Add module-vendor mode to bashkit-coreutils-port

[chaliy]

Background

bashkit-coreutils-port currently has one mode: rewrite a uu_app() clap definition into a generated <util>_args.rs (#1529). For reuse beyond argument surfaces — e.g. swapping printf.rs's handwritten format-spec parser for uucore's — we need a second mode that vendors entire library modules from uucore.

Verification on a probe branch ruled out depending on uucore at runtime:

Metric	Baseline	uucore runtime dep	Delta
bashkit-cli size	24,260,000 B	24,269,344 B	+9 KB (declared, unused)
Cold release build time	1m 37s	3m 15s	+98 s
wasm32-unknown-unknown build	✅ pass	❌ fail (uucore → rustix → errno)	hard blocker

The format module inside uucore is platform-clean — its imports are std, bigdecimal, num-traits, unit-prefix, os_display, plus a small number of uucore-internal types (extendedbigdecimal, UError, NonUtf8OsStrError). The drag comes from other parts of uucore that get pulled in via the parent crate, not from the modules themselves.

Conclusion: vendor selected modules at port time using the same machinery (and drift CI) that the args path uses today. This issue covers the tool capability only; actual vendoring of uucore::format and the printf migration is the first user of this capability and lives in #1534.

Acceptance criteria

Tool surface

• bashkit-coreutils-port gains a port-module invocation (subcommand or distinct CLI verb) accepting:
  • uutils source dir
  • module path under src/uucore/src/lib/features/<name>/
  • output dir (under crates/bashkit/src/builtins/generated/<name>/)
  • a small config (TOML or inline flags) declaring per-import substitutions: (internal_path, action) where action is one of inline (copy the type's defining file in too), replace_with(local_path) (rewrite the import to a bashkit-side equivalent), or error (abort the port — surface unexpected internal references rather than silently emit).
• The tool's existing args mode is unchanged.
• Output files carry the same banner shape as args output: // GENERATED by bashkit-coreutils-port. DO NOT EDIT. + uutils rev + regenerate command + MIT attribution.
• Verbatim copy by default; a future enhancement may add a syn-based rewriter for modules that need it. format/ does not — recommend landing this issue with verbatim-copy only and adding rewrite when the first module demands it.
• Fluent / i18n imports trigger a hard error: unresolved import: 'fluent::*' or similar, telling the operator the module is not safely vendorable without code changes.
• Module-vendoring metadata lives in a single manifest (e.g. crates/bashkit-coreutils-port/vendored.toml) listing each vendored module + its source path + uutils revision. Drift workflow reads from this manifest; adding a new vendored module is one TOML stanza.

CI

• .github/workflows/coreutils-args-drift.yml extends to also iterate the manifest and re-run port-module for each vendored module against uutils HEAD. The same auto-PR fires when args or module output diffs. Single PR per drift run, not one PR per module.

Documentation

• specs/coreutils-args-port.md gains a "Module mode" section:
  • When to use it (vendoring small platform-clean modules) vs args mode (rewriting uu_app()).
  • The substitution model for uucore-internal types.
  • The manifest format and drift-CI integration.
• crates/bashkit-coreutils-port/src/main.rs module-level doc updated to reflect the second mode.
• specs/maintenance.md § Coreutils Argument-Surface Drift broadens to cover module-mode review, not just args.

Out of scope

• Actually vendoring uucore::format and migrating printf.rs. That is the first consumer and lives in Vendor uucore::format and migrate printf.rs to use it #1534.
• Vendoring ranges, parser-num, version-cmp, quoting-style, etc. — file separate consumer issues per module on demand.
• A syn-based module rewriter (only added when a module needs it).

Open questions

• Manifest location. crates/bashkit-coreutils-port/vendored.toml (next to the tool) vs crates/bashkit/src/builtins/generated/vendored.toml (next to the output). Recommend the former — the tool owns it, the workflow only reads it.
• License attribution. Extend the existing THIRD_PARTY_LICENSES entry that covers args codegen with a module list, vs new entry. Recommend extending — same upstream, same MIT terms.
• Internal-type substitution UX. Verbose TOML vs declarative comments inside the source like // codegen: substitute uucore::error::UError -> crate::error::Error. Recommend the TOML for centralization; the substitution rules are policy, not source.

Related

• feat(builtins): port uutils argument surfaces via codegen (POC: cat, tac) #1529 — codegen POC (args mode)
• Differential testing harness for ported coreutils builtins (vs. uutils binaries) #1530 — differential testing harness
• Vendor uucore::format and migrate printf.rs to use it #1534 — first consumer (vendor uucore::format for printf) — depends on this issue
• specs/coreutils-args-port.md
• crates/bashkit-coreutils-port/