PyInstaller collect_all fails on Windows: VERSION file collides with arcade.version submodule

[benjamin-kirkbride]

Summary

When using PyInstaller's collect_all("arcade") on Windows, the build fails because arcade ships both a VERSION file and an arcade.version submodule. On case-insensitive filesystems (Windows), these collide during the COLLECT phase.

Error

ERROR: Pyinstaller needs to create a directory at 'dist\diana\_internal\arcade\VERSION', but there already exists a file at that path!

PyInstaller tries to bundle:

1. arcade/VERSION (the data file)
2. arcade/version/ (the arcade.version submodule directory)

On Windows, VERSION and version resolve to the same path, causing the conflict.

Reproduction

On Windows:

pip install arcade pyinstaller

Create a minimal spec or script that uses collect_all("arcade"):

from PyInstaller.utils.hooks import collect_all
d, b, h = collect_all("arcade")
# Use these in an Analysis/COLLECT - will fail at COLLECT stage

Workaround

Skip collect_all for arcade and manually include only the runtime resources:

shared_datas = [
    (".venv/Lib/site-packages/arcade/resources", "arcade/resources"),
]

PyInstaller's automatic import analysis handles arcade's Python modules fine without collect_all.

Suggested fix

Either:

• Rename VERSION to something that doesn't collide (e.g., _version.txt, or inline the version string into arcade/__init__.py)
• Or rename the arcade.version submodule (e.g., arcade._version)

This would make arcade compatible with collect_all on Windows, which is the standard PyInstaller mechanism for bundling packages with data files and hidden imports.

Environment

• arcade: latest (installed via pip)
• PyInstaller: latest
• OS: Windows (GitHub Actions windows-latest)
• Python: 3.14

[Cleptomania]

I think renaming the VERSION file is the best approach here, it can become _VERSION or something similar.

[yxshee]

I’d like to work on this. I’ll follow the maintainer-suggested approach: rename the top-level arcade/VERSION data file, update every reader/package-data reference, and add a platform-independent regression test for case-insensitive package path collisions. I’ll target the development branch and include focused test results in the PR.