You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently Python projects in RAPIDS contain README files that are simply symlinks to the top-level READMEs. This approach was taken during the initial wheels spike because that was the easiest way to generate reasonably suitable READMEs for every package. The build backends that we use (setuptools and scikit-build-core) support symlinked READMEs, so the actual README is properly pulled into wheels when the packages are built. However, this approach has a couple of drawbacks
The READMEs often contain invalid links when viewed through the Github GUI because the relative paths are all wrong.
The READMEs contain information that isn't strictly relevant to the package at hand because the top-level READMEs encompass all components of the repo, including other Python packages as well as C++ libraries.
To resolve this issues, I would propose that we write separate README files for each Python package. To minimize the risk of these going stale, we should aim to minimize the contents of READMEs and attempt to link out to other resources as much as possible so that there remains a single source of truth for information as much as possible.
Note that for conda packages the README isn't relevant since only installed components of the Python package end up included in the conda package anyway, so
The text was updated successfully, but these errors were encountered:
Currently Python projects in RAPIDS contain README files that are simply symlinks to the top-level READMEs. This approach was taken during the initial wheels spike because that was the easiest way to generate reasonably suitable READMEs for every package. The build backends that we use (setuptools and scikit-build-core) support symlinked READMEs, so the actual README is properly pulled into wheels when the packages are built. However, this approach has a couple of drawbacks
To resolve this issues, I would propose that we write separate README files for each Python package. To minimize the risk of these going stale, we should aim to minimize the contents of READMEs and attempt to link out to other resources as much as possible so that there remains a single source of truth for information as much as possible.
Note that for conda packages the README isn't relevant since only installed components of the Python package end up included in the conda package anyway, so
The text was updated successfully, but these errors were encountered: