mirror of
https://github.com/stablyai/orca.git
synced 2026-10-19 16:03:26 +00:00
fix(e2e): repair release e2e suite — parking regression tests, stale/flaky specs, profile switcher gate
Diagnosed 20 failing tests across the release e2e shards. Most are test debt, plus two genuine product-side issues. Product fixes: - OrcaProfileSwitcher: the PROD gate hid the "Switch profile" button in the e2e build (electron-vite build bakes NODE_ENV=production). Exempt MODE==='e2e' so the specs render it while packaged prod builds stay hidden. Parking cluster (8 tests): #8262 intentionally keeps the most-recently-hidden tab warm (exempt from cold-park). The specs hid exactly one tab — always the exempt one — so it never parked. Open a throwaway decoy tab that absorbs the last-active exemption so the target parks. (terminal-hidden-view-parking, terminal-pane-close-layout-consistency) Stale tests updated to match intended product behavior: - rich-markdown-link-bubble: match Edit link by aria-label (title dropped in #8307) - terminal-codex-hidden-startup-background: drop the dead hiddenRendererSkipCount poll (Phase-4 main-side delivery gate #7214 bypasses that renderer path) Brittle threshold/geometry/timing hardening (no product regression): - agent-session-log-tail-stability: assert full-model length instead of a machine-specific word-wrap pixel baseline - artificial-opencode revisit: dedicated under-backpressure latency bound - terminal-history-size-typing-latency: gate p90 not max (tolerate one checkpoint-in-window spike; median stays strict) - combined-diff-scroll-restore: assert viewport barely moved vs exact anchor key - terminal-shortcuts: idempotent kitty-flag reset instead of a racing stack pop - agent-session-live-force-exit-resume: drive the product quit-capture path - renderer-crash-recovery-terminal-input: poll the transport probe over the recovery budget (still flags a permanently frozen pane) terminal-push-delivery-loss-recovery left unchanged (no safe test-only improvement; recovery is wall-clock bounded with ample slack).
This commit is contained in:
Jinjing
@@ -28,7 +28,7 @@
|
||||
"providers": ["local"],
|
||||
"coveredPlatforms": ["macos"],
|
||||
"coveredProviders": ["local"],
|
||||
"coverageNotes": "Focused tests and real-Monaco 9/50 MiB benchmarks are platform-independent. Local macOS Electron evidence opens a synthetic 9 MiB transcript through Agent Session History at fixed 900x720 viewport, 13px font, 1x zoom, and the measured 2,775,880px legacy-failure geometry, then verifies three five-second-cadence watcher appends with Find open and closed. Live Windows/Linux evidence remains uncollected.",
|
||||
"coverageNotes": "Focused tests and real-Monaco 9/50 MiB benchmarks are platform-independent. Local macOS Electron evidence opens a synthetic 9 MiB transcript through Agent Session History at fixed 900x720 viewport, 13px font, 1x zoom, and asserts the full 9 MiB model length loaded as a font-metric-independent containment check (word-wrap pixel geometry varies ~10% across runners, so a generous content-height floor is only a collapsed/truncated-render smoke check), then verifies three five-second-cadence watcher appends with Find open and closed. Live Windows/Linux evidence remains uncollected.",
|
||||
"motivatingLinks": ["https://github.com/stablyai/orca/pull/8432"],
|
||||
"invariant": "Append-only external file growth changes only Monaco's model suffix, retaining the viewport, selection, Find state, and renderer liveness above the append point; arbitrary rewrites continue to replace the model content.",
|
||||
"oracle": "Focused tests assert one post-mount content owner, actual outer lifecycle remount ordering across retained path models, exact end-of-model suffix edits with one model read, no-op equality, and full replacement for non-appends. With Node forced GC, real-Monaco benchmarks alternate 30 suffix and 30 replacement samples after five warmups on fresh equivalent models at 9 and 50 MiB, forcing GC and event-loop settlement between every arm. The Electron scenario alternates an e2e-only legacy setValue red control and the fixed watcher append from restored equivalent model/geometry at the measured legacy-failure cadence, asserting that the control disrupts anchor state while the fixed path preserves visible ranges, selection, complete Find state, scroll offset, renderer survival, and forced-GC heap/native-memory budgets.",
|
||||
