The best way to control Claude Code, Codex and more from any computer or phone
Mirrored from https://github.com/Camo67/t3code.git
Synced Nov 21, 04:45 UTC
| 1 | // @effect-diagnostics nodeBuiltinImport:off - resume coverage writes, appends |
| 2 | // to, and truncates real transcript files byte-exactly, mirroring the reader's |
| 3 | // own deliberate node:fs usage. |
| 4 | import * as NodeFSP from "node:fs/promises"; |
| 5 | import * as NodeOS from "node:os"; |
| 6 | import * as NodePath from "node:path"; |
| 7 | import * as NodeZlib from "node:zlib"; |
| 8 | |
| 9 | import { afterEach, assert, beforeEach, describe, it } from "@effect/vitest"; |
| 10 | import { totalTokens, type TranscriptUsageFormat } from "@t3tools/provider-core/server/usage"; |
| 11 | import * as Schema from "effect/Schema"; |
| 12 | |
| 13 | import { TEST_FORMATS } from "./usageTestFormats.ts"; |
| 14 | import { |
| 15 | listTranscriptFiles, |
| 16 | physicalSourceIdOf, |
| 17 | readPhysicalSourceId, |
| 18 | readTranscriptRecords, |
| 19 | transcriptLineGate, |
| 20 | } from "./usageTranscriptReader.ts"; |
| 21 | |
| 22 | let dir: string; |
| 23 | |
| 24 | beforeEach(async () => { |
| 25 | dir = await NodeFSP.mkdtemp(NodePath.join(NodeOS.tmpdir(), "usage-reader-test-")); |
| 26 | }); |
| 27 | |
| 28 | afterEach(async () => { |
| 29 | await NodeFSP.rm(dir, { recursive: true, force: true }); |
| 30 | }); |
| 31 | |
| 32 | function claudeLine(id: number, outputTokens: number): string { |
| 33 | return `${JSON.stringify({ |
| 34 | type: "assistant", |
| 35 | timestamp: "2026-08-01T10:00:00Z", |
| 36 | requestId: `req_${id}`, |
| 37 | sessionId: "session-1", |
| 38 | message: { |
| 39 | id: `msg_${id}`, |
| 40 | model: "claude-fable-5", |
| 41 | usage: { input_tokens: 10, output_tokens: outputTokens }, |
| 42 | }, |
| 43 | })}\n`; |
| 44 | } |
| 45 | |
| 46 | function codexMetaLine(): string { |
| 47 | return `${JSON.stringify({ |
| 48 | type: "session_meta", |
| 49 | timestamp: "2026-08-01T10:00:00Z", |
| 50 | payload: { type: "session_meta", id: "codex-session-1" }, |
| 51 | })}\n`; |
| 52 | } |
| 53 | |
| 54 | function codexModelLine(model: string): string { |
| 55 | return `${JSON.stringify({ |
| 56 | type: "turn_context", |
| 57 | timestamp: "2026-08-01T10:00:01Z", |
| 58 | payload: { type: "turn_context", model }, |
| 59 | })}\n`; |
| 60 | } |
| 61 | |
| 62 | function codexTierLine(serviceTier: string): string { |
| 63 | return `${JSON.stringify({ |
| 64 | type: "event_msg", |
| 65 | timestamp: "2026-08-01T10:00:01Z", |
| 66 | payload: { type: "thread_settings_applied", thread_settings: { service_tier: serviceTier } }, |
| 67 | })}\n`; |
| 68 | } |
| 69 | |
| 70 | function codexUsageLine(outputTokens: number, secondsOffset: number): string { |
| 71 | return `${JSON.stringify({ |
| 72 | type: "event_msg", |
| 73 | timestamp: `2026-08-01T10:00:${String(secondsOffset).padStart(2, "0")}Z`, |
| 74 | payload: { |
| 75 | type: "token_count", |
| 76 | info: { last_token_usage: { input_tokens: 100, output_tokens: outputTokens } }, |
| 77 | }, |
| 78 | })}\n`; |
| 79 | } |
| 80 | |
| 81 | describe("compressed transcripts", () => { |
| 82 | it("discovers nested plain and zstd JSONL files within the requested window", async () => { |
| 83 | const nested = NodePath.join(dir, "nested"); |
| 84 | await NodeFSP.mkdir(nested); |
| 85 | const plain = NodePath.join(dir, "active.jsonl"); |
| 86 | const compressed = NodePath.join(nested, "archived.jsonl.zst"); |
| 87 | const old = NodePath.join(nested, "old.jsonl.zst"); |
| 88 | await NodeFSP.writeFile(plain, claudeLine(1, 5)); |
| 89 | await NodeFSP.writeFile(compressed, NodeZlib.zstdCompressSync(claudeLine(2, 7))); |
| 90 | await NodeFSP.writeFile(old, NodeZlib.zstdCompressSync(claudeLine(3, 11))); |
| 91 | await NodeFSP.writeFile(NodePath.join(nested, "unrelated.zst"), "ignored"); |
| 92 | await NodeFSP.writeFile(NodePath.join(nested, "partial.jsonl.zst.tmp"), "ignored"); |
| 93 | await NodeFSP.utimes(old, 1, 1); |
| 94 | |
| 95 | const listed = await listTranscriptFiles(dir, Date.parse("2026-01-01T00:00:00Z")); |
| 96 | assert.deepStrictEqual( |
| 97 | listed.files.map((file) => file.path).sort(), |
| 98 | [plain, compressed].sort(), |
| 99 | ); |
| 100 | assert.strictEqual(listed.failedPaths, 0); |
| 101 | assert.isAbove(listed.files.find((file) => file.path === compressed)?.size ?? 0, 0); |
| 102 | }); |
| 103 | |
| 104 | it("lists files in directory order when more directories wait than it reads at once", async () => { |
| 105 | for (let index = 0; index < 40; index += 1) { |
| 106 | const nested = NodePath.join(dir, `project-${index}`, "sessions"); |
| 107 | await NodeFSP.mkdir(nested, { recursive: true }); |
| 108 | await NodeFSP.writeFile(NodePath.join(nested, "a.jsonl"), ""); |
| 109 | await NodeFSP.writeFile(NodePath.join(nested, "..", "b.jsonl"), ""); |
| 110 | } |
| 111 | // Usage that two transcripts share counts in the first one listed, so the order is part of the result. |
| 112 | const inOrder = async (folder: string): Promise<string[]> => { |
| 113 | const paths: string[] = []; |
| 114 | for (const entry of await NodeFSP.readdir(folder, { withFileTypes: true })) { |
| 115 | const child = NodePath.join(folder, entry.name); |
| 116 | if (entry.isDirectory()) paths.push(...(await inOrder(child))); |
| 117 | else paths.push(child); |
| 118 | } |
| 119 | return paths; |
| 120 | }; |
| 121 | const listed = await listTranscriptFiles(dir, 0); |
| 122 | assert.deepStrictEqual( |
| 123 | listed.files.map((file) => file.path), |
| 124 | await inOrder(dir), |
| 125 | ); |
| 126 | }); |
| 127 | |
| 128 | it.each([ |
| 129 | { parser: "native", streamingThresholdBytes: undefined }, |
| 130 | { parser: "projected", streamingThresholdBytes: 1024 }, |
| 131 | ])( |
| 132 | "matches plain Codex records and unfinished tails through long lines ($parser)", |
| 133 | async ({ streamingThresholdBytes }) => { |
| 134 | const plain = NodePath.join(dir, "rollout.jsonl"); |
| 135 | const compressed = plain + ".zst"; |
| 136 | const large = codexUsageLine(9, 5).replace( |
| 137 | '"payload":', |
| 138 | '"padding":' + JSON.stringify("日本語 é".repeat(128 * 1024)) + ',"payload":', |
| 139 | ); |
| 140 | const content = |
| 141 | codexMetaLine() + |
| 142 | codexModelLine("gpt-6-astra") + |
| 143 | codexTierLine("ultrafast") + |
| 144 | large + |
| 145 | codexUsageLine(13, 6).replace('"output_tokens":13', '"output_tokens":"invalid"') + |
| 146 | codexUsageLine(21, 8).trimEnd(); |
| 147 | await NodeFSP.writeFile(plain, content); |
| 148 | await NodeFSP.writeFile(compressed, NodeZlib.zstdCompressSync(content)); |
| 149 | |
| 150 | const options = |
| 151 | streamingThresholdBytes === undefined ? undefined : { streamingThresholdBytes }; |
| 152 | const expected = await readTranscriptRecords(plain, TEST_FORMATS.codex, undefined, options); |
| 153 | const actual = await readTranscriptRecords( |
| 154 | compressed, |
| 155 | TEST_FORMATS.codex, |
| 156 | undefined, |
| 157 | options, |
| 158 | ); |
| 159 | assert.isNotNull(expected); |
| 160 | assert.isNotNull(actual); |
| 161 | assert.strictEqual(expected.records.length, 1); |
| 162 | assert.strictEqual(expected.tailRecords.length, 1); |
| 163 | assert.deepStrictEqual(actual.records, expected.records); |
| 164 | assert.deepStrictEqual(actual.tailRecords, expected.tailRecords); |
