mixdog 0.9.54 → 0.9.55

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (49) hide show
  1. package/package.json +2 -1
  2. package/scripts/desktop-session-bridge-test.mjs +47 -0
  3. package/scripts/interrupted-turn-history-test.mjs +28 -0
  4. package/scripts/pending-completion-drop-test.mjs +160 -4
  5. package/scripts/pending-messages-lock-nonblocking-test.mjs +62 -0
  6. package/scripts/process-lifecycle-test.mjs +5 -2
  7. package/scripts/prompt-input-parity-test.mjs +145 -0
  8. package/scripts/streaming-tail-window-test.mjs +146 -0
  9. package/scripts/tui-runtime-load-bench-entry.jsx +608 -0
  10. package/scripts/tui-runtime-load-bench.mjs +45 -0
  11. package/scripts/tui-store-frame-batch-test.mjs +99 -0
  12. package/scripts/tui-transcript-perf-test.mjs +367 -2
  13. package/scripts/write-backpressure-test.mjs +147 -0
  14. package/src/cli.mjs +5 -5
  15. package/src/lib/keychain-cjs.cjs +114 -25
  16. package/src/runtime/agent/orchestrator/session/manager/ask-session.mjs +94 -18
  17. package/src/runtime/agent/orchestrator/session/manager/message-sanitize.mjs +20 -4
  18. package/src/runtime/agent/orchestrator/session/manager/pending-messages.mjs +316 -63
  19. package/src/runtime/agent/orchestrator/session/store.mjs +46 -1
  20. package/src/runtime/memory/lib/trace-store.mjs +66 -4
  21. package/src/runtime/shared/buffered-appender.mjs +83 -4
  22. package/src/runtime/shared/process-lifecycle.mjs +56 -12
  23. package/src/runtime/shared/process-shutdown.mjs +2 -1
  24. package/src/session-runtime/model-route-api.mjs +4 -1
  25. package/src/session-runtime/native-search.mjs +4 -2
  26. package/src/session-runtime/provider-auth-api.mjs +8 -1
  27. package/src/session-runtime/provider-models.mjs +8 -3
  28. package/src/session-runtime/resource-api.mjs +2 -1
  29. package/src/session-runtime/runtime-core.mjs +32 -0
  30. package/src/session-runtime/warmup-schedulers.mjs +5 -1
  31. package/src/standalone/agent-tool.mjs +19 -1
  32. package/src/tui/App.jsx +10 -4
  33. package/src/tui/app/text-layout.mjs +9 -1
  34. package/src/tui/app/transcript-window.mjs +146 -60
  35. package/src/tui/app/use-mouse-input.mjs +16 -8
  36. package/src/tui/app/use-transcript-scroll.mjs +71 -19
  37. package/src/tui/app/use-transcript-window.mjs +21 -10
  38. package/src/tui/components/PromptInput.jsx +13 -6
  39. package/src/tui/dist/index.mjs +1063 -232
  40. package/src/tui/engine/context-state.mjs +31 -31
  41. package/src/tui/engine/frame-batched-store.mjs +75 -0
  42. package/src/tui/engine/session-api-ext.mjs +6 -1
  43. package/src/tui/engine/session-api.mjs +9 -3
  44. package/src/tui/engine/session-flow.mjs +54 -27
  45. package/src/tui/engine/turn.mjs +13 -3
  46. package/src/tui/engine.mjs +515 -36
  47. package/src/tui/input-editing.mjs +33 -13
  48. package/src/tui/markdown/measure-rendered-rows.mjs +117 -5
  49. package/vendor/ink/build/ink.js +8 -16
@@ -0,0 +1,608 @@
1
+ import React from 'react';
2
+ import { Writable } from 'node:stream';
3
+ import { mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
4
+ import { tmpdir } from 'node:os';
5
+ import { dirname, join, resolve } from 'node:path';
6
+ import { performance } from 'node:perf_hooks';
7
+ import { fileURLToPath } from 'node:url';
8
+ import { Box, render } from 'ink';
9
+
10
+ import { Item } from '../src/tui/components/TranscriptItem.jsx';
11
+ import { useTranscriptWindow } from '../src/tui/app/use-transcript-window.mjs';
12
+ import { createTranscriptWriter } from '../src/runtime/shared/transcript-writer.mjs';
13
+ import {
14
+ drainPathSync,
15
+ getBufferedAppenderStats,
16
+ hasInFlightWrite,
17
+ } from '../src/runtime/shared/buffered-appender.mjs';
18
+
19
+ /**
20
+ * Headless TUI/runtime load-regression bench.
21
+ *
22
+ * Each measured frame covers producer state updates, React reconciliation,
23
+ * production transcript windowing, TranscriptItem rendering, Yoga layout,
24
+ * measured-height harvest/re-render, Ink serialization, and the final write
25
+ * callback on a capture TTY. Producer records go through the production
26
+ * transcript writer/appender and the same text is appended to rendered state.
27
+ *
28
+ * Gate precedence is explicit CLI/env override, then the checked-in baseline
29
+ * p95 times a fixed margin, then the 16.67 ms absolute fallback. Warm-up
30
+ * calibration is informational only: a render regression cannot raise its own
31
+ * threshold. Re-record the baseline intentionally with --update-baseline after
32
+ * a hardware/runner change, never as part of an ordinary regression run.
33
+ */
34
+ const DEFAULTS = Object.freeze({
35
+ items: 10_000,
36
+ frames: 120,
37
+ producers: 8,
38
+ burst: 12,
39
+ columns: 120,
40
+ viewportRows: 40,
41
+ thresholdFloorMs: 16.67,
42
+ calibrationFrames: 8,
43
+ baselineRegressionMargin: 1.5,
44
+ });
45
+ const PROJECT_ROOT = resolve(dirname(fileURLToPath(import.meta.url)), '..');
46
+ const BASELINE_PATH = join(PROJECT_ROOT, 'bench-results', 'tui-runtime-load-baseline.json');
47
+ const identity = (value) => value;
48
+ const noop = () => {};
49
+
50
+ function argValue(name, fallback) {
51
+ const index = process.argv.indexOf(name);
52
+ if (index >= 0 && index + 1 < process.argv.length) return process.argv[index + 1];
53
+ const prefix = `${name}=`;
54
+ const entry = process.argv.find((value) => value.startsWith(prefix));
55
+ return entry ? entry.slice(prefix.length) : fallback;
56
+ }
57
+
58
+ function positiveInt(name, fallback) {
59
+ const value = Number.parseInt(argValue(name, String(fallback)), 10);
60
+ if (!Number.isFinite(value) || value <= 0) throw new Error(`${name} must be a positive integer`);
61
+ return value;
62
+ }
63
+
64
+ function optionalPositiveNumber(name, envName) {
65
+ const cli = argValue(name, null);
66
+ const raw = cli ?? process.env[envName] ?? null;
67
+ if (raw == null || raw === '') return null;
68
+ const value = Number(raw);
69
+ if (!Number.isFinite(value) || value <= 0) {
70
+ throw new Error(`${name}/${envName} must be a positive number`);
71
+ }
72
+ return { value, source: cli != null ? name : envName };
73
+ }
74
+
75
+ function percentile(sorted, fraction) {
76
+ if (sorted.length === 0) return 0;
77
+ const index = Math.min(sorted.length - 1, Math.max(0, Math.ceil(sorted.length * fraction) - 1));
78
+ return sorted[index];
79
+ }
80
+
81
+ function round(value, digits = 3) {
82
+ const scale = 10 ** digits;
83
+ return Math.round(value * scale) / scale;
84
+ }
85
+
86
+ function readPinnedBaselineP95(path) {
87
+ try {
88
+ const parsed = JSON.parse(readFileSync(path, 'utf8'));
89
+ const value = Number(parsed?.frame_time?.p95_ms);
90
+ return Number.isFinite(value) && value > 0 ? value : null;
91
+ } catch {
92
+ return null;
93
+ }
94
+ }
95
+
96
+ function makeTranscriptItems(count) {
97
+ const items = new Array(count);
98
+ for (let index = 0; index < count; index += 1) {
99
+ const id = `history-${index}`;
100
+ switch (index % 10) {
101
+ case 0:
102
+ case 5:
103
+ items[index] = { id, kind: 'user', text: `Investigate session ${index} and preserve behavior.` };
104
+ break;
105
+ case 1:
106
+ case 4:
107
+ case 7:
108
+ items[index] = {
109
+ id,
110
+ kind: 'assistant',
111
+ text: `Completed transcript item ${index}.\n\n- checked runtime state\n- retained ordering`,
112
+ streaming: false,
113
+ };
114
+ break;
115
+ case 2:
116
+ case 6:
117
+ case 8:
118
+ items[index] = {
119
+ id,
120
+ kind: 'tool',
121
+ name: index % 2 ? 'agent' : 'shell',
122
+ args: JSON.stringify({ task_id: `task-${index}`, status: 'completed', path: `src/file-${index % 31}.mjs` }),
123
+ result: `result ${index}: ok`,
124
+ completedCount: 1,
125
+ count: 1,
126
+ };
127
+ break;
128
+ case 3:
129
+ items[index] = { id, kind: 'notice', tone: 'plain', text: `agent session ${index} completed` };
130
+ break;
131
+ default:
132
+ items[index] = { id, kind: 'turndone', elapsedMs: 100 + index };
133
+ break;
134
+ }
135
+ }
136
+ return items;
137
+ }
138
+
139
+ function makeProducer({ writer, lane, burst }) {
140
+ return async function produce(frame) {
141
+ await Promise.resolve();
142
+ const chunks = [];
143
+ let recordCount = 0;
144
+ for (let entry = 0; entry < burst; entry += 1) {
145
+ const text = `lane=${lane} frame=${frame} chunk=${entry} ${'stream '.repeat(6)}\n`;
146
+ writer.appendAssistant(text);
147
+ chunks.push(text);
148
+ recordCount += 1;
149
+ }
150
+ const stateItems = [{
151
+ id: `producer-${lane}-frame-${frame}`,
152
+ kind: 'assistant',
153
+ text: chunks.join(''),
154
+ streaming: false,
155
+ }];
156
+ if (frame % 6 === lane % 6) {
157
+ const command = `producer-${lane}-frame-${frame}`;
158
+ const result = `producer ${lane} frame ${frame} complete`;
159
+ writer.appendToolUse('shell', { command });
160
+ writer.appendToolResult({ stdout: result, exitCode: 0 });
161
+ recordCount += 2;
162
+ stateItems.push({
163
+ id: `producer-tool-${lane}-frame-${frame}`,
164
+ kind: 'tool',
165
+ name: 'shell',
166
+ args: JSON.stringify({ command }),
167
+ result,
168
+ count: 1,
169
+ completedCount: 1,
170
+ });
171
+ }
172
+ return { lane, chunks, recordCount, stateItems };
173
+ };
174
+ }
175
+
176
+ function captureTty(columns, rows) {
177
+ const stream = new Writable({
178
+ write(chunk, _encoding, callback) {
179
+ stream.capturedBytes += Buffer.byteLength(chunk);
180
+ stream.writeCalls += 1;
181
+ callback();
182
+ },
183
+ });
184
+ stream.columns = columns;
185
+ stream.rows = rows;
186
+ stream.isTTY = true;
187
+ stream.capturedBytes = 0;
188
+ stream.writeCalls = 0;
189
+ stream.getColorDepth = () => 1;
190
+ stream.hasColors = () => false;
191
+ stream.setRawMode = () => stream;
192
+ return stream;
193
+ }
194
+
195
+ function TranscriptHarness({ settledItems, streamingTail, structureRevision, columns, viewportRows }) {
196
+ const [scrollOffset, setScrollOffset] = React.useState(0);
197
+ const [measuredRowsVersion, setMeasuredRowsVersion] = React.useState(0);
198
+ const transcriptAnchorRef = React.useRef(null);
199
+ const transcriptAnchorDirtyRef = React.useRef(false);
200
+ const scrollTargetRef = React.useRef(0);
201
+ const scrollPositionRef = React.useRef(0);
202
+ const maxScrollRowsRef = React.useRef(0);
203
+ const transcriptGeomRef = React.useRef({});
204
+ const followingRef = React.useRef(true);
205
+ const dragRef = React.useRef({ active: false, rect: null });
206
+ const transcriptViewportRef = React.useRef({ top: 0 });
207
+ const selectionLayoutRef = React.useRef(null);
208
+
209
+ const {
210
+ transcriptWindow,
211
+ renderedTranscriptItems,
212
+ transcriptTailPinned,
213
+ transcriptMeasureRef,
214
+ } = useTranscriptWindow({
215
+ items: settledItems,
216
+ structureRevision,
217
+ streamingTail,
218
+ themeEpoch: 0,
219
+ frameColumns: columns,
220
+ toolOutputExpanded: false,
221
+ transcriptContentHeight: viewportRows,
222
+ transcriptBottomSlackRows: 1,
223
+ transcriptGuardRows: 1,
224
+ floatingPanelRows: 0,
225
+ overlayHintRequested: false,
226
+ scrollOffset,
227
+ setScrollOffset,
228
+ transcriptAnchorRef,
229
+ transcriptAnchorDirtyRef,
230
+ scrollTargetRef,
231
+ scrollPositionRef,
232
+ maxScrollRowsRef,
233
+ transcriptGeomRef,
234
+ followingRef,
235
+ dragRef,
236
+ transcriptViewportRef,
237
+ selectionLayoutRef,
238
+ withSelectionClip: identity,
239
+ paintSelectionRect: noop,
240
+ stopSmoothScroll: noop,
241
+ measuredRowsVersion,
242
+ setMeasuredRowsVersion,
243
+ });
244
+
245
+ return (
246
+ <Box flexDirection="column" width={columns} height={viewportRows} overflow="hidden" justifyContent="flex-end">
247
+ <Box
248
+ flexDirection="column"
249
+ width="100%"
250
+ flexShrink={0}
251
+ marginBottom={-transcriptWindow.effectiveScrollOffset}
252
+ >
253
+ {renderedTranscriptItems.map((item, index, all) => {
254
+ const measureRef = transcriptMeasureRef(item);
255
+ const itemNode = (
256
+ <Item
257
+ item={item}
258
+ prevKind={index > 0 ? all[index - 1].kind : null}
259
+ columns={columns}
260
+ toolOutputExpanded={false}
261
+ streamingWindowRows={transcriptTailPinned && item.id === streamingTail?.id
262
+ ? viewportRows + 4
263
+ : 0}
264
+ />
265
+ );
266
+ return measureRef ? (
267
+ <Box key={item.id} ref={measureRef} flexDirection="column" flexShrink={0}>
268
+ {itemNode}
269
+ </Box>
270
+ ) : <React.Fragment key={item.id}>{itemNode}</React.Fragment>;
271
+ })}
272
+ {transcriptWindow.bottomSpacerRows > 0
273
+ ? <Box height={transcriptWindow.bottomSpacerRows} flexShrink={0} />
274
+ : null}
275
+ </Box>
276
+ </Box>
277
+ );
278
+ }
279
+
280
+ async function flushFrame(instance) {
281
+ // The first wait covers the requested React/Ink commit and terminal callback;
282
+ // the second covers a measured-height harvest update scheduled by layout
283
+ // effects, matching the production two-commit correction path when it fires.
284
+ await instance.waitUntilRenderFlush();
285
+ await instance.waitUntilRenderFlush();
286
+ }
287
+
288
+ async function waitForAppender(paths, timeoutMs = 5_000) {
289
+ const deadline = performance.now() + timeoutMs;
290
+ for (;;) {
291
+ const stats = paths.map((path) => ({
292
+ path,
293
+ inFlight: hasInFlightWrite(path),
294
+ ...getBufferedAppenderStats(path),
295
+ }));
296
+ if (stats.every((entry) => !entry.inFlight && entry.bufferedBytes === 0)) {
297
+ return { settled: true, stats };
298
+ }
299
+ if (performance.now() >= deadline) return { settled: false, stats };
300
+ await new Promise((resolveWait) => setTimeout(resolveWait, 10));
301
+ }
302
+ }
303
+
304
+ function readPersistedRecords(paths) {
305
+ let records = 0;
306
+ let malformed = 0;
307
+ let bytes = 0;
308
+ for (const path of paths) {
309
+ const content = readFileSync(path, 'utf8');
310
+ bytes += Buffer.byteLength(content);
311
+ for (const line of content.split(/\r?\n/)) {
312
+ if (!line) continue;
313
+ try {
314
+ JSON.parse(line);
315
+ records += 1;
316
+ } catch {
317
+ malformed += 1;
318
+ }
319
+ }
320
+ }
321
+ return { records, malformed, bytes };
322
+ }
323
+
324
+ function usage() {
325
+ process.stdout.write(`tui-runtime-load-bench
326
+
327
+ Headless production React/Ink + transcript-writer stress bench. Gate precedence:
328
+ --threshold-ms / MIXDOG_TUI_LOAD_THRESHOLD_MS, then checked-in baseline p95 *
329
+ ${DEFAULTS.baselineRegressionMargin}, then ${DEFAULTS.thresholdFloorMs}ms. Warm-up calibration is informational
330
+ only. Run --update-baseline intentionally after changing hardware/runner; normal
331
+ runs never rewrite the checked-in baseline.
