@mayeut , @henryiii - does this look like the right sort of approach for splitting out utils? I've detailed my thought patterns in the description and would welcome your insights.

IMHO, we don't need the first item. cibuildwheel is not a library so this module is internal.
I would move download to files.py

I think the next step is correcting the imports so we can avoid import *.

Hey @MusicalNinjaDad ! Thanks for putting this together. However, I must say I'm not totally convinced by the need for this change. Admittedly, the utils module is large, but utils are generally self-contained chunks of code that have very few interdependencies - i.e. they're simple. But happy to be outvoted, perhaps I'm missing some extra context here, please let me know if so - I've been out of the project for a few weeks!

(I think perhaps this started from a problem with circular imports. Regarding that, utils should never import another module in the codebase, with the possible exception of a file of basic types or constants or something.)

Having said that, your points around docstrings are valid, and a bit of effort here could improve things for sure.

IMO, having a bit more separation here would improve readability and one massive "utils" file for everything is not ideal. (See various articles on this, like https://breadcrumbscollector.tech/stop-naming-your-python-modules-utils/).

I think it will help when this PR is closer to a final form, without the import *s.

This could also be done as utils.py and fileutils.py, rather than nested. I like nested, but utils/misc.py seems even worse than utils.py! :D

This could also be done as utils.py and fileutils.py, rather than nested. I like nested, but utils/misc.py seems even worse than utils.py! :D

The goal would be that utils/misc.py is only temporary, and a few more PRs over the (northern) summer would break it down into more meaningful units.

@joerick , I agree this is very much a style question and not a must-do.

As a new contributor I personally found that utils.py came across as a jumble of functions and it was hard jumping around in the file to find the "flow" I was trying to understand. Not a big problem, but "could be better". My personal view on style is: if you need to implement an internal structure and grouping within a module; it's worthwhile thinking about why not to split it into more logical units.

The original need was driven by a circular import with logging.py which needs to use CI-identification functionality from utils. Currently there is no logging used in utils, but I would guess that if / as additional clean-ups are done around consolidating output, error handling, etc. then this kind of circular import risk will grow.

For that reason I'm happy to invest time on this change, if it is considered worthwhile, or, at least, harmless and possibly useful.

I think that the change is either going to be worth it done at once, or not at all. Here's a go at how we might organise things- how does this feel to people?

util/cmd.py:
    call
    shell

util/file.py:
    download
    extract_zip
    extract_tar
    move_file
    chdir
    resources_dir
    install_certifi_script
    free_thread_enable_313
    test_fail_cwd_file
    DEFAULT_CIBW_CACHE_PATH
    CIBW_CACHE_PATH

util/selectors.py:
    BuildSelector
    TestSelector
    selector_matches

util/ci.py:
    detect_ci_provider
    CIProvider
    fix_ansi_codes_for_github_actions

util/packaging.py:
    virtualenv
    _parse_constraints_for_virtualenv
    _ensure_virtualenv
    find_compatible_wheel
    BuildFrontendName
    BuildFrontendConfig
    DependencyConstraints
    get_pip_version
    combine_constraints
    find_uv
    get_build_verbosity_extra_flags
    split_config_settings

util/helpers.py:
    format_safe
    prepare_command
    strtobool
    unwrap
    parse_key_value_string
    read_python_configs

move to __main__:
    FileReport
    print_new_wheels

move to pyodide.py:
    ensure_node

move to options.py:
    MANYLINUX_ARCHS
    MUSLLINUX_ARCHS

inline where needed:
    IS_WIN

I think that could work. I don't think there are any circular dependencies there, but that is a concern with breaking the file up.