@@ -68,7 +68,7 @@
|
||||
{
|
||||
"file": "tests/e2e/agent-session-log-tail-stability.spec.ts",
|
||||
"assertions": [
|
||||
"production Agent Session History opens a synthetic 9 MiB View Log at the measured 2,775,880px legacy-failure geometry",
|
||||
"production Agent Session History opens a synthetic 9 MiB View Log and confirms the full model length loaded as font-metric-independent containment, with a generous content-height floor as a collapsed/truncated-render smoke check",
|
||||
"an executable e2e-only legacy setValue control disrupts selection/Find/anchor state at each fixed-geometry five-second sample, then restores the equivalent model state before the fixed arm",
|
||||
"three alternating watcher suffix appends preserve visible ranges, selection, scroll offset, Find open/query/active-match state, and exact suffix content",
|
||||
"the renderer remains responsive with no render-process-gone event and forced-GC JS-heap/working-set/private-memory peak and retained budgets hold against paired legacy controls"
|
||||
@@ -106,7 +106,7 @@
|
||||
],
|
||||
"runtimeBudget": { "p95Seconds": 600, "scope": "local focused renderer tests plus one Electron production-journey scenario" },
|
||||
"flakeHistory": { "status": "unknown", "evidence": "New deterministic gate with local macOS passes; CI soak history is not yet available." },
|
||||
"redGreenEvidence": { "status": "complete", "evidence": "The retained Electron gate itself executes the former read-only wrapper setValue behavior behind MODE=e2e at fixed 900x720, 13px, 1x zoom, 2,775,880px content height, and five-second cadence. The control retains a flat incoming value like the former controlled IPC prop. Each legacy arm must disrupt selection, Find active-match state, visible range, or scroll anchor, then restore identical model length/tail/geometry and anchor state before the fixed watcher arm; every fixed arm must preserve them. Production builds never install the control. The original exact-file dev repro also recorded renderer exit code 5." },
|
||||
"redGreenEvidence": { "status": "complete", "evidence": "The retained Electron gate itself executes the former read-only wrapper setValue behavior behind MODE=e2e at fixed 900x720, 13px, 1x zoom, the full 9 MiB model length loaded (with a generous content-height floor, since word-wrap pixel geometry is runner-dependent), and five-second cadence. The control retains a flat incoming value like the former controlled IPC prop. Each legacy arm must disrupt selection, Find active-match state, visible range, or scroll anchor, then restore identical model length/tail/geometry and anchor state before the fixed watcher arm; every fixed arm must preserve them. Production builds never install the control. The original exact-file dev repro also recorded renderer exit code 5." },
|
||||
"performanceBudget": { "required": true, "evidence": "Every update retrieves the model value once and performs at most one equality-or-prefix comparison; a matching append submits only the suffix. The registered Node command requires --expose-gc and --pool=threads so worker GC is available; alternating fresh real-Monaco arms force GC and deterministic event-loop settlement after every operation. Observed p95 ranges: 9 MiB append 4.02-5.51ms versus replacement 81.42-83.76ms; 50 MiB append 22.72-26.60ms versus replacement 445.11-449.21ms. Electron alternates each legacy replacement control and fixed watcher suffix from equivalent restored model state/geometry, samples forced-GC JS heap plus app.getAppMetrics renderer working set and OS private memory before/after/settled, asserts retained memory within max(20MiB,10%), and requires suffix peak deltas for jsHeapMb, workingSetMb, and privateMb not exceed the paired legacy controls." },
|
||||
"promotionCriteria": [
|
||||
"Collect stable soak history on macOS, Linux, and Windows.",
|
||||
|
||||
@@ -87,11 +87,11 @@ export function OrcaProfileSwitcher({
|
||||
}, [fetchProfiles, loading, profiles.length])
|
||||
|
||||
// Why: the Orca Cloud account UX isn't ready for production users yet, so the
|
||||
// trigger stays hidden in packaged builds. Dev builds still show it when
|
||||
// cloud auth is configured, and a dev-only Settings > Dev Tools > Orca Cloud
|
||||
// section mirrors it. import.meta.env.PROD is true only in packaged builds
|
||||
// (false in dev and under vitest), so tests still exercise the render path.
|
||||
if (import.meta.env.PROD) {
|
||||
// trigger stays hidden in packaged builds. Dev builds still show it when cloud
|
||||
// auth is configured. electron-vite build forces NODE_ENV=production, so PROD
|
||||
// is baked true even for the e2e bundle; exempt MODE==='e2e' so the switcher
|
||||
// specs can exercise the render path while real packaged prod builds stay hidden.
|
||||
if (import.meta.env.PROD && import.meta.env.MODE !== 'e2e') {
|
||||
return null
|
||||
}
|
||||
|
||||
|
||||
@@ -196,6 +196,11 @@ test('resumes a live agent record after force-exit restart when pane PTY ownersh
|
||||
}
|
||||
)
|
||||
|
||||
// Drive the product's own quit-capture path (what the 60s timer / beforeunload
|
||||
// run) so the live record is flushed deterministically instead of racing the
|
||||
// debounced session writer before the poll below.
|
||||
await page.evaluate(() => window.__store?.getState().captureAllSleepingAgentSessions())
|
||||
|
||||
await expect
|
||||
.poll(() => persistedLiveRecordExists(session.userDataDir), {
|
||||
timeout: 15_000,
|
||||
|
||||
@@ -57,9 +57,12 @@ test.describe('Agent Session History live log', () => {
|
||||
|
||||
let baseline = await restoreAnchorState(orcaPage, true)
|
||||
console.log(`[live-log-stability] geometry ${JSON.stringify(baseline)}`)
|
||||
// The legacy whole-model baseline at this fixed viewport/font/zoom moved
|
||||
// the anchor at 2,775,880px; missing that geometry cannot claim containment.
|
||||
expect(baseline.contentHeight).toBeGreaterThanOrEqual(2_775_880)
|
||||
// Why: containment is proven by the full 9 MiB model length, which is
|
||||
// font-independent. Word-wrap pixel geometry varies ~10% across runner font
|
||||
// metrics (macOS baseline 2,775,880px; Linux CI ~2,498,292px), so keep only a
|
||||
// generous content-height floor as a collapsed/truncated-render smoke check.
|
||||
expect(baseline.valueLength).toBe(fixture.initialLength)
|
||||
expect(baseline.contentHeight).toBeGreaterThanOrEqual(2_000_000)
|
||||
expect(baseline.visibleRanges.length).toBeGreaterThan(0)
|
||||
expect(baseline.find).toMatchObject({ open: true, query: ANCHOR_TOKEN })
|
||||
expect(baseline.find.activeMatch).not.toBe('')
|
||||
|
||||
@@ -90,6 +90,7 @@ export async function runRendererBackpressureRevisitScenario<
|
||||
deps,
|
||||
maxMedianKeyLatencyMs,
|
||||
maxRendererSchedulerQueuedChars,
|
||||
maxRevisitLatencyMs,
|
||||
maxTimerDriftMs,
|
||||
maxWorstKeyLatencyMs,
|
||||
mainRendererPressureTargetChars,
|
||||
@@ -102,6 +103,7 @@ export async function runRendererBackpressureRevisitScenario<
|
||||
deps: RevisitPressureDeps<TMeasurement, TDebug, TScheduler, TMainPressure, TAckGate>
|
||||
maxMedianKeyLatencyMs: number
|
||||
maxRendererSchedulerQueuedChars: number
|
||||
maxRevisitLatencyMs: number
|
||||
maxTimerDriftMs: number
|
||||
maxWorstKeyLatencyMs: number
|
||||
mainRendererPressureTargetChars: number
|
||||
@@ -205,7 +207,10 @@ export async function runRendererBackpressureRevisitScenario<
|
||||
1
|
||||
)}ms heldAckChars=${ackGate?.heldAckChars ?? 0}`
|
||||
})
|
||||
expect(revisitLatencyMs).toBeLessThan(maxWorstKeyLatencyMs)
|
||||
// Why: this printf is measured after a worktree switch/focus while the
|
||||
// background panes are still ACK-gate-held, so it gets its own under-load
|
||||
// bound rather than the unloaded worst-key budget.
|
||||
expect(revisitLatencyMs).toBeLessThan(maxRevisitLatencyMs)
|
||||
|
||||
await deps.releaseTerminalAckGate(orcaPage)
|
||||
await deps.focusPane(orcaPage, loadPanes[0]?.paneKey ?? revisitPane.paneKey)
|
||||
|
||||
@@ -131,6 +131,11 @@ const MAX_WORST_KEY_LATENCY_MS = 300
|
||||
// a catastrophic-hang detector. Mirrors ssh-docker-relay-perf's 2s worst-key
|
||||
// tolerance and the hidden-pressure scenario's relaxed worst budget.
|
||||
const MAX_WORST_KEY_LATENCY_UNDER_LOAD_MS = 3_000
|
||||
// Why: the post-revisit printf is sampled while the background panes are still
|
||||
// ACK-gate-held and through a whole-buffer serialize poll, so it inherits the
|
||||
// same environment-dominated worst-case as typing under load; the unloaded
|
||||
// 300ms budget has no margin for this switch-and-focus moment.
|
||||
const MAX_REVISIT_LATENCY_UNDER_LOAD_MS = 3_000
|
||||
// Why: GitHub's two-worker Electron shards can briefly starve renderer timers
|
||||
// without visible typing lag. Keep this as a smoke gate, not a CPU lottery.
|
||||
const MAX_TIMER_DRIFT_MS = 250
|
||||
@@ -604,6 +609,7 @@ test.describe('Artificial OpenCode terminal load', () => {
|
||||
mainRendererPressureTargetChars: MAIN_RENDERER_PRESSURE_TARGET_CHARS,
|
||||
maxMedianKeyLatencyMs: MAX_MEDIAN_KEY_LATENCY_MS,
|
||||
maxRendererSchedulerQueuedChars: MAX_RENDERER_SCHEDULER_QUEUED_CHARS,
|
||||
maxRevisitLatencyMs: MAX_REVISIT_LATENCY_UNDER_LOAD_MS,
|
||||
maxTimerDriftMs: MAX_TIMER_DRIFT_MS,
|
||||
maxWorstKeyLatencyMs: MAX_WORST_KEY_LATENCY_MS,
|
||||
orcaPage,
|
||||
|
||||
@@ -423,7 +423,10 @@ test.describe('Combined diff scroll restore', () => {
|
||||
await clickVisibleDiffLine(orcaPage)
|
||||
const afterLineClick = await waitForStableViewportAnchor(orcaPage)
|
||||
|
||||
expect(afterLineClick.key).toBe(afterSwitch.key)
|
||||
// Assert the viewport barely moved rather than an exact anchor key: sections
|
||||
// are ~viewport-sized, so a sub-pixel focus scroll from the click can flip the
|
||||
// topmost-visible key by one without meaningfully moving the scroll position.
|
||||
expect(Math.abs(afterLineClick.scrollTop - afterSwitch.scrollTop)).toBeLessThan(40)
|
||||
expect(Math.abs(afterLineClick.top - afterSwitch.top)).toBeLessThan(80)
|
||||
} finally {
|
||||
rmSync(fixture.repoPath, { recursive: true, force: true })
|
||||
|
||||
@@ -139,7 +139,28 @@ test.describe('Renderer crash recovery keeps terminal input alive', () => {
|
||||
// should reattach to the same session rather than spawn a fresh shell.
|
||||
const postPtyIds = await mainGetStorePtyIds(electronApp)
|
||||
|
||||
const transportAlive = await mainProbeTransportPaste(electronApp, `PASTE_POST_${cycle}_OK`)
|
||||
// Why: the post-crash transport reattach is async — a single fire-and-forget
|
||||
// probe can race the reconnect window. Poll over the recovery budget so a
|
||||
// transient unbound-transport window does not fail the test; a transport
|
||||
// still dead after the full budget is the real frozen-pane bug reported below.
|
||||
let transportAlive = false
|
||||
try {
|
||||
await expect
|
||||
.poll(
|
||||
async () => {
|
||||
transportAlive = await mainProbeTransportPaste(
|
||||
electronApp,
|
||||
`PASTE_POST_${cycle}_OK`,
|
||||
5_000
|
||||
)
|
||||
return transportAlive
|
||||
},
|
||||
{ timeout: RECOVERY_TIMEOUT_MS }
|
||||
)
|
||||
.toBe(true)
|
||||
} catch {
|
||||
transportAlive = false
|
||||
}
|
||||
const directAlive =
|
||||
postPtyIds.length > 0 &&
|
||||
(await mainProbeDirectWrite(electronApp, postPtyIds[0], `DIRECT_POST_${cycle}_OK`))
|
||||
|
||||
@@ -102,7 +102,9 @@ test.describe('Rich markdown link bubble stacking', () => {
|
||||
expect(overlap.overlapWidth).toBeGreaterThan(8)
|
||||
expect(overlap.topElementIsBubble).toBe(true)
|
||||
|
||||
const editButton = bubble.getByTitle('Edit link')
|
||||
// The Edit action exposes its label via aria-label (a shadcn Button +
|
||||
// Radix tooltip), not a title attribute, so match by accessible name.
|
||||
const editButton = bubble.getByRole('button', { name: 'Edit link' })
|
||||
const editButtonBounds = await editButton.boundingBox()
|
||||
const explorerBounds = await explorer.boundingBox()
|
||||
expect(editButtonBounds).not.toBeNull()
|
||||
|
||||
@@ -13,12 +13,6 @@ import {
|
||||
import { getTerminalContent, waitForActiveTerminalManager } from './helpers/terminal'
|
||||
import { BACKGROUND_MOUNT_TERMINAL_WORKTREE_EVENT } from '../../src/renderer/src/constants/terminal'
|
||||
|
||||
type HiddenOutputDebugSnapshot = {
|
||||
hiddenRendererSkipCount: number
|
||||
hiddenRendererSkippedChars: number
|
||||
hiddenRendererMode2031ReplyCount: number
|
||||
}
|
||||
|
||||
type CodexStartupBackgroundTarget = {
|
||||
clip: { x: number; y: number; width: number; height: number }
|
||||
cellWidth: number
|
||||
@@ -28,13 +22,6 @@ type CodexStartupBackgroundTarget = {
|
||||
modelBackgroundCells: number
|
||||
}
|
||||
|
||||
type CodexStartupBackgroundWindow = Window & {
|
||||
__terminalPtyOutputDebug?: {
|
||||
reset: () => void
|
||||
snapshot: () => HiddenOutputDebugSnapshot
|
||||
}
|
||||
}
|
||||
|
||||
const COMPOSER_BG = { red: 72, green: 72, blue: 72 }
|
||||
|
||||
function codexLikeStartupCommand(marker: string): string {
|
||||
@@ -97,18 +84,6 @@ async function waitForHiddenTabPtyId(page: Page, tabId: string): Promise<string>
|
||||
return ptyId
|
||||
}
|
||||
|
||||
async function readHiddenOutputDebug(page: Page): Promise<HiddenOutputDebugSnapshot | null> {
|
||||
return page.evaluate(() => {
|
||||
return (window as CodexStartupBackgroundWindow).__terminalPtyOutputDebug?.snapshot() ?? null
|
||||
})
|
||||
}
|
||||
|
||||
async function resetHiddenOutputDebug(page: Page): Promise<void> {
|
||||
await page.evaluate(() => {
|
||||
;(window as CodexStartupBackgroundWindow).__terminalPtyOutputDebug?.reset()
|
||||
})
|
||||
}
|
||||
|
||||
async function mainSnapshotContains(page: Page, ptyId: string, text: string): Promise<boolean> {
|
||||
return page.evaluate(
|
||||
async ({ targetPtyId, expectedText }) => {
|
||||
@@ -266,7 +241,6 @@ test.describe('Codex hidden startup composer background', () => {
|
||||
|
||||
const marker = `CODEX_STARTUP_BG_${Date.now()}`
|
||||
const command = codexLikeStartupCommand(marker)
|
||||
await resetHiddenOutputDebug(orcaPage)
|
||||
const hiddenTabId = await orcaPage.evaluate(
|
||||
({ worktreeId, command, eventName }) => {
|
||||
const store = window.__store
|
||||
@@ -312,12 +286,10 @@ test.describe('Codex hidden startup composer background', () => {
|
||||
message: 'Hidden Codex startup background never reached the main buffer snapshot'
|
||||
})
|
||||
.toBe(true)
|
||||
await expect
|
||||
.poll(async () => (await readHiddenOutputDebug(orcaPage))?.hiddenRendererSkipCount ?? 0, {
|
||||
timeout: 10_000,
|
||||
message: 'Codex-marked hidden startup did not take the skipped renderer path'
|
||||
})
|
||||
.toBeGreaterThan(0)
|
||||
// Why: the renderer skip counter is dead under the Phase-4 main-side delivery
|
||||
// gate (#7214) — hidden bytes are dropped in main before reaching the renderer.
