in For these files, you may expect the output .pyx.in -> .pyx. Thus, with aliasing this would look like:

[build.targets.wheel.hooks.cython.options.files]
aliases = {"abclib._somemod" = "abclib.somemod"}

• 1. Source files somemod.pyi.in, _somemod.pyx.in
• 1. Processed templates somemod.pyi, _somemod.pyx
• 1. Compiled module abclib.somemod{.pyi,.pyx}

An example of this is included in:

• pyi stub file
• pyx cython source file
• pyi stub (rendered)
• pyx cython source (rendered)

Template Arguments

You may also supply arguments for per-file matched namespaces. This follows the above platforms, arch, & marker formats, where if supplied & passing the condition the argument is passed to the template as a named series of keyword arguments.

You supply an index value, and all other kwargs to templates are keywords for each index value. Follows FIFO priority for all keys except global, which is evaluated first and overriden if there are other matching index directives. The engine will attempt to merge the items of the keywords, roughly following:

args = {
    "index": [
        {"keyword": "global", ...},
        {"keyword": "thisenv", ...},
    ],
    "global": {"abc": 1, "other": 2},
    "thisenv": {"other": 3},
}

merge(args) -> {"abc": 1, "other": 3}

In hatch.toml:

[build.targets.wheel.hooks.cython.options.templates]
index = [
  {keyword = "global", matches = "*" },
  {keyword = "templated_mac", matches = "templated.*.in",  platforms = ["darwin"] },
  {keyword = "templated_mac_py38", matches = "templated.*.in",  platforms = ["darwin"], marker = "python == '3.8'" },
  {keyword = "templated_win", matches = "templated.*.in",  platforms = ["windows"] },
  {keyword = "templated_win_x86_64", matches = "templated.*.in",  platforms = ["windows"], arch = ["x86_64"] },

]

# these are passed as arguments for templating

# 'global' is a special directive reserved & overriden by all other matched values
global = { supported = ["int"] }

templated_mac = { supported = ["int", "float"] }
templated_mac_py38 = { supported = ["int", "float"] }

templated_win = { supported = ["int", "float", "complex"] }

# assuming numpy is cimported in the template
templated_win_x86_64 = { supported = ["int", "float", "np.double"]}

Development

Requirements

• a c / c++ compiler
• python 3.8-<=3.12
• git-cliff (pip install git-cliff)
• tasks

Scripts

• test: library & coverage
  • hatch run cov
• test: src structure example
  • task example
• test: simple structure example
  • task simple-structure
• commit: precommitt
  • task precommit

License

hatch-cython is distributed under the terms of the MIT license.