@threadbase-sh/streamer 1.27.3 → 1.28.1
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.
- package/dist/cli.cjs +495 -54
- package/dist/cli.cjs.map +1 -1
- package/dist/index.cjs +464 -33
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.cts +117 -1
- package/dist/index.d.ts +117 -1
- package/dist/index.js +445 -11
- package/dist/index.js.map +1 -1
- package/dist/migrations/009_create_offset_index.sql +28 -0
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -3124,9 +3124,15 @@ async function createDirectory(parentAbsolutePath, name) {
|
|
|
3124
3124
|
}
|
|
3125
3125
|
|
|
3126
3126
|
// src/conversation-cache.ts
|
|
3127
|
+
import {
|
|
3128
|
+
createJsonlParseState,
|
|
3129
|
+
parseJsonlLine
|
|
3130
|
+
} from "@threadbase-sh/scanner";
|
|
3127
3131
|
import Database from "better-sqlite3";
|
|
3128
3132
|
import { closeSync as closeSync2, existsSync as existsSync5, mkdirSync as mkdirSync2, openSync as openSync2, readSync as readSync2, statSync as statSync2 } from "fs";
|
|
3133
|
+
import { open as openAsync } from "fs/promises";
|
|
3129
3134
|
import { dirname as dirname6 } from "path";
|
|
3135
|
+
import { setImmediate as yieldToEventLoop } from "timers/promises";
|
|
3130
3136
|
|
|
3131
3137
|
// src/db/sqlite-migrate.ts
|
|
3132
3138
|
import { readdirSync as readdirSync2, readFileSync as readFileSync5 } from "fs";
|
|
@@ -3238,6 +3244,33 @@ function parseAgentEntrypointsEnv(raw) {
|
|
|
3238
3244
|
return new Set(parts);
|
|
3239
3245
|
}
|
|
3240
3246
|
|
|
3247
|
+
// src/utils/fileIdentity.ts
|
|
3248
|
+
import { createHash } from "crypto";
|
|
3249
|
+
function fileIdentity(stat3, headBytes) {
|
|
3250
|
+
if (stat3.ino && stat3.ino > 0) return `inode:${stat3.dev}:${stat3.ino}`;
|
|
3251
|
+
const head = headBytes ?? Buffer.alloc(0);
|
|
3252
|
+
return `fp:${createHash("sha1").update(head).digest("hex")}`;
|
|
3253
|
+
}
|
|
3254
|
+
function splitCompleteLines(buf, baseOffset) {
|
|
3255
|
+
const spans = [];
|
|
3256
|
+
let lineStart = 0;
|
|
3257
|
+
let consumed = 0;
|
|
3258
|
+
for (let i = 0; i < buf.length; i++) {
|
|
3259
|
+
if (buf[i] !== 10) continue;
|
|
3260
|
+
const lineLen = i - lineStart;
|
|
3261
|
+
if (lineLen > 0) {
|
|
3262
|
+
spans.push({
|
|
3263
|
+
byteOffset: baseOffset + lineStart,
|
|
3264
|
+
byteLength: lineLen,
|
|
3265
|
+
text: buf.toString("utf-8", lineStart, i)
|
|
3266
|
+
});
|
|
3267
|
+
}
|
|
3268
|
+
lineStart = i + 1;
|
|
3269
|
+
consumed = lineStart;
|
|
3270
|
+
}
|
|
3271
|
+
return { spans, consumed };
|
|
3272
|
+
}
|
|
3273
|
+
|
|
3241
3274
|
// src/conversation-cache.ts
|
|
3242
3275
|
function shortProjectName(fullPath) {
|
|
3243
3276
|
const parts = fullPath.split(/[/\\]/).filter(Boolean);
|
|
@@ -3286,6 +3319,14 @@ var ConversationCache = class _ConversationCache {
|
|
|
3286
3319
|
tailSize;
|
|
3287
3320
|
fileIndex = /* @__PURE__ */ new Map();
|
|
3288
3321
|
fileIndexLoaded = false;
|
|
3322
|
+
// Per-file scanner parse state for the incremental offset-index writer. The
|
|
3323
|
+
// reducer is stateful across lines (pending tool_uses, latest timestamp), so
|
|
3324
|
+
// it must persist between watcher reads of the same file. Cleared on
|
|
3325
|
+
// truncation/backfill.
|
|
3326
|
+
indexParseState = /* @__PURE__ */ new Map();
|
|
3327
|
+
// Single-flight guard for backfillIndex — concurrent detail requests for the
|
|
3328
|
+
// same cold file await one walk, not N. Entry dropped on settle.
|
|
3329
|
+
backfillInFlight = /* @__PURE__ */ new Map();
|
|
3289
3330
|
// Monotonically increasing counter for tail updated_at — guarantees strict
|
|
3290
3331
|
// ordering even when multiple updateFromLine() calls land within the same ms.
|
|
3291
3332
|
tailSeq = Date.now();
|
|
@@ -3449,6 +3490,42 @@ var ConversationCache = class _ConversationCache {
|
|
|
3449
3490
|
GROUP BY project_path
|
|
3450
3491
|
ORDER BY cnt DESC
|
|
3451
3492
|
LIMIT ?`
|
|
3493
|
+
),
|
|
3494
|
+
getFileState: db.prepare("SELECT * FROM conversation_file_state WHERE path = ?"),
|
|
3495
|
+
upsertFileState: db.prepare(
|
|
3496
|
+
`INSERT INTO conversation_file_state
|
|
3497
|
+
(path, identity, size, mtime_ms, byte_offset, last_message_index)
|
|
3498
|
+
VALUES (@path, @identity, @size, @mtime_ms, @byte_offset, @last_message_index)
|
|
3499
|
+
ON CONFLICT(path) DO UPDATE SET
|
|
3500
|
+
identity = excluded.identity,
|
|
3501
|
+
size = excluded.size,
|
|
3502
|
+
mtime_ms = excluded.mtime_ms,
|
|
3503
|
+
byte_offset = excluded.byte_offset,
|
|
3504
|
+
last_message_index = excluded.last_message_index`
|
|
3505
|
+
),
|
|
3506
|
+
deleteFileState: db.prepare("DELETE FROM conversation_file_state WHERE path = ?"),
|
|
3507
|
+
deleteMessageIndex: db.prepare(
|
|
3508
|
+
"DELETE FROM conversation_message_index WHERE conversation_id = ?"
|
|
3509
|
+
),
|
|
3510
|
+
insertMessageIndexRow: db.prepare(
|
|
3511
|
+
`INSERT INTO conversation_message_index
|
|
3512
|
+
(conversation_id, message_index, byte_offset, byte_length, uuid, role, ts)
|
|
3513
|
+
VALUES (@conversation_id, @message_index, @byte_offset, @byte_length, @uuid, @role, @ts)
|
|
3514
|
+
ON CONFLICT(conversation_id, message_index) DO UPDATE SET
|
|
3515
|
+
byte_offset = excluded.byte_offset,
|
|
3516
|
+
byte_length = excluded.byte_length,
|
|
3517
|
+
uuid = excluded.uuid,
|
|
3518
|
+
role = excluded.role,
|
|
3519
|
+
ts = excluded.ts`
|
|
3520
|
+
),
|
|
3521
|
+
getMessageIndexWindow: db.prepare(
|
|
3522
|
+
`SELECT message_index, byte_offset, byte_length, uuid, role, ts
|
|
3523
|
+
FROM conversation_message_index
|
|
3524
|
+
WHERE conversation_id = ? AND message_index >= ? AND message_index < ?
|
|
3525
|
+
ORDER BY message_index ASC`
|
|
3526
|
+
),
|
|
3527
|
+
getIndexedMessageCount: db.prepare(
|
|
3528
|
+
"SELECT COUNT(*) as cnt FROM conversation_message_index WHERE conversation_id = ?"
|
|
3452
3529
|
)
|
|
3453
3530
|
};
|
|
3454
3531
|
}
|
|
@@ -3459,6 +3536,273 @@ var ConversationCache = class _ConversationCache {
|
|
|
3459
3536
|
getDatabase() {
|
|
3460
3537
|
return this.db;
|
|
3461
3538
|
}
|
|
3539
|
+
// ── Offset index (design 1b) ────────────────────────────────────────────
|
|
3540
|
+
// conversation_file_state + conversation_message_index back the windowed
|
|
3541
|
+
// detail read path (SQL window select + pread of byte ranges). All methods
|
|
3542
|
+
// are thin wrappers over the prepared statements above.
|
|
3543
|
+
getFileState(path) {
|
|
3544
|
+
return this.stmts.getFileState.get(path) ?? null;
|
|
3545
|
+
}
|
|
3546
|
+
upsertFileState(row) {
|
|
3547
|
+
this.stmts.upsertFileState.run(row);
|
|
3548
|
+
}
|
|
3549
|
+
/** Drop a file's index rows + file_state (truncation / identity change). */
|
|
3550
|
+
deleteFileIndex(path, conversationId) {
|
|
3551
|
+
const tx = this.db.transaction(() => {
|
|
3552
|
+
this.stmts.deleteMessageIndex.run(conversationId);
|
|
3553
|
+
this.stmts.deleteFileState.run(path);
|
|
3554
|
+
});
|
|
3555
|
+
tx();
|
|
3556
|
+
this.indexParseState.delete(path);
|
|
3557
|
+
}
|
|
3558
|
+
/** Append/replace index rows in one transaction. */
|
|
3559
|
+
appendMessageIndexRows(rows) {
|
|
3560
|
+
const tx = this.db.transaction((batch) => {
|
|
3561
|
+
for (const r of batch) this.stmts.insertMessageIndexRow.run(r);
|
|
3562
|
+
});
|
|
3563
|
+
tx(rows);
|
|
3564
|
+
}
|
|
3565
|
+
/** Rows for message_index in [fromIndex, toIndex), ordered ascending. */
|
|
3566
|
+
getMessageIndexWindow(conversationId, fromIndex, toIndex) {
|
|
3567
|
+
return this.stmts.getMessageIndexWindow.all(
|
|
3568
|
+
conversationId,
|
|
3569
|
+
fromIndex,
|
|
3570
|
+
toIndex
|
|
3571
|
+
);
|
|
3572
|
+
}
|
|
3573
|
+
getIndexedMessageCount(conversationId) {
|
|
3574
|
+
return this.stmts.getIndexedMessageCount.get(conversationId).cnt;
|
|
3575
|
+
}
|
|
3576
|
+
/**
|
|
3577
|
+
* Conversation id for a JSONL path — the filename stem (matches the pseudo-id
|
|
3578
|
+
* updateFromLine derives and the uuid the detail read path resolves). The
|
|
3579
|
+
* offset index keys on this so the window select and the cursor agree.
