[feat] Batch GitHub PR status fetching via GraphQL for faster sidebar updates #2958

Describe the feature

Problem

The sidebar currently fetches GitHub PR status one workspace at a time using individual gh pr view CLI calls. Each call spawns a child process, waits for the GitHub REST API round-trip, and parses the JSON response — all sequentially gated behind a hover interaction. This means:

• PR status is invisible until the user hovers over a workspace item.
• With 10+ workspaces, the N+1 pattern causes noticeable latency and unnecessary GitHub API usage.
• Each gh pr view call is ~200-400ms, so 15 workspaces = 3-6 seconds of sequential shell spawns.

Solution

Replace the per-workspace gh pr view calls with a single batched GitHub GraphQL query that fetches PR status for all active workspaces at once, then distributes results into per-workspace React Query caches.

Architecture

Main process (Electron)

• New batchFetchGitHubPRStatuses() function that:
  • Groups all active worktree workspaces by GitHub repository (via cached RepoContext)
  • Builds a single GraphQL query with aliased repository and pullRequests fields
  • Executes via gh api graphql (one shell spawn total, with 30s timeout)
  • Parses response with Zod discriminatedUnion schemas, normalizes GraphQL shape → existing GHPRResponse format
  • Populates the in-memory githubStatusResource cache and persists meaningful changes to SQLite
• New batchGetGitHubStatuses tRPC procedure exposes this to the renderer

Renderer (React)

• New useBatchGitHubStatus() hook called at the WorkspaceSidebar level
  • Polls batchGetGitHubStatuses every 15 seconds
  • Seeds empty per-workspace getGitHubStatus React Query cache slots via setData
  • Never overwrites richer data from per-workspace fetches (which include previewUrl, accurate branchExistsOnRemote)
• Removed hasHovered gate from WorkspaceListItem and WorkspaceRow — PR status appears immediately
• Aligned workspace-list-item stale time with batch polling interval (15s)

Key design decisions

1. GraphQL over REST: One query fetches PRs for all repos/branches with status checks and review requests. Stays well within GitHub's 5,000 points/hour rate limit.
2. Seed-only cache strategy: Batch results only fill empty cache slots. Per-workspace queries (triggered on hover/focus) return richer data — we never regress it.
3. Discriminated unions: Zod schemas use __typename as the discriminator for CheckRun/StatusContext and User/Team unions — correct GraphQL pattern, better parse performance.
4. Branch deduplication: Multiple workspaces on the same branch within a repo share a single pullRequests field in the query, with O(1) Map-based dedup.
5. Graceful degradation: If the batch query fails (auth, network, timeout), the sidebar falls back to the existing per-workspace fetching on hover — zero breakage.

Files changed

Performance impact