Files
orca/src/renderer
Jinjing 2f4e9aed70 fix(worktrees): address review feedback on the worktrees slice split (#14669)
* fix(worktrees): address review feedback on the worktrees slice split

Follow-ups to #14643. All of these predate the split (the constructs moved
verbatim from worktrees.ts), so they are behavior fixes, not refactor fallout.

- fetchAllWorktrees: route runtime-scope-forbidden errors through the toast in
  the hydration path, matching the fast path
- updateWorktreeMeta / ensureHostedReviewPushTarget: use
  trySettingsForWorktreeOwner so an ambiguous owner skips push-target resolution
  instead of throwing past the { ok, error } contract
- updateWorktreeMeta: compare linkedPR with ?? null so an unknown worktree does
  not trigger a spurious hosted-review refresh (matches the four sibling providers)
- persistWorktreeMeta: localize the two runtime-capability messages
- buildWorktreeRenameState: fix a doc comment that contradicted the code
- setWorktreesPinnedAndReveal: skip updateWorktreesMeta on an empty map
- pruneLastVisitedTimestamps: treat an empty worktreesByRepo list as unhydrated,
  and clear activeWorkspaceKey alongside a stale activeWorktreeId
- forceDeletePreservedBranch: fail closed when several retained cleanups match
  and no host was given, rather than routing to the active runtime
- purgeOrphanedRuntimeSshProjects: drop blank target ids so a repo with no
  connectionId cannot match
- markWorktreesDeleting: make the skip guard phase-aware so queued rows are
  promoted to deleting
- worktree purge: validate right-sidebar tabs with normalizeRightSidebarRoute so
  pr-checks and plugin panels are not silently dropped

* test(worktrees): add edge case tests from review feedback

Add defensive tests covering phase-aware deletion state transitions, tab
preservation across worktree purges, graceful handling of ambiguous owner
scenarios, and stale workspace cleanup. These catch edge cases that could
lead to silent failures or invalid state transitions.

* fix(worktrees): preserve workspace key pointing at live worktree

Replace overly broad scope check with direct equality check, so pruning
only drops the stale worktree's own workspace key while preserving keys
pointing to other live worktrees or folder workspaces.

* fix(worktrees): clear a legacy unprefixed active workspace key on prune

pruneLastVisitedTimestamps only dropped the `worktree:`-prefixed derived key,
so sessions predating the prefix kept a phantom workspace selected. Match the
purge path, which already treats a bare worktree id as a workspace key.

* chore(i18n): re-key the two runtime-capability strings to the canonical scheme

The keys added when persistWorktreeMeta was localized were hand-written. Derive
them the way localize-renderer-strings.mjs does — module path plus
sha1(path:text) — so a drift audit against the generator does not flag them.

* fix(worktrees): stop updateWorktreeLineage rejecting past its never-rejects contract

The sidebar's "Remove parent link" (WorktreeContextMenu.handleRemoveParentLink)
awaits updateWorktreeLineage in a bare `void Promise.all(...)` with no catch, so
the action must always resolve. Two paths broke that:

- settingsForWorktreeOwner ran outside the try and throws on an ambiguous owner.
  Skip with warnAmbiguousOwnerOnce instead, matching the ensure-push-target fix.
- the recovery refresh inside the catch has no internal try/catch, so a failing
  lineage RPC rejected out of the handler meant to absorb the failure.

assignWorktreeParent keeps rethrowing (both its callers catch and toast), but its
recovery refresh is now best-effort too so it can't mask the original cause.

Also: localize the new ambiguous-host preserved-branch throw (it reaches a toast
description; the throw above it stays verbatim because it mirrors a main-process
message), and drop blank SSH target ids once so both purge lookups agree.

* fix(worktrees): surface un-nest failures instead of silencing them

Supersedes the approach in 789edeaa7f. That commit fixed the unhandled rejection
in WorktreeContextMenu.handleRemoveParentLink by making updateWorktreeLineage
never reject, which had two costs:

- it made the .catch in use-lineage-drop-commit unreachable, so a drag-unnest
  failure went from an error toast to a silent no-op; and
- trySettingsForWorktreeOwner returns null for BOTH 'ambiguous' and 'missing'
  route resolutions (worktree-operation-route.ts collapses the three-way result),
  so the skip also swallowed unresolvable-route cases while logging "identity is
  ambiguous", which is the wrong cause for a legacy row on a multi-runtime setup.

Fix the caller instead: updateWorktreeLineage keeps rejecting, and both call
sites go through a shared unnestWorktrees() that catches and toasts. The
best-effort recovery refresh from 789edeaa7f is kept — that part was correct and
stops a failing refresh masking the original error on the assign path.

unnestWorktrees also de-duplicates the identical console.error/toast pair the two
callers had, and reuses the existing failedUnnestWorkspace key (already
translated in all five locales), so no new string ships.
2026-08-16 01:00:36 -07:00
..