GeezerVison

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

Sweep of the next audited, unswept rows of coverage/files.tsv in crates/socket-patch-cli/src/commands/: hosted_bundle.rs, hosted_unwind.rs, list.rs, lock_cli.rs, mod.rs and pypi_reinstall.rs (clean). No other open PR touches these files. This follows the architecture audit's per-file notes (discussion #1830, living document Part 4).

Changes

hosted_unwind.rs (+ rollback.rs)

• Deleted the dead HostedLegOutcome::unsupported field. Nothing ever wrote to it. Removing it also removes rollback's hosted_unsupported event loop and its exit-code conjunct (−11 lines in rollback.rs). The hosted_unsupported code appears nowhere else: not in the contract and not in any test.
• Merged the two flush_error.is_none() blocks into one. Warning order is unchanged.
• A dry run now skips the bun live_user_patch_keys read, since only the wet run uses it.
• The module doc now names scan's withdrawn-patch caller. Removed 3 bare issue references.

hosted_bundle.rs

• Bundle option fields are now moved with take() instead of cloned.
• The stdin schema in the module doc was missing policySha256; it is now listed.

list.rs

• Removed the Store as Source alias and its 6-line comment.
• Replaced the PatchRecord literal with ledgers::uuid_only_record.
• Replaced 3 RunWarning literals with RunWarning::new.
• Removed a stale comment about read_manifest.
• Inlined vendor_state_lenient, which had one caller and lived in mod.rs.
• Moved a misplaced fixture doc onto vendor_entry.
• Deleted ordering_is_deterministic_across_builds. It built the envelope twice from the same HashMap, so both builds iterate in the same order and the test can't fail.

mod.rs

• Moved mod user_patches up with the other module declarations.
• hosted_state_from_pins now uses uuid_only_record.
• Tests fixed (3):
  • an_uncovered_manifest_record_beside_a_ledger_is_ambiguous asserted only .contains("--mode").
  • a_malformed_manifest_is_not_a_hosted_project asserted only is_err().
  • a_hosted_pin_over_a_ledger_entry_is_ambiguous asserted only is_err().
  • All three now compare the whole usage error, including the named purl. The expected strings are written out as literals.
  • hosted_pins_beside_a_ledger_are_ambiguous now does the same.
  • Red→green: I temporarily changed the (true, true) arm to return "<root> --mode". The old asserts still passed. The new tests failed (an_uncovered_… and a_malformed_…), and they pass again after the revert.

lock_cli.rs

• Inlined emit, held_hint and held_message into their only callers.
• The human branch of acquire_or_emit now prints through format_lock_error, so it shares one formatter with the sites that build their own envelope.
• Removed the 24-line lifecycle essay; core apply_lock already documents the lifecycle.
• Removed history notes about the "historical" try-once, --break-lock and the acquire-mkdirs change.
• Fixed the doc overclaim on error_envelope_propagates_dry_run_and_command.
• Tests:
  • The 3 held_message tests became one, lock_failure_renders_the_waited_budget.
  • Deleted held_hint_depends_on_whether_we_already_waited, a duplicate of format_lock_error_matches_acquire_or_emit.
  • waiting_message is still a separate function, because its status-line text is tested only through it.

Not done here, because each would reach files that #1698 touches or would change output:

• acquire_or_emit(&GlobalArgs) (T09/Acquire apply.lock through one lock_cli helper that takes &GlobalArgs; the sequence is open-coded 9 times and renders lock_held 3 ways #1858)
• the patch_server_origins forwarder (callers in get.rs / vendor.rs)
• global_scope_flag vs ui::global_scope_arg (they print different text)
• hosted_bundle's json! error and its hand-built ApiClient (both would change output or behavior)

Size

git diff --stat: 6 files, +160 / −300.

• Production: about −105
• Tests: about −35

Behavior change

None. Output bytes, JSON, exit codes and error codes are unchanged. Removing the dead hosted_unsupported branch changes nothing, because that list was always empty.

Verification

• cargo clippy --workspace --all-features -- -D warnings: clean.
• cargo test -p socket-patch-cli --lib: 936 passed.
• --test rollback, remove, in_process_rollback_hosted, in_process_rollback_vendored, scan_get_mode_from_project_state, e2e_safety_lock, cli_parse_list, command_module_layering and hosted_memory_parity: all pass.
• No platform-specific code was touched.

/code-review high

Six findings. I fixed three:

• acquire_or_emit now reuses format_lock_error.
• The lock_failure doc no longer overclaims.
• The merged test is renamed.

I did not fix three:

• list --json doesn't report an unreadable vendor ledger in warnings[]. This is pre-existing; I only moved the code. Fixing it would change JSON output, so it belongs in its own issue.
• The deleted determinism test "wasn't a tautology." It was: it reused one manifest instance, so both builds iterate the same HashMap.
• The mode tests copy the production message. That is deliberate: expected values are written independently of the code under test.

Risk

Low. The changes are mechanical inlining and deletions, and the touched paths are covered by the suites above.

🤖 Generated with Claude Code

https://claude.ai/code/session_011f9aeQq8rjUYApktQhf6WR

Generated by Claude Code