|
||||
// The main-buffer snapshot above proves the hidden output was handled; the
|
||||
// reveal restore below proves it repaints when the worktree first shows.
|
||||
|
||||
await switchToWorktree(orcaPage, secondWorktreeId)
|
||||
await expect
|
||||
|
||||
@@ -173,6 +173,20 @@ async function waitForTabParked(page: Page, tabId: string): Promise<number> {
|
||||
return Date.now() - parkWaitStartedAt
|
||||
}
|
||||
|
||||
// Why: #8262 exempts the single most-recently-hidden tab from cold-park to keep
|
||||
// the just-left view instantly warm, so a lone hidden tab never parks. Opening
|
||||
// one more tab hides the current view (which then holds that exemption) and
|
||||
// leaves the older `targetTabId` free to cold-park. Returns waitForTabParked's
|
||||
// elapsed-ms so callers keep their parking annotations.
|
||||
async function parkHiddenTabBehindDecoy(
|
||||
page: Page,
|
||||
worktreeId: string,
|
||||
targetTabId: string
|
||||
): Promise<number> {
|
||||
await createActiveTerminalTab(page, worktreeId)
|
||||
return waitForTabParked(page, targetTabId)
|
||||
}
|
||||
|
||||
async function activateTerminalTab(page: Page, tabId: string): Promise<void> {
|
||||
await page.evaluate((targetTabId) => {
|
||||
const store = window.__store
|
||||
@@ -306,15 +320,16 @@ test.describe('Terminal hidden view parking', () => {
|
||||
.toContain(finalMarker)
|
||||
|
||||
const tabBId = await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
const parkDetectedAfterMs = await waitForTabParked(orcaPage, tabAId)
|
||||
const parkDetectedAfterMs = await parkHiddenTabBehindDecoy(orcaPage, worktreeId, tabAId)
|
||||
const wiring = await readParkingWiring(orcaPage)
|
||||
testInfo.annotations.push({
|
||||
type: 'terminal-parking',
|
||||
description: `parkDelayMs=${wiring.parkDelayMs ?? PARKING_DELAY_MS} parkDetectedAfterMs=${parkDetectedAfterMs}`
|
||||
})
|
||||
|
||||
// Why: parking must be scoped to the hidden tab — the visible tab keeps
|
||||
// a live pane manager and xterm.
|
||||
// Why: parking must be scoped to the parked tab — tab B (hidden more
|
||||
// recently, so #8262 keeps it warm) still holds a live pane manager and
|
||||
// xterm while tab A tore down.
|
||||
const tabBState = await readTerminalTabViewState(orcaPage, tabBId)
|
||||
expect(tabBState.hasManager).toBe(true)
|
||||
expect(tabBState.paneCount).toBeGreaterThan(0)
|
||||
@@ -370,7 +385,7 @@ test.describe('Terminal hidden view parking', () => {
|
||||
const { worktreeId, tabAId, tabAPtyId } = setup
|
||||
|
||||
await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
await waitForTabParked(orcaPage, tabAId)
|
||||
await parkHiddenTabBehindDecoy(orcaPage, worktreeId, tabAId)
|
||||
|
||||
const runId = randomUUID()
|
||||
const parkedTitle = `Parked side effects ${runId}`
|
||||
@@ -443,8 +458,10 @@ test.describe('Terminal hidden view parking', () => {
|
||||
.toBe(tabBId)
|
||||
|
||||
// Why: tab A parking proves the machinery ran past the delay in this app
|
||||
// instance, so the tab C assertion below is not vacuously green.
|
||||
await waitForTabParked(orcaPage, tabAId)
|
||||
// instance, so the tab C assertion below is not vacuously green. A decoy
|
||||
// takes the #8262 last-active exemption (tab C is excluded, not a candidate)
|
||||
// so tab A is the one that cold-parks.
|
||||
await parkHiddenTabBehindDecoy(orcaPage, worktreeId, tabAId)
|
||||
await orcaPage.waitForTimeout(PARKING_DELAY_MS * 3)
|
||||
|
||||
// Premise guard: nothing consumed the pending startup while hidden.
|
||||
@@ -483,8 +500,11 @@ test.describe('Terminal hidden view parking', () => {
|
||||
.toContain(marker)
|
||||
|
||||
// Tab B stays visible whenever tab A is parked; toggling the active tab
|
||||
// between them is the deterministic hide/reveal driver.
|
||||
// between them is the deterministic hide/reveal driver. The decoy tab
|
||||
// absorbs the #8262 last-active exemption each cycle (hidden after B) so
|
||||
// tab A — not the just-hidden view — is the one that cold-parks.
|
||||
const tabBId = await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
const decoyTabId = await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
|
||||
// One park/reveal cycle to run the frame through the snapshot restore for a
|
||||
// baseline. Why not compare against the visible-before-park content: an
|
||||
@@ -495,6 +515,8 @@ test.describe('Terminal hidden view parking', () => {
|
||||
// both sides pass through identical machinery, so any later diff is drift.
|
||||
const runOneParkRevealCycle = async (cycle: number): Promise<string[]> => {
|
||||
await activateTerminalTab(orcaPage, tabBId)
|
||||
// Hide tab B behind the decoy so B (not A) holds the #8262 exemption.
|
||||
await activateTerminalTab(orcaPage, decoyTabId)
|
||||
await waitForTabParked(orcaPage, tabAId)
|
||||
await activateTerminalTab(orcaPage, tabAId)
|
||||
await waitForActiveTerminalManager(orcaPage, 30_000)
|
||||
|
||||
@@ -163,6 +163,12 @@ function median(values: number[]): number {
|
||||
return sorted[Math.floor(sorted.length / 2)] ?? 0
|
||||
}
|
||||
|
||||
function percentile(values: number[], fraction: number): number {
|
||||
const sorted = [...values].sort((a, b) => a - b)
|
||||
const index = Math.min(sorted.length - 1, Math.max(0, Math.ceil(fraction * sorted.length) - 1))
|
||||
return sorted[index] ?? 0
|
||||
}
|
||||
|
||||
type PhaseLatency = {
|
||||
label: string
|
||||
bufferRows: number
|
||||
@@ -281,9 +287,12 @@ test.describe('Terminal typing latency vs scrollback history size', () => {
|
||||
phase.medianMs,
|
||||
`${phase.label}: median latency regressed with history size`
|
||||
).toBeLessThan(MAX_MEDIAN_KEY_LATENCY_MS)
|
||||
// Why: the fill keeps the session dirty so a daemon checkpoint
|
||||
// serialization can land inside a single sample; guard the p90 so one
|
||||
// environment-dominated spike is not a failure while median stays strict.
|
||||
expect(
|
||||
phase.worstMs,
|
||||
`${phase.label}: worst latency regressed with history size`
|
||||
percentile(phase.samples, 0.9),
|
||||
`${phase.label}: sustained (p90) latency regressed with history size`
|
||||
).toBeLessThan(MAX_WORST_KEY_LATENCY_MS)
|
||||
}
|
||||
} finally {
|
||||
|
||||
@@ -221,6 +221,19 @@ async function waitForTabParked(page: Page, tabId: string): Promise<void> {
|
||||
.toBe(false)
|
||||
}
|
||||
|
||||
// Why: #8262 exempts the single most-recently-hidden tab from cold-park to keep
|
||||
// the just-left view warm, so a lone hidden tab never parks. Opening one more
|
||||
// tab hides the current view (which then holds that exemption) and leaves the
|
||||
// older `targetTabId` free to cold-park.
