Test files are in binary package distribution

[daskol]

Test files are not filtered properly. The issue is that setuptools.find_package finds packages not modules while tests are organized as a separate modules. In order to mitigate the issue, one should filter test files manually as follows. This patch are created and tested on chex v0.1.86.

--- a/setup.py	2024-03-19 12:58:11.000000000 +0300
+++ b/setup.py	2024-04-17 15:19:33.593293889 +0300
@@ -15,8 +15,11 @@
 """Install script for setuptools."""
 
 import os
-from setuptools import find_packages
-from setuptools import setup
+from pathlib import Path
+
+from setuptools import find_packages, setup
+from setuptools.command.build_py import build_py as _build_py
+
 
 _CURRENT_DIR = os.path.dirname(os.path.abspath(__file__))
 
@@ -40,6 +43,15 @@
     ]
 
 
+class build_py(_build_py):
+
+    def find_package_modules(self, package, package_dir):
+        modules = super().find_package_modules(package, package_dir)
+        return [(pkg, mod, file)
+                for pkg, mod, file in modules
+                if not Path(file).match('**/*_test.py')]
+
+
 setup(
     name='chex',
     version=_get_version(),
@@ -51,7 +63,7 @@
     long_description_content_type='text/markdown',
     author_email='chex-dev@google.com',
     keywords='jax testing debugging python machine learning',
-    packages=find_packages(exclude=['*_test.py']),
+    packages=find_packages(),
     install_requires=_parse_requirements(
         os.path.join(_CURRENT_DIR, 'requirements', 'requirements.txt')),
     tests_require=_parse_requirements(
@@ -73,4 +85,5 @@
         'Topic :: Software Development :: Testing :: Unit',
         'Topic :: Software Development :: Libraries :: Python Modules',
     ],
+    cmdclass={'build_py': build_py},
 )

[shreed27]

Hi @daskol 👋

I’ve gone through the issue description and the proposed approach around filtering test files from the binary/package distribution. I understand the root cause is setuptools.find_packages including test modules, which then end up in the built artifacts.

I’d like to take this up and work on a proper fix. My plan would be to:
Reproduce the issue locally using the current release
Verify the existing patch/approach (or adjust it if needed)
Ensure test files are excluded cleanly from the final distribution
Add/verify tests to prevent regressions

If this sounds good, I’m happy to submit a PR addressing this. Please let me know if I can proceed or if you’d like the fix structured in a specific way.

Thanks!