Fix dists linux #737
Fix dists linux #737
Conversation
There was a problem hiding this comment.
Looks good. I have tested on MacOS and Linux using the conda env YAMLs to set up test envs.
- Confirmed
hatchis the only prerequisite to runhatch buildin a clean env. - Wheels can be successfully pip installed on the arch they were created for.
- Tarball can be successfully installed on Mac and Linux.
- Can do editable install from repo with
pip install -e '.[tests]' - Built Cython modules are no longer included in the tarball.
All tests pass if using the conda envs (inc bedtools and ImageMagick) on Mac, if on Linux you also need to install ImageMagick manually (have added a note to the readme).
I think that the abi tag in the custom wheel name is incorrect. Please check/fix before merge.
build.py
Outdated
| platform_tag = sysconfig.get_platform().replace("-", "_").replace(".", "_") | ||
| python_version = sysconfig.get_python_version().replace(".", "") # e.g., "310" | ||
| python_impl = "cp" # Assuming CPython. Modify if using PyPy or others. | ||
| abi_tag = f"{python_impl}{python_version}" |
There was a problem hiding this comment.
I think the duplicated implementation+version tag "cp312" is incorrect here.
Second instance should be Application Binary Interface (ABI) tag.
Maybe abi3? I don't know how to automate ABI tag selection.
Example generated wheel names:
jcvi-1.4.25.dev6+gbfef2f34-cp312-cp312-linux_x86_64.whl
jcvi-1.4.25.dev6+gbfef2f34-cp312-cp312-macosx_10_9_x86_64.whl
There was a problem hiding this comment.
thanks. Calling a method from wheel package instead (but this adds a dependency at build_system.requires).
|
When installing from PyPi I think pip will first check for compatible wheels and if not found will download the tarball and install/build locally using the instructions in pyproject.toml. In that case pip will fetch the
|
Changes
.sofrom source tarball.whlwith platform specific tagTesting
pip installon both source tarball and wheelOne caveat, I may have to install
hatchling,hatch-vcs, orhatchdespite them being in[build-system].I am not sure if the standalone tarball and wheel, once hosted on PyPI, will be able to trigger installing the dependencies first on the client machine.