| 165 | assert.strictEqual(expected.malformedRecords, 1); |
| 166 | assert.strictEqual(actual.malformedRecords, expected.malformedRecords); |
| 167 | assert.strictEqual(actual.records[0]?.speed, "ultrafast"); |
| 168 | assert.isFalse(actual.resumed); |
| 169 | assert.strictEqual(actual.position.resumeOffset, 0); |
| 170 | }, |
| 171 | ); |
| 172 | |
| 173 | it("ignores a prior resume position and rereads compressed rollouts from their start", async () => { |
| 174 | const plain = NodePath.join(dir, "rollout.jsonl"); |
| 175 | const content = codexMetaLine() + codexModelLine("gpt-6-astra") + codexUsageLine(9, 5); |
| 176 | await NodeFSP.writeFile(plain, content); |
| 177 | const first = await readTranscriptRecords(plain, TEST_FORMATS.codex); |
| 178 | assert.isNotNull(first); |
| 179 | |
| 180 | const compressed = plain + ".zst"; |
| 181 | await NodeFSP.writeFile(compressed, NodeZlib.zstdCompressSync(content + codexUsageLine(21, 8))); |
| 182 | const actual = await readTranscriptRecords(compressed, TEST_FORMATS.codex, first.position); |
| 183 | assert.isNotNull(actual); |
| 184 | assert.isFalse(actual.resumed); |
| 185 | assert.deepStrictEqual( |
| 186 | actual.records.map((record) => record.totals.outputTokens), |
| 187 | [9, 21], |
| 188 | ); |
| 189 | const repeated = await readTranscriptRecords(compressed, TEST_FORMATS.codex, actual.position); |
| 190 | assert.isNotNull(repeated); |
| 191 | assert.isFalse(repeated.resumed); |
| 192 | assert.deepStrictEqual(repeated.records, actual.records); |
| 193 | }); |
| 194 | |
| 195 | it("returns null for zstd decoder errors rather than caching an empty parse", async () => { |
| 196 | const path = NodePath.join(dir, "corrupt.jsonl.zst"); |
| 197 | await NodeFSP.writeFile(path, "not a zstd frame"); |
| 198 | assert.isNull(await readTranscriptRecords(path, TEST_FORMATS.claude)); |
| 199 | }); |
| 200 | }); |
| 201 | |
| 202 | describe("readTranscriptRecords resume", () => { |
| 203 | it.each([ |
| 204 | { mode: "buffered", streamingThresholdBytes: Infinity }, |
| 205 | { mode: "streaming", streamingThresholdBytes: 0 }, |
| 206 | ])("counts malformed tails once across $mode reads and resume", async (options) => { |
| 207 | const format: TranscriptUsageFormat<{ errors: number }> = { |
| 208 | ...TEST_FORMATS.claude, |
| 209 | selectFields: { ...TEST_FORMATS.claude.selectFields, usage: true }, |
| 210 | state: { |
| 211 | initial: () => ({ errors: 0 }), |
| 212 | schema: Schema.Struct({ errors: Schema.Number }), |
| 213 | }, |
| 214 | parseLine(line, state) { |
| 215 | return this.parseProjected(JSON.parse(line), state); |
| 216 | }, |
| 217 | parseProjected(value, state) { |
| 218 | if ( |
| 219 | typeof value === "object" && |
| 220 | value !== null && |
| 221 | "usage" in value && |
| 222 | value.usage === "malformed" |
| 223 | ) { |
| 224 | state.errors += 1; |
| 225 | return []; |
| 226 | } |
| 227 | return TEST_FORMATS.claude.parseProjected(value, undefined); |
| 228 | }, |
| 229 | readMalformedRecords: (state: { errors: number }) => state.errors, |
| 230 | }; |
| 231 | const path = NodePath.join(dir, "diagnostics.jsonl"); |
| 232 | const malformed = JSON.stringify({ usage: "malformed" }); |
| 233 | // Completed lines enter the projected parser only when they cross a stream chunk. |
| 234 | const committed = `${JSON.stringify({ usage: "malformed", padding: "x".repeat(256 * 1024) })}\n${claudeLine(1, 5)}`; |
| 235 | await NodeFSP.writeFile(path, committed + malformed); |
| 236 | |
| 237 | const first = await readTranscriptRecords(path, format, undefined, options); |
| 238 | assert.isNotNull(first); |
| 239 | assert.deepStrictEqual( |
