Im using pyenv to support having different versions of python.
In a project using python 3.7 I also want linting with pre-commit to run when doing code changes.
But when the lint rules run, pre-commit seem to be looking in a cache folder under the user for python packages and can thus not find them.. even though they are installed on system level. For example:
>git ci -m'lint test'
...
...
/Users/[]/.cache/pre-commit/repo180cws4u/py_env-python3.7/lib/python3.7/site-packages/
...
ModuleNotFoundError: No module named 'six'
>pip list | grep six
six 1.11.0
How can I solve this. (A colleague did by running /Users/[]/.cache/pre-commit/repo180cws4u/bin/python install -m six which solved the issue, but feels wrong.. must be possible to configure to use system packages or similar..?
pre-commit installs isolated environments for each of the tools such that they don't interfere with local development. It sounds like you're missing dependencies in that environment
Running pip install inside the cache environments is very not supported and you are likely to break pre-commit by doing that. They are intentionally private implementation detail so you don't mess with them!
You have two main options:
set up your tools such that they specify their dependencies properly (you haven't shown your configuration, but my guess is you're calling some script?)
for repositories that get installed, the dependencies of your tool should be listed in your setup metadata (usually setup.py / setup.cfg / pyproject.toml)
For example, pre-commit/pre-commit-hooks specifies these dependencies:
# setup.cfg
install_requires =
ruamel.yaml>=0.15
toml
for local repositories with language: python, you should specify your dependencies in additional_dependencies -- for example:
- repo: local
hooks:
- id: run-the-thing
name: run the thing
entry: ./scripts/lint
additional_dependencies: [six, ...]
language: python
types: [python]
with language: system / language: script, pre-commit does not provision an environment. it is the responsibility of each user to set up such an environment. these are the escape hatch from the normal way because they kind of defeat the purpose of the framework in the first place (which is to provision the tools necessary on its own -- without every developer needing to carefully manage a bunch of tools)
For repo: local you'd use language: system and just hope your script produces a useful error message if the user's environment is misconfigured
- repo: local
hooks:
- id: run-the-thing
name: run the thing
entry: ./scripts/lint
language: system
types: [python]
disclaimer: I created pre-commit