mirror of
https://github.com/openai/codex.git
synced 2026-10-20 19:17:37 +00:00
Celia Chen
5e0d1ef83e
New local TUI sessions in 0.155 request detailed reasoning summaries by default, which causes request rejection on providers that do not support reasoning summaries. Restore `none` as the default while honoring explicit `auto`, `concise`, and `detailed` settings and concurrent delivery when summaries are explicitly enabled. This focused backport targets `release/0.155`, based on the content commit for `rust-v0.155.0`. It includes regression coverage for new-thread settings and the outbound Responses request. Validation: - Patch applies cleanly to the 0.155 release source; `git diff --check` passes. - All 9 focused reasoning-default and reasoning-status tests passed, including the outbound Responses request and existing snapshots. - `just fmt` completed; the hotfix Rust files were unchanged. Unrelated justfile rewrites from the locally installed formatter were discarded. - All 4,589 non-ignored TUI tests passed across the rerun and a follow-up of the remaining 196 tests after a runner disconnection. The earlier cursor failures were caused by `NO_COLOR=1` / `TERM=dumb`; the tests pass with normal terminal settings and no code or snapshot changes. - Release-build checks passed on Linux, macOS, and Windows. CI currently reports failures in unchanged release-base files: - [Codespell](https://github.com/openai/codex/actions/runs/35380630185/job/105715678757) flags the existing `OFlags::WRONLY` identifier. - [Manifest validation](https://github.com/openai/codex/actions/runs/35380630185/job/105715679235) finds a stale feature exception for `codex-rs/code-mode/Cargo.toml`. - [Dependency audit](https://github.com/openai/codex/actions/runs/35380630185/job/105715679009) reports RUSTSEC-2026-0285 for the existing `rustls 0.23.36` lockfile entry. - [cargo shear](https://github.com/openai/codex/actions/runs/35380630185/job/105716110806) reports an existing unlinked file, `codex-rs/login/src/internal_identity.rs`. Additional CI failures are also outside the changed TUI files: an unused `body_json` import in `core/tests/suite/openai_file_mcp.rs` (Bazel Clippy), a format-driver expectation in the Python SDK tests, and a transient `Text file busy` error while spawning a code-mode test helper (musl). These paths are unchanged from the release base.