| 240 | first.records.map((record) => record.totals.outputTokens), |
| 241 | [5], |
| 242 | ); |
| 243 | assert.deepStrictEqual(first.tailRecords, []); |
| 244 | assert.strictEqual(first.position.resumeOffset, Buffer.byteLength(committed)); |
| 245 | assert.deepStrictEqual(first.position.state, { errors: 1 }); |
| 246 | assert.strictEqual(first.malformedRecords, 2); |
| 247 | |
| 248 | const savedPosition = JSON.parse(JSON.stringify(first.position)); |
| 249 | await NodeFSP.appendFile(path, `\n${claudeLine(2, 7)}${malformed}\n`); |
| 250 | const resumed = await readTranscriptRecords(path, format, savedPosition, options); |
| 251 | assert.isNotNull(resumed); |
| 252 | assert.isTrue(resumed.resumed); |
| 253 | assert.deepStrictEqual( |
| 254 | resumed.records.map((record) => record.totals.outputTokens), |
| 255 | [7], |
| 256 | ); |
| 257 | assert.deepStrictEqual(resumed.position.state, { errors: 3 }); |
| 258 | assert.strictEqual(resumed.malformedRecords, 3); |
| 259 | |
| 260 | const full = await readTranscriptRecords(path, format, undefined, options); |
| 261 | assert.isNotNull(full); |
| 262 | assert.deepStrictEqual([...first.records, ...resumed.records], full.records); |
| 263 | assert.deepStrictEqual(resumed.tailRecords, full.tailRecords); |
| 264 | assert.deepStrictEqual(resumed.position, full.position); |
| 265 | assert.strictEqual(resumed.malformedRecords, full.malformedRecords); |
| 266 | }); |
| 267 | |
| 268 | it("parses only appended lines when resuming a grown file", async () => { |
| 269 | const path = NodePath.join(dir, "claude.jsonl"); |
| 270 | await NodeFSP.writeFile(path, claudeLine(1, 5) + claudeLine(2, 7)); |
| 271 | const first = await readTranscriptRecords(path, TEST_FORMATS.claude); |
| 272 | assert.isNotNull(first); |
| 273 | assert.strictEqual(first.records.length, 2); |
| 274 | assert.isFalse(first.resumed); |
| 275 | assert.strictEqual(first.malformedRecords, 0); |
| 276 | |
| 277 | await NodeFSP.appendFile(path, claudeLine(3, 11)); |
| 278 | const second = await readTranscriptRecords(path, TEST_FORMATS.claude, first.position); |
| 279 | assert.isNotNull(second); |
| 280 | assert.isTrue(second.resumed); |
| 281 | assert.strictEqual(second.records.length, 1); |
| 282 | assert.strictEqual(second.records[0]?.totals.outputTokens, 11); |
| 283 | |
| 284 | const full = await readTranscriptRecords(path, TEST_FORMATS.claude); |
| 285 | assert.isNotNull(full); |
| 286 | assert.deepStrictEqual([...first.records, ...second.records], [...full.records]); |
| 287 | }); |
| 288 | |
| 289 | it("carries the Codex reducer state across the resume boundary", async () => { |
| 290 | const path = NodePath.join(dir, "rollout.jsonl"); |
| 291 | await NodeFSP.writeFile( |
| 292 | path, |
| 293 | codexMetaLine() + codexModelLine("gpt-5.2-codex") + codexTierLine("ultrafast"), |
| 294 | ); |
| 295 | const first = await readTranscriptRecords(path, TEST_FORMATS.codex); |
| 296 | assert.isNotNull(first); |
| 297 | assert.strictEqual(first.records.length, 0); |
| 298 | |
| 299 | await NodeFSP.appendFile(path, codexUsageLine(9, 5)); |
| 300 | const second = await readTranscriptRecords(path, TEST_FORMATS.codex, first.position); |
| 301 | assert.isNotNull(second); |
| 302 | assert.isTrue(second.resumed); |
| 303 | assert.strictEqual(second.records.length, 1); |
| 304 | assert.strictEqual(second.records[0]?.model, "gpt-5.2-codex"); |
| 305 | assert.strictEqual(second.records[0]?.speed, "ultrafast"); |
| 306 | assert.strictEqual(second.records[0]?.sessionId, "codex-session-1"); |
| 307 | }); |
| 308 | |
| 309 | it("suppresses a Codex duplicate usage event that straddles the boundary", async () => { |
| 310 | const path = NodePath.join(dir, "rollout.jsonl"); |
| 311 | await NodeFSP.writeFile( |
| 312 | path, |
| 313 | codexMetaLine() + codexModelLine("gpt-5.2-codex") + codexUsageLine(9, 5), |
| 314 | ); |
| 315 | const first = await readTranscriptRecords(path, TEST_FORMATS.codex); |
| 316 | assert.isNotNull(first); |
| 317 | assert.strictEqual(first.records.length, 1); |
| 318 | |
| 319 | // Codex duplicates token_count at stream boundaries, including after resume points. |
| 320 | await NodeFSP.appendFile(path, codexUsageLine(9, 5) + codexUsageLine(21, 8)); |
| 321 | const second = await readTranscriptRecords(path, TEST_FORMATS.codex, first.position); |
| 322 | assert.isNotNull(second); |
| 323 | assert.isTrue(second.resumed); |
| 324 | assert.deepStrictEqual( |
| 325 | second.records.map((record) => record.totals.outputTokens), |
| 326 | [21], |
| 327 | ); |
| 328 | }); |
| 329 | |
| 330 | it("defers an unterminated trailing line to tailRecords, then consumes it once terminated", async () => { |
| 331 | const path = NodePath.join(dir, "claude.jsonl"); |
| 332 | const unterminated = claudeLine(2, 7).trimEnd(); |
| 333 | await NodeFSP.writeFile(path, claudeLine(1, 5) + unterminated); |
| 334 | const first = await readTranscriptRecords(path, TEST_FORMATS.claude); |
| 335 | assert.isNotNull(first); |
| 336 | assert.strictEqual(first.records.length, 1); |
| 337 | assert.strictEqual(first.tailRecords.length, 1); |
| 338 | assert.strictEqual(first.tailRecords[0]?.totals.outputTokens, 7); |
| 339 | |
| 340 | await NodeFSP.appendFile(path, `\n${claudeLine(3, 11)}`); |
| 341 | const second = await readTranscriptRecords(path, TEST_FORMATS.claude, first.position); |
| 342 | assert.isNotNull(second); |
| 343 | assert.isTrue(second.resumed); |
| 344 | assert.deepStrictEqual( |
| 345 | second.records.map((record) => record.totals.outputTokens), |
| 346 | [7, 11], |
| 347 | ); |
| 348 | assert.strictEqual(second.tailRecords.length, 0); |
| 349 | }); |
| 350 | |
| 351 | it.each([ |
| 352 | { mode: "buffered", streamingThresholdBytes: Infinity }, |
| 353 | { mode: "streaming", streamingThresholdBytes: 0 }, |
| 354 | ])( |
| 355 | "resumes an initial Codex carried-total tail without rebilling it ($mode)", |
| 356 | async (options) => { |
| 357 | const path = NodePath.join(dir, "continuation.jsonl"); |
| 358 | const usage = (input: number, output: number) => ({ |
| 359 | input_tokens: input, |
| 360 | output_tokens: output, |
| 361 | }); |
| 362 | const event = (last: unknown, cumulative: unknown) => |
| 363 | JSON.stringify({ |
| 364 | type: "event_msg", |
| 365 | timestamp: "2026-08-01T10:00:05Z", |
| 366 | payload: { |
| 367 | type: "token_count", |
| 368 | info: { last_token_usage: last, total_token_usage: cumulative }, |
| 369 | }, |
| 370 | }); |
| 371 | const firstLine = event(usage(100, 20), usage(500, 60)); |
| 372 | await NodeFSP.writeFile( |
| 373 | path, |
| 374 | codexMetaLine() + codexModelLine("gpt-6-astra") + codexTierLine("ultrafast") + firstLine, |
| 375 | ); |
| 376 | const first = await readTranscriptRecords(path, TEST_FORMATS.codex, undefined, options); |
| 377 | assert.isNotNull(first); |
| 378 | assert.strictEqual(first.records.length, 0); |
| 379 | assert.deepStrictEqual( |
| 380 | first.tailRecords.map((r) => [r.totals.outputTokens, r.speed]), |
| 381 | [[20, "ultrafast"]], |
| 382 | ); |
| 383 | assert.strictEqual(first.malformedRecords, 0); |
| 384 | |
| 385 | await NodeFSP.appendFile(path, `\n${firstLine}\n${event(usage(10, 2), usage(510, 62))}\n`); |
| 386 | const resumed = await readTranscriptRecords( |
| 387 | path, |
| 388 | TEST_FORMATS.codex, |
| 389 | JSON.parse(JSON.stringify(first.position)), |
| 390 | options, |
| 391 | ); |
| 392 | const full = await readTranscriptRecords(path, TEST_FORMATS.codex, undefined, options); |
| 393 | assert.isNotNull(resumed); |
| 394 | assert.isNotNull(full); |
| 395 | assert.isTrue(resumed.resumed); |
| 396 | assert.deepStrictEqual(resumed.records, full.records); |
| 397 | assert.deepStrictEqual( |
| 398 | resumed.records.map((r) => [r.totals.outputTokens, r.speed]), |
| 399 | [ |
| 400 | [20, "ultrafast"], |
| 401 | [2, "ultrafast"], |
| 402 | ], |
| 403 | ); |
| 404 | assert.deepStrictEqual(resumed.tailRecords, []); |
| 405 | assert.strictEqual(resumed.malformedRecords, 0); |
| 406 | assert.deepStrictEqual(resumed.position, full.position); |
| 407 | }, |
| 408 | ); |
| 409 | |
| 410 | it.each([ |
| 411 | { mode: "buffered", streamingThresholdBytes: Infinity }, |
| 412 | { mode: "streaming", streamingThresholdBytes: 0 }, |
| 413 | ])( |
| 414 | "bills a Codex request whose turn was interrupted before its token_count ($mode)", |
| 415 | async (options) => { |
| 416 | const path = NodePath.join(dir, "rollout.jsonl"); |
| 417 | const usage = (input: number, output: number) => ({ |
| 418 | input_tokens: input, |
| 419 | cached_input_tokens: 0, |
| 420 | cache_write_input_tokens: 0, |
| 421 | output_tokens: output, |
| 422 | reasoning_output_tokens: 0, |
| 423 | total_tokens: input + output, |
| 424 | }); |
| 425 | const context = (turn: number) => ({ |
| 426 | type: "turn_context", |
| 427 | timestamp: `2026-11-02T10:00:0${turn * 4 - 4}.000Z`, |
| 428 | payload: { turn_id: `turn-${turn}`, model: "gpt-6-sol" }, |
| 429 | }); |
| 430 | const request = (turn: number, id: number, timestamp: string, last: unknown) => ({ |
| 431 | type: "token_usage_record", |
| 432 | timestamp, |
| 433 | payload: { turn_id: `turn-${turn}`, response_id: `resp_${id}`, usage: last }, |
| 434 | }); |
| 435 | const tokenCount = (timestamp: string, last: unknown, total: unknown) => ({ |
| 436 | type: "event_msg", |
| 437 | timestamp, |
| 438 | payload: { |
| 439 | type: "token_count", |
| 440 | info: { last_token_usage: last, total_token_usage: total }, |
| 441 | }, |
| 442 | }); |
| 443 | const lines = [ |
| 444 | { |
| 445 | type: "session_meta", |
| 446 | timestamp: "2026-11-02T10:00:00.000Z", |
| 447 | payload: { id: "aborted-turn-repro", model_provider: "openai" }, |
| 448 | }, |
| 449 | context(1), |
| 450 | request(1, 1, "2026-11-02T10:00:01.000Z", usage(100, 20)), |
| 451 | tokenCount("2026-11-02T10:00:01.001Z", usage(100, 20), usage(100, 20)), |
| 452 | // Interrupted after this request finished: no token_count reports it. |
| 453 | request(1, 2, "2026-11-02T10:00:02.000Z", usage(150, 10)), |
| 454 | { |
| 455 | type: "event_msg", |
| 456 | timestamp: "2026-11-02T10:00:03.000Z", |
| 457 | payload: { type: "turn_aborted", turn_id: "turn-1", reason: "interrupted" }, |
| 458 | }, |
| 459 | context(2), |
| 460 | request(2, 3, "2026-11-02T10:00:05.000Z", usage(200, 30)), |
| 461 | tokenCount("2026-11-02T10:00:05.001Z", usage(200, 30), usage(450, 60)), |
| 462 | ]; |
| 463 | // Lines longer than a read chunk take the projected path when streaming. |
| 464 | const padding = "x".repeat(256 * 1024); |
| 465 | await NodeFSP.writeFile( |
| 466 | path, |
| 467 | lines.map((line) => `${JSON.stringify({ padding, ...line })}\n`).join(""), |
| 468 | ); |
| 469 | |
| 470 | const result = await readTranscriptRecords(path, TEST_FORMATS.codex, undefined, options); |
| 471 | assert.isNotNull(result); |
| 472 | const requests = [...result.records, ...result.tailRecords] |
| 473 | .map((record) => totalTokens(record.totals)) |
| 474 | .toSorted((a, b) => a - b); |
| 475 | assert.deepStrictEqual( |
