Advanced
The following how-to guides are available in this category:
The following how-to guides are available in this category:
Building noarch packages on other platforms
The following guides cover building conda-forge packages that depend on the CUDA Toolkit. They assume familiarity with writing recipes for compiled code (see Compilers and Runtimes) and with the user guide for running CUDA code.
Finding numpy.get_include() in setup.py, numpy in build-system.requires of pyproject.toml, dependency('numpy') in meson.build, or cimport statements in .pyx or .pyd files are a telltale sign that the package links against NumPy.
The following guides cover topics specific to building packages for Windows on conda-forge.
Windows recipes rely on CMD/Batch scripts (.bat) by default.
In order to compile native code (C, C++, etc.) on Windows, you will need to
Only Orin (sm_87) / Jetpack 6 and later Tegra devices are supported.
By default, provisioned CI machines do not feature a GPU. See how to configure self-hosted runners if your feedstock needs access to additional resources like GPUs or long-running builds, and how to distinguish CPU and GPU jobs when your recipe builds both.
conda-forge has access to additional CI resources for feedstocks that can't build their packages in the default runners. Check Reference> Runners for a comprehensive list.
By default, feedstocks are created with support for the main platforms (e.g. linux-64, osx-64, win-64). Additional ones may be available in an opt-in fashion. The configuration can be changed manually in conda-forge.yml, or through automated migrations.
The conda-forge workflow assumes that a push to any branch in the feedstock repository will result in a build being uploaded to the conda-forge channel (and that's why PRs must always be opened from a fork!).
Sometimes you face the possibility of needing to package a fork.
Some parts of the conda-forge automation are exposed as "bot commands" that can be invoked from issue titles and comments. You can check the full list in the admin-web-services documentation.
Recipes for packages that depend on the CUDA Toolkit typically are more complicated compared
Some packages maintain both an Autotools build and a CMake build. Some maintainers
Despite having a successful build, sometimes we can't run all the tests we'd like on CI. Some reasons include:
The first thing that you should know is that you can locally test Windows