332
+
333
+ Options:
334
+ --items N settled transcript items (default ${DEFAULTS.items})
335
+ --frames N measured frames (default ${DEFAULTS.frames})
336
+ --producers N concurrent session-like writers (default ${DEFAULTS.producers})
337
+ --burst N assistant records per producer/frame (default ${DEFAULTS.burst})
338
+ --threshold-ms N explicit p95 gate (env MIXDOG_TUI_LOAD_THRESHOLD_MS)
339
+ --update-baseline replace bench-results/tui-runtime-load-baseline.json on PASS
340
+ --json print JSON only
341
+ --output PATH also save the JSON result
342
+ `);
343
+ }
344
+
345
+ if (process.argv.includes('--help') || process.argv.includes('-h')) {
346
+ usage();
347
+ process.exit(0);
348
+ }
349
+
350
+ const thresholdOverride = optionalPositiveNumber('--threshold-ms', 'MIXDOG_TUI_LOAD_THRESHOLD_MS');
351
+ const pinnedBaselineP95 = readPinnedBaselineP95(BASELINE_PATH);
352
+ const updateBaseline = process.argv.includes('--update-baseline');
353
+ const config = {
354
+ items: positiveInt('--items', DEFAULTS.items),
355
+ frames: positiveInt('--frames', DEFAULTS.frames),
356
+ producers: positiveInt('--producers', DEFAULTS.producers),
357
+ burst: positiveInt('--burst', DEFAULTS.burst),
358
+ columns: positiveInt('--columns', DEFAULTS.columns),
359
+ viewportRows: positiveInt('--viewport-rows', DEFAULTS.viewportRows),
360
+ thresholdFloorMs: DEFAULTS.thresholdFloorMs,
361
+ calibrationFrames: DEFAULTS.calibrationFrames,
362
+ baselineRegressionMargin: DEFAULTS.baselineRegressionMargin,
363
+ };
364
+ const jsonOnly = process.argv.includes('--json');
365
+ const outputPath = argValue('--output', updateBaseline ? BASELINE_PATH : null);
366
+ const root = mkdtempSync(join(tmpdir(), 'mixdog-tui-load-bench-'));
367
+ const mixdogHome = join(root, 'home');
368
+ const cwd = join(root, 'project');
369
+ mkdirSync(cwd, { recursive: true });
370
+
371
+ let result;
372
+ let exitCode = 1;
373
+ let instance = null;
374
+ let rssTimer = null;
375
+ const pathsToDrain = [];
376
+ const startedAt = performance.now();
377
+ try {
378
+ let settledItems = makeTranscriptItems(config.items);
379
+ const writers = Array.from({ length: config.producers }, (_, lane) => {
380
+ const writer = createTranscriptWriter({
381
+ mixdogHome,
382
+ sessionId: `bench-agent-${lane}`,
383
+ cwd,
384
+ pid: process.pid * 100 + lane,
385
+ });
386
+ writer.ensureTranscriptFile();
387
+ pathsToDrain.push(writer.transcriptPath);
388
+ return writer;
389
+ });
390
+ const producers = writers.map((writer, lane) => makeProducer({ writer, lane, burst: config.burst }));
391
+ const droppedBytesAtStart = getBufferedAppenderStats(pathsToDrain[0]).totalDroppedBytes;
392
+
393
+ let peakRss = process.memoryUsage().rss;
394
+ const sampleRss = () => {
395
+ peakRss = Math.max(peakRss, process.memoryUsage().rss);
396
+ };
397
+ rssTimer = setInterval(sampleRss, 5);
398
+ rssTimer.unref?.();
399
+
400
+ const stdout = captureTty(config.columns, config.viewportRows);
401
+ const stderr = captureTty(config.columns, config.viewportRows);
402
+ const stdin = captureTty(config.columns, config.viewportRows);
403
+ let inkRenderCalls = 0;
404
+ let streamingText = 'Synthetic live response\n';
405
+ let streamingTail = { id: 'bench-live-tail', kind: 'assistant', text: streamingText, streaming: true };
406
+ let structureRevision = 1;
407
+
408
+ instance = render(
409
+ <TranscriptHarness
410
+ settledItems={settledItems}
411
+ streamingTail={streamingTail}
412
+ structureRevision={structureRevision}
413
+ columns={config.columns}
414
+ viewportRows={config.viewportRows}
415
+ />,
416
+ {
417
+ stdout,
418
+ stderr,
419
+ stdin,
420
+ interactive: true,
421
+ patchConsole: false,
422
+ exitOnCtrlC: false,
423
+ incrementalRendering: true,
424
+ maxFps: 1000,
425
+ onRender: () => { inkRenderCalls += 1; },
426
+ },
427
+ );
428
+ await flushFrame(instance);
429
+
430
+ const calibrationTimes = [];
431
+ for (let index = 0; index < config.calibrationFrames; index += 1) {
432
+ const begin = performance.now();
433
+ streamingTail = { ...streamingTail, text: `${streamingText}calibration ${index}\n` };
434
+ instance.rerender(
435
+ <TranscriptHarness
436
+ settledItems={settledItems}
437
+ streamingTail={streamingTail}
438
+ structureRevision={structureRevision}
439
+ columns={config.columns}
440
+ viewportRows={config.viewportRows}
441
+ />,
442
+ );
443
+ await flushFrame(instance);
444
+ calibrationTimes.push(performance.now() - begin);
445
+ }
446
+ const rawCalibrationP95 = percentile([...calibrationTimes].sort((a, b) => a - b), 0.95);
447
+ const effectiveThresholdMs = thresholdOverride?.value
448
+ ?? (pinnedBaselineP95 != null
449
+ ? pinnedBaselineP95 * config.baselineRegressionMargin
450
+ : config.thresholdFloorMs);
451
+ const thresholdSource = thresholdOverride?.source
452
+ || (pinnedBaselineP95 != null
453
+ ? `checked-in baseline p95 * ${config.baselineRegressionMargin}`
454
+ : 'absolute fallback (checked-in baseline unavailable)');
455
+
456
+ const frameTimes = [];
457
+ let expectedRecords = 0;
458
+ let producerItemsRendered = 0;
459
+ let framesWithoutInkRender = 0;
460
+ let framesWithoutTerminalWrite = 0;
461
+ for (let frame = 0; frame < config.frames; frame += 1) {
462
+ const frameStarted = performance.now();
463
+ const priorRenderCalls = inkRenderCalls;
464
+ const priorTerminalBytes = stdout.capturedBytes;
465
+ const produced = await Promise.all(producers.map((produce) => produce(frame)));
466
+ const newItems = produced.flatMap((lane) => lane.stateItems);
467
+ const frameText = produced.flatMap((lane) => lane.chunks).join('');
468
+ expectedRecords += produced.reduce((sum, lane) => sum + lane.recordCount, 0);
469
+ producerItemsRendered += newItems.length;
470
+ settledItems = [...settledItems, ...newItems];
471
+ structureRevision += 1;
472
+ streamingText += frameText;
473
+ streamingTail = { ...streamingTail, text: streamingText };
474
+
475
+ instance.rerender(
476
+ <TranscriptHarness
477
+ settledItems={settledItems}
478
+ streamingTail={streamingTail}
479
+ structureRevision={structureRevision}
480
+ columns={config.columns}
481
+ viewportRows={config.viewportRows}
482
+ />,
483
+ );
484
+ await flushFrame(instance);
485
+ const elapsed = performance.now() - frameStarted;
486
+ frameTimes.push(elapsed);
487
+ if (inkRenderCalls === priorRenderCalls) framesWithoutInkRender += 1;
488
+ if (stdout.capturedBytes === priorTerminalBytes) framesWithoutTerminalWrite += 1;
489
+ sampleRss();
490
+ }
491
+
492
+ instance.unmount();
493
+ await instance.waitUntilExit();
494
+ instance.cleanup();
495
+ instance = null;
496
+
497
+ const appender = await waitForAppender(pathsToDrain);
498
+ const persisted = readPersistedRecords(pathsToDrain);
499
+ const endAppenderStats = getBufferedAppenderStats(pathsToDrain[0]);
500
+ const droppedBytes = endAppenderStats.totalDroppedBytes - droppedBytesAtStart;
501
+ const bufferedBytes = appender.stats.reduce((sum, entry) => sum + entry.bufferedBytes, 0);
502
+ sampleRss();
503
+
504
+ const sorted = [...frameTimes].sort((a, b) => a - b);
505
+ const rawP50 = percentile(sorted, 0.50);
506
+ const rawP95 = percentile(sorted, 0.95);
507
+ const rawMax = sorted.at(-1) || 0;
508
+ const stats = {
509
+ p50_ms: round(rawP50),
510
+ p95_ms: round(rawP95),
511
+ max_ms: round(rawMax),
512
+ over_threshold_frames: frameTimes.filter((ms) => ms > effectiveThresholdMs).length,
513
+ };
514
+ const failures = [];
515
+ // Compare the unrounded percentile; rounded values are reporting only.
516
+ if (rawP95 > effectiveThresholdMs) {
517
+ failures.push(`raw p95 ${rawP95}ms exceeds ${effectiveThresholdMs}ms threshold`);
518
+ }
519
+ if (!appender.settled) failures.push('buffered appender did not settle within 5s');
520
+ if (persisted.records !== expectedRecords) {
521
+ failures.push(`persisted record count ${persisted.records} != expected ${expectedRecords}`);
522
+ }
523
+ if (persisted.malformed !== 0) failures.push(`${persisted.malformed} malformed transcript records`);
524
+ if (droppedBytes !== 0) failures.push(`buffered appender dropped ${droppedBytes} bytes`);
525
+ if (bufferedBytes !== 0) failures.push(`buffered appender retained ${bufferedBytes} bytes`);
526
+ if (framesWithoutInkRender !== 0) failures.push(`${framesWithoutInkRender} frames skipped Ink rendering`);
527
+ if (framesWithoutTerminalWrite !== 0) failures.push(`${framesWithoutTerminalWrite} frames skipped terminal flush`);
528
+
529
+ result = {
530
+ bench: 'tui-runtime-load',
531
+ status: failures.length === 0 ? 'PASS' : 'FAIL',
532
+ threshold: {
533
+ metric: 'raw_p95_ms',
534
+ max_ms: round(effectiveThresholdMs),
535
+ source: thresholdSource,
536
+ host_dependent: true,
537
+ baseline_path: 'bench-results/tui-runtime-load-baseline.json',
538
+ baseline_p95_ms: pinnedBaselineP95,
539
+ baseline_regression_margin: config.baselineRegressionMargin,
540
+ calibration_p95_ms: round(rawCalibrationP95),
541
+ calibration_role: 'informational only',
542
+ },
543
+ config,
544
+ load: {
545
+ transcript_items: settledItems.length + 1,
546
+ producer_state_items: producerItemsRendered,
547
+ expected_records: expectedRecords,
548
+ persisted_records: persisted.records,
549
+ malformed_records: persisted.malformed,
550
+ persisted_bytes: persisted.bytes,
551
+ dropped_buffer_bytes: droppedBytes,
552
+ buffered_bytes_after_settle: bufferedBytes,
553
+ ink_render_calls: inkRenderCalls,
554
+ terminal_write_calls: stdout.writeCalls,
555
+ terminal_bytes: stdout.capturedBytes,
556
+ },
557
+ frame_time: stats,
558
+ peak_rss_mb: round(peakRss / (1024 * 1024), 1),
559
+ wall_ms: round(performance.now() - startedAt, 1),
560
+ runtime: { node: process.version, platform: process.platform, arch: process.arch },
561
+ failures,
562
+ };
563
+ exitCode = failures.length === 0 ? 0 : 1;
564
+ } catch (error) {
565
+ result = {
566
+ bench: 'tui-runtime-load',
567
+ status: 'FAIL',
568
+ config,
569
