perf: Move ShowWorkspace CI polling off the UI IPC path

[Ziinc]

Summary

When a workspace is opened, queue a single out-of-band PR info + CI checks refresh so the Rust background cache is warmed before the next poll tick. The UI keeps reading cache-only; the user should not see stale PR/CI state after switching workspaces.

Changes

• Rust: PrStatusManager::queue_branch_refresh / refresh_branch_now — one call fetches PR + CI for a branch; queue is fire-and-forget and coalesces in-flight (repo, branch).
• Tauri / NAPI: refresh_pr_branch_status returns immediately (does not block the JS event loop on gh).
• Frontend: usePrInfoViaGh and usePrCiStatus call refreshPrBranchStatus on mount when a branch is present.
• Tests: Rust coverage for async queue + coalescing; hook tests assert the refresh is queued; screenshot specs mock the new API.

Test plan

• cargo test --lib pr_status
• npm run test:run -- test/integration/useMergeQueueStatus.test.ts
• npm run test:run -- src/components/CiStatusIndicator.test.tsx
• App-qa: ci-status-indicator success / failure / pending screenshots

  

[cursor]

Same stutter class as #218, but for CI: ShowWorkspace’s 15s gh pr checks was still on the command path.

• Background poller now also caches CI; only shells out when a PR exists for the branch
• UI reads stay cache-only; pr-statuses-updated carries ci_statuses so React Query updates without another gh

poll_one_repo

[cursor]

Force-fetch / panel paths still need gh, but must not stall IPC the way sync commands did (same pattern as gh_create_pr in #217).

• get_pr_info_via_gh, get_pr_checks_*, and gh_list_pr_review_threads now use spawn_blocking
• Force-fetch also warms the background CI cache so the next UI read is immediate

get_pr_checks_via_gh

[cursor]

usePrCiStatus mirrors usePrInfoViaGh: start the Rust poller, listen for cache events, refetch from memory every 15s.

• Removes the ShowWorkspace stutter source (sync getPrChecksViaGh on interval)
• invalidatePrStatuses still force-warms both PR + CI after create-PR mutations

usePrCiStatus

[cursor]

CI now has its own 30s cadence (DEFAULT_CI_POLL_INTERVAL), independent of 60s PR metadata.

• Mid-cycle tick calls refresh_ci_now / poll_ci_only — gh pr checks only, no gh pr view
• UI still only reads get_cached_pr_ci_status; force-fetch get_pr_checks_via_gh is spawn_blocking + cache warm

DEFAULT_CI_POLL_INTERVAL · poll_ci_only

[cursor]

• Workspace open already starts the poller; this queues an immediate branch warm so the first paint is not stuck on a stale 30s/60s tick.
• Same call from usePrCiStatus — coalesced on the Rust side when both mount together.
• Permalink:

  treq/src/hooks/useMergeQueueStatus.ts

  Lines 160 to 165 in 5fffc89

  	useEffect(() => {
  	if (!repoPath) return;
  	void startPrStatusPolling(repoPath);
  	if (branchName) {
  	void refreshPrBranchStatus(repoPath, branchName);
  	}

[cursor]

• One call warms both PR metadata and CI for the opened branch; UI stays on cache reads.
• Fire-and-forget + inflight coalesce so double-mount (PR hook + CI hook) does not spawn two gh shells.
• Permalink:

  treq/src-tauri/src/pr_status.rs

  Lines 210 to 227 in 5fffc89

  	pub fn queue_branch_refresh(&self, repo_path: String, branch_name: String) {
  	self.ensure_started();
  	let key = format!("{repo_path}\0{branch_name}");
  	{
  	let mut inflight = self.inner.branch_refresh_inflight.lock().unwrap();
  	if !inflight.insert(key.clone()) {
  	return;
  	}
  	}
  	let inner = Arc::clone(&self.inner);
  	thread::Builder::new()
  	.name("pr-status-branch-refresh".into())
  	.spawn(move || {
  	poll_one_branch(&inner, &repo_path, &branch_name);
  	inner.branch_refresh_inflight.lock().unwrap().remove(&key);
  	})
  	.expect("failed to spawn pr-status-branch-refresh thread");
  	}

[cursor]

• Must return before gh finishes: invokeSync runs on the Node event loop, so a blocking refresh previously timed out Vitest and stuttered the UI path.
• Permalink:

  treq/crates/treq-napi/src/dispatch.rs

  Lines 106 to 114 in 5fffc89

  	"refresh_pr_branch_status" => {
  	let repo_path = get_str(&args, "repoPath")?;
  	let branch_name = get_str(&args, "branchName")?;
  	treq_lib::pr_status::global().watch_repo(&repo_path);
  	// Fire-and-forget — same as the Tauri command. Must not block the
  	// JS event loop on `gh` (which is what caused test timeouts when
  	// a spy was briefly restored to the real invoke).
  	treq_lib::pr_status::global().queue_branch_refresh(repo_path, branch_name);
  	Ok(Value::Null)