| 476 | { |
| 477 | billedTokens: requests.reduce((sum, tokens) => sum + tokens, 0), |
| 478 | requests, |
| 479 | malformedRecords: result.malformedRecords, |
| 480 | }, |
| 481 | { billedTokens: 510, requests: [120, 160, 230], malformedRecords: 0 }, |
| 482 | ); |
| 483 | }, |
| 484 | ); |
| 485 | |
| 486 | it("re-parses from the start when the guard bytes no longer match", async () => { |
| 487 | const path = NodePath.join(dir, "claude.jsonl"); |
| 488 | await NodeFSP.writeFile(path, claudeLine(1, 5)); |
| 489 | const first = await readTranscriptRecords(path, TEST_FORMATS.claude); |
| 490 | assert.isNotNull(first); |
| 491 | |
| 492 | await NodeFSP.writeFile(path, claudeLine(4, 13) + claudeLine(5, 17)); |
| 493 | const second = await readTranscriptRecords(path, TEST_FORMATS.claude, first.position); |
| 494 | assert.isNotNull(second); |
| 495 | assert.isFalse(second.resumed); |
| 496 | assert.deepStrictEqual( |
| 497 | second.records.map((record) => record.totals.outputTokens), |
| 498 | [13, 17], |
| 499 | ); |
| 500 | }); |
| 501 | |
| 502 | it("re-parses from the start when the file shrank below the resume point", async () => { |
| 503 | const path = NodePath.join(dir, "claude.jsonl"); |
| 504 | await NodeFSP.writeFile(path, claudeLine(1, 5) + claudeLine(2, 7)); |
| 505 | const first = await readTranscriptRecords(path, TEST_FORMATS.claude); |
| 506 | assert.isNotNull(first); |
| 507 | |
| 508 | await NodeFSP.writeFile(path, claudeLine(3, 11)); |
| 509 | const second = await readTranscriptRecords(path, TEST_FORMATS.claude, first.position); |
| 510 | assert.isNotNull(second); |
| 511 | assert.isFalse(second.resumed); |
| 512 | assert.deepStrictEqual( |
| 513 | second.records.map((record) => record.totals.outputTokens), |
| 514 | [11], |
| 515 | ); |
| 516 | }); |
| 517 | |
| 518 | it("parses a line larger than one stream chunk", async () => { |
| 519 | const path = NodePath.join(dir, "claude.jsonl"); |
| 520 | const bigLine = `${JSON.stringify({ |
| 521 | type: "assistant", |
| 522 | timestamp: "2026-08-01T10:00:00Z", |
| 523 | requestId: "req_big", |
| 524 | sessionId: "session-1", |
| 525 | padding: "x".repeat(512 * 1024), |
| 526 | message: { |
| 527 | id: "msg_big", |
| 528 | model: "claude-fable-5", |
| 529 | usage: { input_tokens: 10, output_tokens: 42 }, |
| 530 | }, |
| 531 | })}\n`; |
| 532 | await NodeFSP.writeFile(path, bigLine + claudeLine(2, 7)); |
| 533 | |
| 534 | const parsed = await readTranscriptRecords(path, TEST_FORMATS.claude); |
| 535 | assert.isNotNull(parsed); |
| 536 | assert.deepStrictEqual( |
| 537 | parsed.records.map((record) => record.totals.outputTokens), |
| 538 | [42, 7], |
| 539 | ); |
| 540 | }); |
| 541 | |
| 542 | it("keeps a CRLF Codex reducer line whose marker straddles a stream chunk", async () => { |
| 543 | const path = NodePath.join(dir, "rollout.jsonl"); |
| 544 | const crlf = (line: string) => line.replace("\n", "\r\n"); |
| 545 | const modelLine = crlf(codexModelLine("gpt-5.2-codex")); |
| 546 | const head = crlf(codexMetaLine()); |
| 547 | const fillerBytes = |
| 548 | 256 * 1024 - Buffer.byteLength(head) - modelLine.indexOf("turn_context") - 4; |
| 549 | const fillerEnvelope = crlf(`${JSON.stringify({ type: "response_item", output: "" })}\n`); |
| 550 | const padding = "é".repeat((fillerBytes - Buffer.byteLength(fillerEnvelope)) / 2); |
| 551 | const filler = crlf(`${JSON.stringify({ type: "response_item", output: padding })}\n`); |
| 552 | const content = head + filler + modelLine + crlf(codexUsageLine(40, 2)); |
| 553 | await NodeFSP.writeFile(path, content); |
| 554 | |
| 555 | const result = await readTranscriptRecords(path, TEST_FORMATS.codex); |
