test: tighten relay assertions and repair benchmark checks (#33573)

Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>

config/scripts/ssh-watch-fanout-benchmark.mjs

@@ -1,10 +1,11 @@
 #!/usr/bin/env node
-// Compare repeated per-registration/event NFC normalization with pre-normalized root matchers/event paths for SSH watcher batches.
+// Model repeated per-root/event normalization versus hoisted root matchers/event paths for SSH watcher batches.
 // Assert identical routed results before timing and import production normalizer so folding drift cannot invalidate measurement.
+import assert from 'node:assert/strict'
 import { execFileSync } from 'node:child_process'
-import { readFileSync } from 'node:fs'
 import { performance } from 'node:perf_hooks'
 import { fileURLToPath } from 'node:url'
+import { createJiti } from 'jiti'
 
 const REPO_ROOT = fileURLToPath(new URL('../..', import.meta.url))
 const ITERATIONS = Number(process.env.ORCA_SSH_WATCH_BENCH_ITERATIONS ?? '200')
@@ -19,25 +20,15 @@ for (const [name, value] of [
   }
 }
 
-// Recheck production normalization source: if NFC fold/cost changes, benchmark’s assumption no longer holds.
-const PATH_SOURCE = readFileSync(
-  new URL('../../src/shared/cross-platform-path.ts', import.meta.url),
-  'utf8'
-)
-for (const marker of ['normalize(', 'createNormalizedPathInsideOrEqualMatcher']) {
-  if (!PATH_SOURCE.includes(marker)) {
-    throw new Error(`cross-platform-path.ts no longer contains ${marker}; this benchmark is stale`)
-  }
-}
-
-// Import the real normalizer so both arms fold paths exactly as production does.
+// Jiti resolves the production graph’s extensionless TypeScript imports.
+const jiti = createJiti(import.meta.url)
 const {
   normalizeRuntimePathForComparison,
   isPathInsideOrEqual,
   createNormalizedPathInsideOrEqualMatcher
-} = await import(new URL('../../src/shared/cross-platform-path.ts', import.meta.url).href)
+} = await jiti.import('../../src/shared/cross-platform-path.ts')
 
-// Pre-fix: mirrors the original routeSshFilesystemWatchNotification inner loop.
+// Historical per-pair normalization model.
 function routeBefore(roots, events, sink) {
   for (const rootPath of roots) {
     const matching = events.filter((event) => isPathInsideOrEqual(rootPath, event.absolutePath))
@@ -47,7 +38,7 @@ function routeBefore(roots, events, sink) {
   }
 }
 
-// Post-fix: mirrors the current implementation.
+// Synthetic hoisted-normalization model; this does not invoke the application router.
 function routeAfter(roots, events, sink) {
   const normalizedEvents = events.map((event) => ({
     event,
@@ -65,13 +56,14 @@ function routeAfter(roots, events, sink) {
 }
 
 // Use real paths: length/segments determine normalization cost, which short synthetic paths understate.
-const REPO_PATHS = execFileSync('git', ['ls-files'], {
+const REPO_PATHS = execFileSync('git', ['ls-files', '-z'], {
   cwd: REPO_ROOT,
   maxBuffer: 256 * 1024 * 1024
 })
   .toString()
-  .split('\n')
+  .split('\0')
   .filter(Boolean)
+assert.ok(REPO_PATHS.length > 0, 'Benchmark needs tracked paths')
 
 // Model multi-worktree host with explorer and base-directory watchers per root.
 function makeRoots(count) {
@@ -93,10 +85,8 @@ function makeEvents(roots, count) {
 
 function collect(roots, events, route) {
   const seen = []
-  route(roots, events, (rootPath, matching) =>
-    seen.push(`${rootPath} ${matching.map((event) => event.absolutePath).join(',')}`)
-  )
-  return seen.join('\n')
+  route(roots, events, (rootPath, matching) => seen.push([rootPath, matching]))
+  return seen
 }
 
 // Interleave rounds and compare medians to share CPU/load drift; batch ordering alone varied observed win from 6.7x to 23.3x.
@@ -145,12 +135,11 @@ for (const [rootCount, eventCount] of [
   const events = makeEvents(roots, eventCount)
   const before = collect(roots, events, routeBefore)
   const after = collect(roots, events, routeAfter)
-  if (before !== after) {
-    throw new Error(`routing differs at ${rootCount} roots x ${eventCount} events`)
-  }
-  if (!before.includes(' ')) {
-    throw new Error(`fixture routed nothing at ${rootCount} roots x ${eventCount} events`)
-  }
+  assert.deepEqual(after, before, `routing differs at ${rootCount} roots x ${eventCount} events`)
+  assert.ok(
+    before.length > 0,
+    `fixture routed nothing at ${rootCount} roots x ${eventCount} events`
+  )
   const { beforeMs, afterMs } = measureInterleaved(roots, events)
   console.log(
     `${pad(rootCount, 6)} ${pad(eventCount, 7)} ${pad(`${beforeMs.toFixed(3)} ms`, 11)} ${pad(`${afterMs.toFixed(3)} ms`, 11)} ${pad(`${(beforeMs / afterMs).toFixed(1)}x`, 9)}`
@@ -158,5 +147,5 @@ for (const [rootCount, eventCount] of [
 }
 
 console.log(
-  '\nThis times routing only. The saving scales with roots x events, so it is small\nfor a single-worktree session and largest during a remote checkout storm, which\nis exactly when the main process is already busy.'
+  '\nThis times synthetic routing models with production path normalization.\nIt excludes application dispatch, relay transport, and filesystem work.'
 )

config/scripts/terminal-reattach-payload-scan-benchmark.mjs

@@ -1,9 +1,21 @@
 #!/usr/bin/env node
 // Compare replay signal’s old CSI scan with shipped 256KB-tail/shared-regex path; assert equivalence before timing deterministic seeded payloads.
 // Also measure Kitty-mode scan unchanged and with slower includes prefilter; run node config/scripts/terminal-reattach-payload-scan-benchmark.mjs.
+import assert from 'node:assert/strict'
 import { performance } from 'node:perf_hooks'
-import { CSI_SEQUENCE_PATTERN } from '../../src/shared/ansi-escape-sequences.ts'
-import { TerminalKittyKeyboardModeTracker } from '../../src/shared/terminal-kitty-keyboard-mode-tracker.ts'
+import { fileURLToPath } from 'node:url'
+import { createJiti } from 'jiti'
+
+const jiti = createJiti(import.meta.url, {
+  alias: { '@': fileURLToPath(new URL('../../src/renderer/src', import.meta.url)) }
+})
+const { CSI_SEQUENCE_PATTERN } = await jiti.import('../../src/shared/ansi-escape-sequences.ts')
+const { TerminalKittyKeyboardModeTracker } = await jiti.import(
+  '../../src/shared/terminal-kitty-keyboard-mode-tracker.ts'
+)
+const { hasCursorAgentReattachPayloadScreenSignal: signalSuffixRegex } = await jiti.import(
+  '../../src/renderer/src/components/terminal-pane/pty-connection/cursor-agent-reattach-screen.ts'
+)
 
 const ROUNDS = Number(process.env.ORCA_REATTACH_SCAN_BENCH_ROUNDS ?? '7')
 const MIN_ITERATION_MS = 120
@@ -86,7 +98,7 @@ function buildPayloads() {
   ]
 }
 
-// Baseline: verbatim copy of pty-connection.ts:448 (not exported).
+// Historical full-payload CSI character scan for comparison.
 function stripBaseline(data) {
   let normalized = ''
   let index = 0
@@ -112,13 +124,9 @@ const stripRegex = (data) => data.replace(CSI_SEQUENCE_PATTERN, '')
 
 const HEADER = 'Cursor Agent'
 const TAIL_CHARS = 5000
-// Mirrors CURSOR_AGENT_REATTACH_SCAN_TAIL_LIMIT_CHARS in pty-connection.ts.
-const SUFFIX_RAW_CHARS = 256 * 1024
 
 const signalBaseline = (data) => hasSignal(stripBaseline(data))
 const signalRegex = (data) => hasSignal(stripRegex(data))
-const signalSuffixRegex = (data) =>
-  hasSignal(stripRegex(data.length > SUFFIX_RAW_CHARS ? data.slice(-SUFFIX_RAW_CHARS) : data))
 
 function hasSignal(normalized) {
   const headerIndex = normalized.lastIndexOf(HEADER)
@@ -197,12 +205,13 @@ const kittyScan = (data) => {
 }
 // The includes() introducer pre-filter a daemon-side mode scan would use.
 const kittyGated = (data) => {
-  if (!data.includes('\x1b[?') && !data.includes('\x1bc') && !data.includes('\x9b')) {
+  if (!data.includes('\x1b[') && !data.includes('\x1bc') && !data.includes('\x9b')) {
     return 0
   }
   return kittyScan(data)
 }
 for (const { name, data } of payloads) {
+  assert.equal(kittyGated(data), kittyScan(data), `Kitty gate disagrees on ${name}`)
   console.log(
     `${pad(name, 36)} ${pad(kb(data), 8)} ${pad(measure(kittyScan, data).toFixed(3), 11)} ${pad(measure(kittyGated, data).toFixed(3), 15)}`
   )
@@ -212,19 +221,22 @@ for (const { name, data } of payloads) {
 const liveChunk = buildColoredScrollback(4 * 1024, 5).replaceAll('\x1b', '')
 const introducerRows = [
   ['4KB live chunk, no escapes', liveChunk],
-  ['4KB live chunk, \\x1b[?25l present', `${liveChunk}\x1b[?25l`]
+  ['4KB live chunk, \\x1b[?25l present', `${liveChunk}\x1b[?25l`],
+  ['4KB live chunk, Kitty flags 3', `${liveChunk}\x1b[>3u`],
+  ['4KB live chunk, Kitty flags 7', `${liveChunk}\x1b[=7u`]
 ]
 console.log(
   `\n${pad('payload', 36)} ${pad('size', 8)} ${pad('scanReplay', 11)} ${pad('+includes gate', 15)}`
 )
 for (const [name, data] of introducerRows) {
+  assert.equal(kittyGated(data), kittyScan(data), `Kitty gate disagrees on ${name}`)
   console.log(
     `${pad(name, 36)} ${pad(kb(data), 8)} ${pad(measure(kittyScan, data).toFixed(3), 11)} ${pad(measure(kittyGated, data).toFixed(3), 15)}`
   )
 }
 
 console.log(
-  '\nSnapshot/replay payloads always contain \\x1b[?, so the includes() gate never fires on\nthe reattach path — it only helps the live per-chunk path.'
+  '\nThese replay fixtures contain CSI escapes, so the includes() gate never skips\ntheir scan. The live plain-output row measures the no-escape fast path.'
 )
 
 // Time xterm on identical bytes as reference cost for deciding whether scan optimization matters.

src/shared/mobile-relay-phone-protocol.test.ts

@@ -11,6 +11,13 @@ describe('relay phone outer protocol', () => {
         leaseExpiresAt: 1
       }).success
     ).toBe(true)
+    expect(
+      RelayPhoneHelloSchema.safeParse({
+        type: 'relay-hello',
+        ok: false,
+        code: 4404
+      }).success
+    ).toBe(true)
     expect(
       RelayPhoneHelloSchema.safeParse({
         type: 'relay-hello',

src/shared/network/server-share-address.test.ts

@@ -5,8 +5,14 @@ import { parseServerShareAddress } from './server-share-address'
 describe('parseServerShareAddress', () => {
   it('accepts a bare hostname or IP', () => {
     expect(parseServerShareAddress('my-host')).toEqual({ ok: true, value: 'my-host' })
-    expect(parseServerShareAddress('my-mac.tail-abcd.ts.net').ok).toBe(true)
-    expect(parseServerShareAddress('192.168.1.50').ok).toBe(true)
+    expect(parseServerShareAddress('my-mac.tail-abcd.ts.net')).toEqual({
+      ok: true,
+      value: 'my-mac.tail-abcd.ts.net'
+    })
+    expect(parseServerShareAddress('192.168.1.50')).toEqual({
+      ok: true,
+      value: '192.168.1.50'
+    })
   })
 
   it('rejects empty, whitespace-containing, and malformed input', () => {

src/shared/tab-has-live-pty.test.ts

@@ -2,10 +2,6 @@ import { describe, expect, it } from 'vitest'
 import { tabHasLivePty } from './tab-has-live-pty'
 
 describe('tabHasLivePty', () => {
-  it('returns false for an empty map', () => {
-    expect(tabHasLivePty({}, 'tab-1')).toBe(false)
-  })
-
   it('returns false for a missing tab id', () => {
     expect(tabHasLivePty({ 'other-tab': ['pty-1'] }, 'tab-1')).toBe(false)
   })