Regression: sdists no longer install package data correctly with 35.0.0 #1016
Comments
Here's a reproducing test script using cartridge-downloads, which uses
(Note that this uses Diff output: --- good-info 2017-04-18 16:25:18.364959212 +0200
+++ bad-info 2017-04-18 16:25:32.172929357 +0200
@@ -37,12 +37,6 @@
cartridge_downloads/models.py
cartridge_downloads/page_processors.py
cartridge_downloads/signals.py
- cartridge_downloads/templates/email/form_response.html
- cartridge_downloads/templates/email/form_response.txt
- cartridge_downloads/templates/email/order_receipt.html
- cartridge_downloads/templates/email/order_receipt.txt
- cartridge_downloads/templates/shop/complete.html
- cartridge_downloads/templates/shop/downloads/index.html
cartridge_downloads/templatetags/__init__.py
cartridge_downloads/templatetags/__pycache__/__init__.cpython-35.pyc
cartridge_downloads/templatetags/__pycache__/downloads.cpython-35.pyc As you can see, setuptools 35.0.0 no longer installs the |
pyuri is another example of an affected package: + diff -u good-info bad-info
--- good-info 2017-04-18 16:36:47.098278222 +0200
+++ bad-info 2017-04-18 16:36:53.994243434 +0200
@@ -19,5 +19,4 @@
pyuri/__pycache__/uri.cpython-35.pyc
pyuri/__pycache__/validators.cpython-35.pyc
pyuri/uri.py
- pyuri/uri.regex
pyuri/validators.py |
Well, now we know. The quick fix is to revert the change. The better fix would be to also write a test to capture this expectation. |
This test is also a regression test for issue pypa#1016.
Okay, I created an integration test case using
This test case passes on 34, and fails on 35. If it looks good, I can PR that as a starting point for fixing this issue? |
This test is also a regression test for issue pypa#1016.
I think that at this point it is better to revert #1014 starting from @pjdelport test case, reopening #436 . Sorry for the regression: when testing #1014 I've completely overlooked the implications beyond sdist creation. |
I went ahead and pulled in the commit. Thanks for putting it together and presenting it in such an accessible way. |
This reverts commit edb6f63. This change was made in reaction to tests failing due to pypa/setuptools#1016. Better to wait for that issue to be resolved than to create new duplication in declaration of package files to later become stale.
Thanks @jaraco! |
Problem description
I have a package that uses
setuptools_scm
andinclude_package_data=True
to include some Django templates and static files.With setuptools 34.4.1, installing an sdist will install these package data files without any problems. However, with setuptools 35.0.0, installing the same sdist file will no longer install these package data files.
Note:
SOURCES.txt
.Cause?
As far as I can tell, this regression is due to #436 / #1014, which no longer reads the sdist package's existing manifest. When the manifest gets regenerated (?), the files that were previously included by
setuptools_scm
while building the sdist are no longer included anymore.The text was updated successfully, but these errors were encountered: