Move canonicalize_pypi_name and the PEP 508 name scanner out of crawlers and vendor into one PyPI name module #883

[agent] Filed by the scheduled architecture audit routine (ecosystems and formats). Register: discussion #560 register.

Kind: refactor (mechanical move). Source: review §2.1 and Part 6.4 ("layering inversion"), register E39 (PyPI-name half).

Problem

On 9c43dfc, PEP 503 name canonicalization lives in a crawler: crawlers/python_crawler.rs#L43-L65 defines pub(crate) fn canonicalize_pypi_name. 29 files outside crawlers/ import it from there:

• 12 in vendor/, including lock_inventory ×3;
• 7 in patch/redirect/ and redirect/upstream/;
• 6 in utils/, including utils::purl;
• 3 in vex/discover/;
• 1 in vendor/test_support.

Every one of those is an edge into crawlers, which should sit at the bottom of the dependency order beside them, not under them. utils::purl → crawlers is the clearest case.

The leading PEP 508 project-name scan ("take [A-Za-z0-9._-]* from the start") is written four times, and two rules have drifted on whether the name must start alphanumeric:

vex/discover has no way to reach vendor::common::pep508_name without another upward edge, which is why it carries its own copy.

Symptoms and impact

I know of no user-visible bug. The cost is layering: crawlers can't be moved behind a locator interface (E36 / #855) while 29 modules reach into it for a string function. A future change to name rules, such as the PEP 685 extras normalization, also has four places to land.

Proposed change

1. Add core/src/formats/pypi_name.rs (or utils/pypi_name.rs; formats/ has no PyPI model yet) holding canonicalize_pypi_name and pep508_name. Move both bodies unchanged.
2. Rewrite the 29 use crate::crawlers::python_crawler::canonicalize_pypi_name; lines and the vendor::common::pep508_name callers to the new path. Keep no re-export in crawlers.
3. Replace the three inline scans (value_identity, pypi_requirements, pep508_direct_reference) with pep508_name. Keep the "must start alphanumeric" check as an explicit caller-side guard where it exists today, so this PR changes no behavior.

Size and scope

• About 35 changed import lines, about 30 moved lines and about 20 deleted inline lines. Production diff is under 120 lines, with no behavior change.
• Out of scope:
  • moving Ecosystem out of crawlers/types.rs (the other half of E39; it overlaps --ecosystems rejects NPM and npm, pypi, which socket.yml patches.ecosystems accepts: the ecosystem name parser is written three times #773 and Tracking: build and classify purls through one validated utils::purl API #748);
  • unifying the alphanumeric-start rule, which is a behavior change for a follow-up.

Acceptance criteria

• grep -rn "crawlers::python_crawler::canonicalize_pypi_name" crates returns nothing outside crawlers/.
• One definition each of canonicalize_pypi_name and pep508_name in core, and no inline [-_.] name scan left in the three files above.
• Unit tests for both helpers move with them, including runs of separators, uppercase, leading whitespace and an empty name.
• cargo test -p socket-patch-core --lib and cargo clippy --workspace --all-features -- -D warnings stay green.

Dependencies

None. It unblocks the Ecosystem move (E39) and the locator split (#855, E36).

Backlog review — 2026-10-08

Closed as not planned following backlog review.

The report explicitly identifies no user-visible bug and proposes a behavior-preserving module/import move. Defer this standalone refactor; it is not P1.

Priority: P1 → P3. Cosmetic, maintenance-only, or subsumed scope; retain at P3 if not closed.