Fix PyPI package specs ignoring PEP 503 spellings (#910)

[Mikola Lysenko (mikolalysenko)]

LLM Description written by Claude Code:claude-opus-5-5

Fixes #910

Summary

socket.yml patches.ignorePackages / patches.packages, scan --package and get's policy check now match PyPI packages by their PEP 503 canonical name. A spec written typing_extensions, Typing_Extensions, typing.extensions or pkg:pypi/typing_extensions now names pkg:pypi/typing-extensions, the same way pip treats them.

Root cause

policy::package_spec_matches is the one matcher behind all four of those entry points. It compared names with to_lowercase() only. Every PyPI purl socket-patch builds is canonical (_ / . / - runs become -), so a spec spelled with _ or . never matched:

• ignore list failed open: the exclusion was silently dropped, and the hosted scan still rewrote requirements.txt. That is a policy bypass.
• allowlist and --package failed closed: nothing was selected, with no hint that the spelling was the cause.

Fix

• For pkg:pypi/ purls, the matcher canonicalizes the name of both the target purl and a purl spec (canonical_pypi_purl), and a bare spec compares with canonicalize_pypi_name.
• Other ecosystems are unchanged: left_pad still does not match pkg:npm/left-pad, and cfg_if still does not match pkg:cargo/cfg-if.
• CLI_CONTRACT.md's --package row now documents the PyPI rule.
• No wrapper changes are needed: npm/, pypi/ and gem/ only dispatch to the binary.

Tests (red → green)

#910 table row	Test
ignorePackages typing_extensions / Typing_Extensions / typing.extensions / pkg:pypi/typing_extensions (hosted)	policy_pypi_names::hosted_ignore_packages_matches_pep503_spellings
ignorePackages typing_extensions, --mode agent	policy_pypi_names::agent_ignore_packages_matches_pep503_spelling
packages allowlist typing_extensions	policy_pypi_names::hosted_packages_allowlist_matches_pep503_spelling
scan --package typing_extensions	policy_pypi_names::hosted_scan_package_flag_matches_pep503_spelling
canonical spelling (control, already passing)	policy_pypi_names::hosted_ignore_packages_canonical_spelling_excludes
unit-level matcher (bare and purl specs, versioned, non-PyPI unchanged)	policy::tests::pypi_package_specs_compare_pep503_canonical_names, policy::tests::pypi_ignore_and_allow_lists_use_pep503_names

Without the fix: the 4 new CLI cases and both new unit tests fail; the control passes.

With the fix:

• cargo test -p socket-patch-cli --all-features --test policy_pypi_names passed, 5/5.
• cargo test -p socket-patch-core --all-features --lib policy:: passed, 69/69.
• e2e_socket_yml_policy passed, 24/24, and spawn_env_hygiene passed, 9/9. The new test spawns through hermetic::binary_command.
• cargo clippy --workspace --all-features -- -D warnings is clean.
• CI: all 412 check runs on 34051e4 passed or were skipped. Bugbot reviewed 34051e4 and found no issues.

Not run locally:

• cargo test --workspace --all-features ran out of disk in this sandbox while compiling every test binary, so CI is the full-suite check (green, see above).
• socket-patch-core --lib has 4 local failures: copy_tree symlinked root, vlt_heal unremovable lock, and two wire_*_failure tests. They depend on chmod 0o555 denials, which root bypasses, and the sandbox runs as uid 0. They pass on CI's non-root runners.

Inherited main failure

Main CI is red: the utils::digest::tests::production_digests_go_through_the_helpers guard fails on the Gradle files, in test, test-release and coverage. Commit 34051e4 ports #878's three-file fix so this PR can go green. It becomes a no-op once #878 lands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Gf2MHGGcCkPZz7pit6onaT

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 34051e4. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Ready for review at head 34051e4.

• CI: all 8 workflows on 34051e4 passed (CI, Gradle, Composer, npm, pnpm, vlt, Benchmarks, Audit GHA). Base main is still 9c43dfc and the branch merges cleanly.
• Bugbot: reviewed 34051e4 and found no issues; there are no open review threads.
• For the reviewer:
  • The main change is policy::package_spec_matches in crates/socket-patch-core/src/policy/mod.rs. PyPI names are now compared by their PEP 503 canonical form, and other ecosystems are unchanged.
  • The Gradle digest commit (34051e4) is a port of Route Gradle digests through utils::digest #878 and becomes a no-op once Route Gradle digests through utils::digest #878 merges.

---

Generated by Claude Code