Wheel name validation broken

It seems like the new wheel release has broken pypdfium2 installation because wheels with multiple .-separated tags are now erroneously considered invalid:

File "/home/mara/projects/pypdfium2/setupsrc/pl_setup/setup_base.py", line 77, in mkwheel
  setuptools.setup(
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/setuptools/__init__.py", line 87, in setup
  return distutils.core.setup(**attrs)
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/setuptools/_distutils/core.py", line 185, in setup
  return run_commands(dist)
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/setuptools/_distutils/core.py", line 201, in run_commands
  dist.run_commands()
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/setuptools/_distutils/dist.py", line 968, in run_commands
  self.run_command(cmd)
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/setuptools/dist.py", line 1217, in run_command
  super().run_command(command)
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/setuptools/_distutils/dist.py", line 987, in run_command
  cmd_obj.run()
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/wheel/bdist_wheel.py", line 385, in run
  with WheelFile(wheel_path, "w", self.compression) as wf:
File "/tmp/pip-build-env-jzav0p36/overlay/lib/python3.10/site-packages/wheel/wheelfile.py", line 45, in __init__
  raise WheelError(f"Bad wheel filename {basename!r}")
wheel.cli.WheelError: Bad wheel filename 'pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl'
[end of output]

pypdfium2-team/pypdfium2#155

I think something's wrong with your WHEEL_INFO_RE regex.

Presumably introduced with commit 88f02bc?

Blacklist broken wheel release

It might even be more severe possibly? The example below is not even a multi-tag

I pulled an example off of the PEP-491 -- specifically distribution-1.0-1-py27-none-any.whl
took the WHEEL_INFO_RE -- which appears to be re.compile('^(?P<namever>(?P<name>[^-]+?)-(?P<ver>[^-]+?))(-(?P<build>\\d[^-]*))?\n -(?P<pyver>[^-]+?)-(?P<abi>[^-]+?)-(?P<plat>[^.]+?)\\.whl$', re.VERBOSE)
Popped that in regex101
And did not get a match....

This is caused by an unfortunate error in the platform part of the regex. I have a fix almost ready and will publish a patch release very soon.

And did not get a match....

>>> import re
>>> WHEEL_INFO_RE = re.compile(
...    r"""^(?P<namever>(?P<name>[^-]+?)-(?P<ver>[^-]+?))(-(?P<build>\d[^-]*))?
...     -(?P<pyver>[^-]+?)-(?P<abi>[^-]+?)-(?P<plat>[^.]+?)\.whl$""",
...    re.VERBOSE,
... )
>>> WHEEL_INFO_RE.match("distribution-1.0-1-py27-none-any.whl")
<re.Match object; span=(0, 36), match='distribution-1.0-1-py27-none-any.whl'>

Thanks for the quick fix!
Looking at the PyPI history, you may want to consider yanking v0.38.1 as well.

pip also has such a regex:
https://github.com/pypa/pip/blob/90f51db1a32592430f2e4f6fbb9efa7a3a249423/src/pip/_internal/models/wheel.py#L15

I think it differs slightly. Is that on purpose?

The differences are negligible, and I have little interest in fine tuning this regex since wheel name parsing will be delegated to packaging from v1.0.0 forward.

Why is the file name not completely matched?

>>> import re
>>> WHEEL_INFO_RE = re.compile(
...     r"""^(?P<namever>(?P<name>[^\s-]+?)-(?P<ver>[^\s-]+?))(-(?P<build>\d[^\s-]*))?
...      -(?P<pyver>[^\s-]+?)-(?P<abi>[^\s-]+?)-(?P<plat>\S+)\.whl$""",
...     re.VERBOSE,
... )
>>> WHEEL_INFO_RE.match("pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl")
<re.Match object; span=(0, 71), match='pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.ma>

It's just the repr of the match object:

>>> import re
>>> WHEEL_INFO_RE = re.compile(
...     r"""^(?P<namever>(?P<name>[^\s-]+?)-(?P<ver>[^\s-]+?))(-(?P<build>\d[^\s-]*))?
...      -(?P<pyver>[^\s-]+?)-(?P<abi>[^\s-]+?)-(?P<plat>\S+)\.whl$""",
...     re.VERBOSE,
... )
>>> WHEEL_INFO_RE.match("pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl")
<re.Match object; span=(0, 71), match='pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.ma>
>>> m = WHEEL_INFO_RE.match("pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl")
>>> m.group(0)
'pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl'

Presumably introduced with commit 88f02bc?

I think it's introduced with commit 88f02bc.

We can use the following method to test：
wheel unpack "pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"

if wheel with commit 88f02bc （and without 4419390）, the result is:
Bad wheel filename 'pypdfium2-3.6.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl'

if wheel without commit 88f02bc, the result is:
Unpacking to: ./pypdfium2-3.6.0...OK

Fixed parsing of wheel file names with multiple platform tags

Navigation Menu

Repository navigation

Wheel name validation broken #485

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Wheel name validation broken #485

Description

Activity

mara004 commented on Nov 4, 2022

jacobdr commented on Nov 4, 2022

agronholm commented on Nov 5, 2022

mara004 commented on Nov 5, 2022

mara004 commented on Nov 6, 2022

agronholm commented on Nov 6, 2022

zhuofeng6 commented on Feb 21, 2023

pfmoore commented on Feb 21, 2023

xiaoge1001 commented on Dec 18, 2023

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions