Skip internal non-Python references

[lkollar]

This should fix #531. While investigating, also discovered a subtle bug when auditwheel can use the wrong manylinux tag when repairing the wheel.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.29%. Comparing base (f296d5a) to head (72c0563).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #683      +/-   ##
==========================================
+ Coverage   95.24%   95.29%   +0.04%     
==========================================
  Files          22       22              
  Lines        1830     1849      +19     
  Branches      343      347       +4     
==========================================
+ Hits         1743     1762      +19     
  Misses         48       48              
  Partials       39       39              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR aims to fix auditwheel repair behavior for wheels that contain internal shared libraries that lddtree fails to resolve (e.g., manylinux_2_28 case in #531), and also fixes a case where repair could graft/retag using the wrong ABI policy.

Changes:

• Add repair_external_refs alongside full_external_refs and introduce filtering logic intended to drop unresolved internal sonames that already exist in the wheel.
• Update repair_wheel to graft/retag using an explicit target_abi (requested policy) rather than relying on abis[0].
• Add unit tests covering the new filtering behavior and the requested-ABI grafting behavior.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/auditwheel/wheel_abi.py	Adds repair_external_refs and filtering of unresolved internal sonames for repair.
src/auditwheel/repair.py	Uses repair_external_refs and a new target_abi parameter to choose which policy drives grafting/retagging.
src/auditwheel/main_repair.py	Passes the requested policy name as target_abi into repair_wheel.
tests/unit/test_wheel_abi.py	Adds tests around filtering internal unresolved references for repair.
tests/unit/test_repair.py	Adds a test ensuring grafting uses the requested ABI policy.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

repair_external_refs for Python extensions is currently set to full_external_refs without applying _filter_repair_refs. That means unresolved DT_NEEDED entries that are actually internal (the library exists somewhere in the wheel, but lddtree couldn't resolve it so the path is None) will still be present for the extension and will still trigger ValueError: required library ... could not be located during repair_wheel. Consider building repair_external_refs for all ELFs (including Python extensions) by filtering against wheel_sonames after you’ve collected the wheel’s sonames, and only keeping refs that still have libs/blacklist after filtering.

Suggested change

	repair_external_refs[fn] = full_external_refs[fn]
	# Apply the same external-reference filtering used for other ELFs
	# so that unresolved DT_NEEDED entries that are actually provided
	# by libraries inside the wheel (wheel_sonames) do not cause
	# spurious repair failures.
	filtered_refs = _filter_repair_refs(
	{fn: full_external_refs[fn]},
	wheel_sonames,
	)
	if fn in filtered_refs:
	repair_external_refs[fn] = filtered_refs[fn]

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

wheel_sonames is built from fn.name (the filenames of ELF entries in the wheel), but _filter_repair_refs and its docstring describe this as a set of SONAMEs. To avoid confusion for future maintainers, consider renaming the parameter/variable (e.g., to reflect DT_NEEDED/filename matching) and updating the docstring accordingly.

Suggested change

	wheel_sonames: frozenset[str],
	) -> dict[str, ExternalReference]:
	"""Return external refs with unresolved sonames already present in the
	wheel removed.
	``external_refs`` maps policy name to the libraries needed by one ELF.
	``wheel_sonames`` is the set of shared-library sonames already present in
	the wheel. The returned mapping keeps only refs that still need repair.
	"""
	filtered_refs = {}
	for name, external_ref in external_refs.items():
	libs = {
	lib: path
	for lib, path in external_ref.libs.items()
	if path is not None or lib not in wheel_sonames
	wheel_lib_names: frozenset[str],
	) -> dict[str, ExternalReference]:
	"""Return external refs with unresolved libraries already present in the
	wheel removed.
	``external_refs`` maps policy name to the libraries needed by one ELF.
	``wheel_lib_names`` is the set of shared-library names/filenames already
	present in the wheel (used for DT_NEEDED/filename-style matching). The
	returned mapping keeps only refs that still need repair.
	"""
	filtered_refs = {}
	for name, external_ref in external_refs.items():
	libs = {
	lib: path
	for lib, path in external_ref.libs.items()
	if path is not None or lib not in wheel_lib_names

[mayeut]

While investigating, also discovered a subtle bug when auditwheel can use the wrong manylinux tag when repairing the wheel.

I fail to understand the issue, I guess this is related to the target_abi addition ?
abis are always sorted - this should likely be documented and/or enforced - so we do want to repair for abis[0] if we're going to add multiple tags as manylinux_x_y1 is always compatible with manylinux_x_y2 when y1<y2 but the reverse is not true.
Can this be split in a different PR ?

I'm trying to understand what the new filtering is doing exactly. Do you have an example before/after to ease this (or an integration test) ?

[lkollar]

You're right, that change is unrelated. I'll move it to a new PR and try to come up with test to reproduce the issue.