refactor(config): split generator-owned contract from settings

[MintsInc]

Context

openapi-transformer's pup generator copies its config asset over src/config.rs on every run. Any hand edit to config.rs in pup gets reverted by the next generated PR unless someone also mirrors it into the asset. That has happened twice: the regional OAuth fix (422e481) had to be re-synced into the asset after a generated PR reverted it. #908 currently deletes auth_config_dir() and the XDG-aware config_file_candidates() from #891 (3f5f913). Merging #908 would quietly move macOS auth storage back to the wrong directory.

This is the same fix as the formatter split (b112d16, 4e45407): a small generator-owned contract, with the real logic in a hand-owned module.

Changes

• src/config.rs becomes a generator-owned shim that re-exports only Config and OutputFormat. Generated code, and the formatter.rs/client.rs assets, use nothing else from crate::config (checked against the generator's assets and templates).
• The re-export lists the two items explicitly instead of using pub use crate::settings::*. A glob would put everything back behind the generator-owned path and make the boundary meaningless.
• The previous config.rs moves to src/settings.rs unchanged apart from dropping the DO NOT EDIT header. Callers of anything beyond Config/OutputFormat now import from crate::settings. The ~140 existing crate::config::{Config, OutputFormat} imports are left alone because they go through the contract.
• Split into two commits so git tracks the config.rs → settings.rs rename and git log --follow keeps working. The first commit doesn't build on its own; the second adds the shim.
• No behavior change.

Tests

• Pure move. Existing unit tests move with the code into settings.rs, including the four auth_config_dir tests and the two macOS config_file_candidates tests from macOS: file token storage writes to ~/Library/Application Support/pup, not the documented ~/.config/pup #891.
• cargo fmt --check passes locally. Clippy, tests and the WASM builds have not been run locally yet; CI covers them.

Appendix

• Sequencing: the generator's config asset must be replaced with a byte-identical copy of this shim right after this merges. Until it is, a generated PR would overwrite the shim with the old 2,645-line asset and fail to compile because of duplicate definitions. Generated PRs are already broken today (opt downtime list into pup generation #908), so nothing working gets worse.
• After the asset swap, opt downtime list into pup generation #908 should be regenerated rather than hand-edited.
• Follow-up: a CI guard that fails hand-written PRs touching files with the Code generated by openapi-transformer header, unless a bypass label is set.

🤖 Generated with Claude Code

[datadog-datadog-prod-us1]

More details

The existing configuration implementation is preserved, helper callers use settings, and both CLI and browser entry points declare the new module. The completed static review found no reportable regression in the Config and OutputFormat contract.

Was this helpful? React 👍 or 👎

_{Open Bits AI session}

_{🤖 Bits Code Review · Commit f8daf4d}