代码拉取完成，页面将自动刷新
This package is deprecated.
Until NumPy 1.25.0 this package was needed to be able to target the oldest supported NumPy version for packages using the NumPy C API. That targeting was necessary in order to obtain wheel for distribution on PyPI. However, a new "C API exporting" feature in NumPy 1.25.0 made that unnecessary. And with NumPy 2.0, it is now actually necessary to use >=2.0 at build time for packages that want to support both NumPy 1.xx and 2.0.
All users should migrate to expressing a build dependency on numpy
directly, according to the guidance given in
Adding a dependency on NumPy.
[build-system]
requires = ["numpy>=2.0,<3"]
Python 3.13 is supported as of NumPy 2.1. For older Python versions this package can still be used to compile against NumPy 1.xx.
[build-system]
requires = [
"oldest-supported-numpy; python_version<='3.8'",
"numpy>=1.25,<2; python_version>'3.8'",
]
This is a meta-package which can be used in pyproject.toml files
to automatically provide as a build-time dependency the oldest version
of NumPy that supports the given Python version and platform. In case
of platforms for which NumPy has prebuilt wheels, the provided version
also has a prebuilt NumPy wheel.
The reason to use the oldest available NumPy version as a build-time dependency is because of ABI compatibility. Binaries compiled with old NumPy versions are binary compatible with newer NumPy versions, but not vice versa. This meta-package exists to make dealing with this more convenient, without having to duplicate the same list manually in all packages requiring it.
In other words:
[build-system]
requires = [
"wheel",
"setuptools",
"numpy==1.13.3; python_version=='3.5'",
"numpy==1.13.3; python_version=='3.6'",
"numpy==1.14.5; python_version=='3.7'",
# more numpy requirements...
]
can be replaced by:
[build-system]
requires = ["wheel", "setuptools", "oldest-supported-numpy"]
And as new Python versions are released, the pyproject.toml file does not
need to be updated.
install_requires in this repository?The NumPy version pinnings are defined inside the setup.cfg file as
install_requires dependencies, rather than as build-time dependencies
inside pyproject.toml. This is deliberate, since NumPy is not actually
required to build wheels of oldest-supported-numpy. What we need here
is to make sure that when oldest-supported-numpy is installed into
the build environment of a package using it, NumPy gets installed too
as a runtime dependency inside the build environment.
Another way to think about this is that since we only publish (universal)
wheels of oldest-supported-numpy, the wheel contains no pyproject.toml,
setup.cfg, or setup.py code - it only contains metadata including
dependencies which get installed by pip when oldest-supported-numpy is
installed.
In many cases, even though your package may require a version of
NumPy that is more recent than the pinned versions here, this
is often a runtime requirement, i.e. for running (rather than
building) your package. In many cases, unless you use recent
features of the NumPy C API, you will still be able to build your
package with an older version of NumPy and therefore you will still
be able to use oldest-supported-numpy. You can still impose a
more recent NumPy requirement in install_requires
If oldest-supported-numpy pins a numpy version that is broken for
everyone using a certain OS/platform/interpreter combination, we can update the
== pin to a newer release. In general, building against a newer bugfix
release (i.e., a higher Y value for a 1.X.Y version number) is safe to
do. Newer minor versions will likely not be ABI-compatible, so are much more
difficult to change. If a bug only affects some uses cases (e.g., versions <
1.20.3 don't work on Windows when using f2py), the pin cannot be updated
because it will affect backwards compatibility of oldest-supported-numpy.
In that case, it is recommended that you add the needed constraint directly
in your own pyproject.toml file. For example:
[build-system]
requires = [
"wheel",
"numpy==1.19.0; python_version<='3.8' and platform_system=='Windows' and platform_python_implementation != 'PyPy'",
"oldest-supported-numpy; python_version>'3.8' or platform_system!='Windows' or platform_python_implementation == 'PyPy'",
# more requirements (if needed) ...
]
Note that when you do this, it is important to ensure the conditions are such
that there is exactly one pin possible for a given platform configuration.
Otherwise your build will fail or pip may refuse to install your package
only on that configuration (so you likely won't see it in CI).
The oldest-supported-numpy repository contains tests, so for safety you
may want to implement your constraints in its setup.cfg and run the
tests with pytest to validate those constraints.
oldest-supported-numpy available for Conda, Homebrew, Debian, etc.?The pyproject.toml format is specific to PyPI. Other packaging systems have
their own metadata formats and ways of specifying dependencies. Typically they
don't need anything like oldest-supported-numpy because either (a) they ship
only a single NumPy version for a given release (typically the case for Linux
distros and Homebrew), or (b) they have a more explicit way of managing ABI
compatibility (see for example conda-forge's pin_compatible feature:
https://conda-forge.org/docs/maintainer/knowledge_base.html#linking-numpy).
The current name is not very catchy as package names go, but it
is very descriptive. This package is only meant to be used in
pyproject.toml files for defining build-time dependencies,
so it's more important to have a descriptive than a catchy name!
Please feel free to open an issue or a pull request if you think something is wrong or could be improved!
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手