diff --git a/ROADMAP.md b/ROADMAP.md index c3bb1d10..6589dfb0 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -720,13 +720,12 @@ scale-&-reach horizon): limitation across different Git repos). - **Richer trust models** beyond trusted-key lists (delegation, expiry) — P22 ships the key-list model. -- **Sidecar-cleanup footgun in `rebase_continue`.** `replay.rs:1059`'s - sidecar sweep blind-unlinks `{path}.theirs` with neither the - non-text-kind gate nor the tracked-path guard that P23 added to - `resolve_path` — so completing a stopped rebase can delete an untracked - user file named `foo.txt.theirs`. Pre-existing (predates P23); flagged - at the P23 final review as the twin of the resolve fix. Apply the same - two guards. +- **Sidecar-cleanup footgun in `rebase_continue` — half fixed (P37c).** + `remove_conflict_sidecars` now skips paths the completion's parent tree + tracks, but it still has no non-text-kind gate: a text conflict writes no + `.theirs` sidecar, so an untracked user file named `foo.txt.theirs` is + still deleted when the stop's conflict path is `foo.txt`. Pass the conflict + kind (or the written sidecar list) to the sweep to close it. - **In-progress guard for the RECEIVING repo on push.** `LocalTransport::update_ref` moves a branch tip with only a CAS/ff gate and no in-progress check on the receiving side — a push into a repo @@ -785,11 +784,9 @@ scale-&-reach horizon): `sc protect … rotate`-style genuine per-path cryptographic cutover — the protected-path analogue of `sc secret rotate` — is now coherent. Recorded in ADR-0043, not built. -- **`sc rewrap` on partial clones (P33 final-review follow-on).** `sc rewrap` - loads each protected plaintext whole into RAM and has no - `PartialCloneUnsupported` guard, unlike merge/rebase/ws-fork/ws-harvest/`sc - work` (P27); a paths-only/streaming tree rebuild (or the same refusal-first - posture) is the follow-on. +- **`sc rewrap` on partial clones — refused (P37).** `Repo::rewrap` calls + `refuse_on_partial`; a paths-only/streaming rebuild that would lift the + refusal remains a possible follow-on. - **Protected-cache hygiene (P33 final-review follow-on).** `.sc/protected-cache` and `.sc/ws/cache-` entries for a deleted or renamed path are never pruned — unbounded but benign growth (a stale entry just goes unmatched).