+ wall_ms: round(performance.now() - startedAt, 1),
570
+ failures: [error?.stack || String(error)],
571
+ };
572
+ exitCode = 1;
573
+ } finally {
574
+ if (rssTimer) clearInterval(rssTimer);
575
+ if (instance) {
576
+ try { instance.unmount(); } catch {}
577
+ try { instance.cleanup(); } catch {}
578
+ }
579
+ for (const path of pathsToDrain) {
580
+ try { drainPathSync(path); } catch {}
581
+ }
582
+ rmSync(root, { recursive: true, force: true });
583
+ }
584
+
585
+ const serialized = `${JSON.stringify(result, null, 2)}\n`;
586
+ if (outputPath && (!updateBaseline || exitCode === 0)) {
587
+ const destination = resolve(outputPath);
588
+ mkdirSync(dirname(destination), { recursive: true });
589
+ writeFileSync(destination, serialized);
590
+ } else if (updateBaseline && exitCode !== 0) {
591
+ process.stderr.write('FAIL: refusing to replace baseline with a failed run\n');
592
+ }
593
+ if (jsonOnly) {
594
+ process.stdout.write(serialized);
595
+ } else {
596
+ const frame = result.frame_time || {};
597
+ process.stdout.write(
598
+ `tui-runtime-load: ${result.status} items=${result.load?.transcript_items ?? config.items} `
599
+ + `frames=${config.frames} producers=${config.producers}\n`
600
+ + `frame ms: p50=${frame.p50_ms ?? '-'} p95=${frame.p95_ms ?? '-'} max=${frame.max_ms ?? '-'} `
601
+ + `(p95 threshold=${result.threshold?.max_ms ?? '-'}ms, over=${frame.over_threshold_frames ?? '-'})\n`
602
+ + `peak RSS=${result.peak_rss_mb ?? '-'} MB records=${result.load?.persisted_records ?? 0}/`
603
+ + `${result.load?.expected_records ?? 0} dropped=${result.load?.dropped_buffer_bytes ?? '-'} `
604
+ + `wall=${result.wall_ms}ms\n`,
605
+ );
606
+ for (const failure of result.failures || []) process.stderr.write(`FAIL: ${failure}\n`);
607
+ }
608
+ process.exitCode = exitCode;
@@ -0,0 +1,45 @@
1
+ #!/usr/bin/env node
2
+ /**
3
+ * Builds the JSX bench entry and runs it in-process. Keeping this public
4
+ * launcher as .mjs matches the other scripts/ benches while the entry can mount
5
+ * the production JSX transcript components without relying on the TUI bundle.
6
+ */
7
+ import { build } from 'esbuild';
8
+ import { rm } from 'node:fs/promises';
9
+ import { dirname, join } from 'node:path';
10
+ import { fileURLToPath, pathToFileURL } from 'node:url';
11
+
12
+ const ROOT = join(dirname(fileURLToPath(import.meta.url)), '..');
13
+ const entry = join(ROOT, 'scripts', 'tui-runtime-load-bench-entry.jsx');
14
+ const outfile = join(ROOT, 'scripts', '.tui-runtime-load-bench.tmp.mjs');
15
+
16
+ const inkAlias = {
17
+ name: 'mixdog-ink-alias',
18
+ setup(ctx) {
19
+ ctx.onResolve({ filter: /^ink$/ }, () => ({
20
+ path: '../vendor/ink/build/index.js',
21
+ external: true,
22
+ }));
23
+ },
24
+ };
25
+
26
+ try {
27
+ await build({
28
+ entryPoints: [entry],
29
+ outfile,
30
+ bundle: true,
31
+ format: 'esm',
32
+ platform: 'node',
33
+ target: 'node22',
34
+ jsx: 'automatic',
35
+ packages: 'external',
36
+ plugins: [inkAlias],
37
+ banner: {
38
+ js: "import { createRequire as __mixdogCreateRequire } from 'node:module';\nconst require = __mixdogCreateRequire(import.meta.url);",
39
+ },
40
+ logLevel: 'silent',
41
+ });
42
+ await import(`${pathToFileURL(outfile).href}?run=${Date.now()}`);
43
+ } finally {
44
+ await rm(outfile, { force: true });
45
+ }