|
||||
async function parkHiddenTabBehindDecoy(
|
||||
page: Page,
|
||||
worktreeId: string,
|
||||
targetTabId: string
|
||||
): Promise<void> {
|
||||
await createActiveTerminalTab(page, worktreeId)
|
||||
await waitForTabParked(page, targetTabId)
|
||||
}
|
||||
|
||||
async function waitForTabRemounted(page: Page, tabId: string): Promise<void> {
|
||||
await expect
|
||||
.poll(async () => page.evaluate((id) => window.__paneManagers?.get(id) !== undefined, tabId), {
|
||||
@@ -298,7 +311,7 @@ test.describe('terminal pane close vs hidden/park lifecycle keeps layout consist
|
||||
const { worktreeId, tabId } = await setUpSplitTab(orcaPage)
|
||||
await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
await closeLastPaneOnTab(orcaPage, tabId)
|
||||
await waitForTabParked(orcaPage, tabId)
|
||||
await parkHiddenTabBehindDecoy(orcaPage, worktreeId, tabId)
|
||||
await activateTerminalTab(orcaPage, tabId)
|
||||
await waitForTabRemounted(orcaPage, tabId)
|
||||
await expectLayoutConsistent(orcaPage, tabId, 1, 'close-while-hidden-mounted')
|
||||
@@ -307,7 +320,7 @@ test.describe('terminal pane close vs hidden/park lifecycle keeps layout consist
|
||||
test('close immediately after reveal remount, before panes settle', async ({ orcaPage }) => {
|
||||
const { worktreeId, tabId } = await setUpSplitTab(orcaPage)
|
||||
await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
await waitForTabParked(orcaPage, tabId)
|
||||
await parkHiddenTabBehindDecoy(orcaPage, worktreeId, tabId)
|
||||
await activateTerminalTab(orcaPage, tabId)
|
||||
await waitForTabRemounted(orcaPage, tabId)
|
||||
// Close as soon as the manager exists — panes may still be attaching.
|
||||
@@ -326,7 +339,7 @@ test.describe('terminal pane close vs hidden/park lifecycle keeps layout consist
|
||||
await closeLastPaneOnTab(orcaPage, tabId)
|
||||
await expectLayoutConsistent(orcaPage, tabId, 1, 'pre-park close')
|
||||
await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
await waitForTabParked(orcaPage, tabId)
|
||||
await parkHiddenTabBehindDecoy(orcaPage, worktreeId, tabId)
|
||||
await activateTerminalTab(orcaPage, tabId)
|
||||
await waitForTabRemounted(orcaPage, tabId)
|
||||
await expectLayoutConsistent(orcaPage, tabId, 1, 'post-park-reveal')
|
||||
@@ -347,7 +360,7 @@ test.describe('terminal pane close vs hidden/park lifecycle keeps layout consist
|
||||
test('split pane shell exits while the tab is parked', async ({ orcaPage }) => {
|
||||
const { worktreeId, tabId, splitPtyId } = await setUpSplitTab(orcaPage)
|
||||
await createActiveTerminalTab(orcaPage, worktreeId)
|
||||
await waitForTabParked(orcaPage, tabId)
|
||||
await parkHiddenTabBehindDecoy(orcaPage, worktreeId, tabId)
|
||||
await sendToTerminal(orcaPage, splitPtyId, 'exit\r')
|
||||
await orcaPage.waitForTimeout(PARKING_DELAY_MS)
|
||||
await activateTerminalTab(orcaPage, tabId)
|
||||
|
||||
@@ -533,9 +533,10 @@ test.describe('Terminal Shortcuts', () => {
|
||||
await expect.poll(() => getKittyKeyboardFlags(orcaPage)).toBe(1)
|
||||
await pressAndExpectWrite(orcaPage, electronApp, 'Shift+Enter', '\x1b[13;2u')
|
||||
|
||||
// The shell is only standing in for a KKP-aware TUI and does not consume
|
||||
// the CSI-u input above, so cancel that synthetic line before its reset.
|
||||
await execInTerminal(orcaPage, ptyId, "\x03printf '\\033[<u'")
|
||||
// The shell is only standing in for a KKP-aware TUI and does not consume the
|
||||
// CSI-u input above. Send an idempotent "set flags to 0" reset rather than a
|
||||
// stack pop so it doesn't race the shell line editor still holding that input.
|
||||
await execInTerminal(orcaPage, ptyId, "\x03printf '\\033[=0u'")
|
||||
await expect.poll(() => getKittyKeyboardFlags(orcaPage)).toBe(0)
|
||||
await pressAndExpectWrite(orcaPage, electronApp, 'Shift+Enter', '\x1b\r')
|
||||
})
|
||||
|
||||
Reference in New Issue
Block a user