| 556 | assert.isNotNull(result); |
| 557 | assert.strictEqual(result.records.length, 1); |
| 558 | assert.strictEqual(result.records[0]?.model, "gpt-5.2-codex"); |
| 559 | assert.strictEqual(result.records[0]?.sessionId, "codex-session-1"); |
| 560 | assert.strictEqual(result.position.resumeOffset, Buffer.byteLength(content)); |
| 561 | }); |
| 562 | }); |
| 563 | |
| 564 | describe("transcriptLineGate", () => { |
| 565 | const gates = { |
| 566 | one: transcriptLineGate(['"usage"']), |
| 567 | several: transcriptLineGate(['"token_count"', '"session_meta"', '"a.b(c)"']), |
| 568 | }; |
| 569 | const markers = { one: ['"usage"'], several: ['"token_count"', '"session_meta"', '"a.b(c)"'] }; |
| 570 | |
| 571 | it("matches markers beside multi-byte and malformed UTF-8", () => { |
| 572 | const invalid = Buffer.from([0xff, 0xc3]); |
| 573 | for (const kind of ["one", "several"] as const) { |
| 574 | for (const marker of markers[kind]) { |
| 575 | assert.isTrue(gates[kind](Buffer.from(`{"text":"日本語 é",${marker}:1}`))); |
| 576 | assert.isTrue(gates[kind](Buffer.concat([invalid, Buffer.from(marker), invalid]))); |
| 577 | assert.isFalse(gates[kind](Buffer.from(`{"text":"日本語 é ${marker.slice(1, -1)}"}`))); |
| 578 | } |
| 579 | } |
| 580 | }); |
| 581 | }); |
| 582 | |
| 583 | describe("readPhysicalSourceId", () => { |
| 584 | const scope = "0123456789abcdef0123456789abcdef"; |
| 585 | const windowsPath = "C:\\Users\\User\\.claude\\projects"; |
| 586 | |
| 587 | // One NTFS directory the Windows server reads natively and the WSL server |
| 588 | // through drvfs. The temp dir stands in for the drive's mount point. |
| 589 | async function drvfsFixture(extraMounts: (escapedMountPoint: string) => string = () => "") { |
| 590 | const mountPoint = NodePath.join(dir, "drive c"); |
| 591 | await NodeFSP.mkdir(NodePath.join(mountPoint, "Users", "User", ".claude", "projects"), { |
| 592 | recursive: true, |
| 593 | }); |
| 594 | const escaped = mountPoint.replaceAll(" ", "\\040"); |
| 595 | const mountInfoPath = NodePath.join(dir, "mountinfo"); |
| 596 | await NodeFSP.writeFile( |
| 597 | mountInfoPath, |
| 598 | [ |
| 599 | "24 1 8:32 / / rw,relatime shared:1 - ext4 /dev/sdc rw,discard", |
| 600 | `81 24 0:61 / ${escaped} rw,noatime shared:2 - 9p C:\\134 rw,dirsync,aname=drvfs;path=C:\\;uid=1000;gid=1000;symlinkroot=/mnt/,mmap,access=client,msize=65536,trans=fd`, |
| 601 | extraMounts(escaped), |
| 602 | ].join("\n"), |
| 603 | ); |
| 604 | return { mountPoint, mountInfoPath }; |
| 605 | } |
| 606 | |
| 607 | it("names a drvfs directory as the Windows server names it, whatever its typed case", async () => { |
| 608 | const { mountPoint, mountInfoPath } = await drvfsFixture(); |
| 609 | const wslId = await readPhysicalSourceId( |
| 610 | `${mountPoint}/users/user/.claude/projects`, |
| 611 | scope, |
| 612 | "linux", |
| 613 | mountInfoPath, |
| 614 | ); |
| 615 | assert.strictEqual(wslId, physicalSourceIdOf(scope, windowsPath)); |
| 616 | }); |
| 617 | |
| 618 | it("gives no identity to native or shadowing Linux mounts, or without a desktop scope", async () => { |
| 619 | const { mountPoint, mountInfoPath } = await drvfsFixture( |
| 620 | (escaped) => `90 81 8:48 / ${escaped}/Users rw,relatime - ext4 /dev/sdd rw`, |
| 621 | ); |
| 622 | const projects = `${mountPoint}/Users/User/.claude/projects`; |
| 623 | assert.isUndefined(await readPhysicalSourceId(projects, scope, "linux", mountInfoPath)); |
| 624 | assert.isUndefined(await readPhysicalSourceId(dir, scope, "linux", mountInfoPath)); |
| 625 | assert.isUndefined(await readPhysicalSourceId(projects, undefined, "linux", mountInfoPath)); |
| 626 | }); |
| 627 | }); |