|
|
3580
|
+
*/
|
|
3581
|
+
static conversationIdForFile(filePath) {
|
|
3582
|
+
return filePath.split(/[/\\]/).pop()?.replace(/\.jsonl$/, "") ?? filePath;
|
|
3583
|
+
}
|
|
3584
|
+
// The offset index understands only claude-code JSONL: parseJsonlLine is the
|
|
3585
|
+
// claude-code reducer, so a codex file "indexes" as zero messages and the
|
|
3586
|
+
// resulting file_state serves empty windows for a real conversation (the
|
|
3587
|
+
// silent-wrong-data bug hotfixed after 1.28.0). Non-claude providers are
|
|
3588
|
+
// excluded from the index entirely; the scanner serves them (it routes each
|
|
3589
|
+
// provider to its own parser).
|
|
3590
|
+
isIndexableFile(filePath) {
|
|
3591
|
+
const meta = this.getMetaById(_ConversationCache.conversationIdForFile(filePath));
|
|
3592
|
+
return (meta?.provider ?? CLAUDE_CODE_PROVIDER) === CLAUDE_CODE_PROVIDER;
|
|
3593
|
+
}
|
|
3594
|
+
/**
|
|
3595
|
+
* Incremental offset-index writer: extend the index for a burst of appended
|
|
3596
|
+
* lines (one watcher read) using their byte spans. Each line is classified
|
|
3597
|
+
* with the scanner's parseJsonlLine (a running per-file reducer state), so the
|
|
3598
|
+
* message ordering can never drift from the scanner's. Message lines get an
|
|
3599
|
+
* index row at the next message_index; non-message lines (summary/sidecar)
|
|
3600
|
+
* get no row but still advance byte_offset. file_state is updated to the end
|
|
3601
|
+
* of the last consumed span.
|
|
3602
|
+
*
|
|
3603
|
+
* Requires an up-to-date `stat` (identity/size/mtime) for the file so the read
|
|
3604
|
+
* path can detect truncation/replacement.
|
|
3605
|
+
*
|
|
3606
|
+
* `readFrom` is the absolute byte offset the watcher read started at, and
|
|
3607
|
+
* `endOffset` is where it ended (readFrom + consumed, i.e. the watcher's new
|
|
3608
|
+
* entry.offset). CONTIGUITY GUARD: the read must begin exactly where the index
|
|
3609
|
+
* left off (`readFrom === existing.byte_offset`, or 0 with no state). If it
|
|
3610
|
+
* doesn't — the watcher attached at EOF after the server was down, or an
|
|
3611
|
+
* append raced an in-flight backfill — extending would assign wrong
|
|
3612
|
+
* message_index values over a hole. In that case this writes nothing and
|
|
3613
|
+
* returns null so the caller drops the index and backfills.
|
|
3614
|
+
*
|
|
3615
|
+
* On success returns the message_index assigned to each input span (null for a
|
|
3616
|
+
* non-message line) so the caller can stamp WS `seq`. Empty array when spans
|
|
3617
|
+
* is empty. `endOffset` is stored verbatim as byte_offset so the watcher's
|
|
3618
|
+
* offset and file_state.byte_offset are the same number by construction.
|
|
3619
|
+
*/
|
|
3620
|
+
extendMessageIndex(filePath, spans, stat3, readFrom, endOffset) {
|
|
3621
|
+
if (!this.isIndexableFile(filePath)) return spans.map(() => null);
|
|
3622
|
+
const existing = this.getFileState(filePath);
|
|
3623
|
+
const expectedStart = existing?.byte_offset ?? 0;
|
|
3624
|
+
if (readFrom !== expectedStart) return null;
|
|
3625
|
+
if (spans.length === 0) return [];
|
|
3626
|
+
const convId = _ConversationCache.conversationIdForFile(filePath);
|
|
3627
|
+
let state = this.indexParseState.get(filePath);
|
|
3628
|
+
if (!state) {
|
|
3629
|
+
state = createJsonlParseState();
|
|
3630
|
+
this.indexParseState.set(filePath, state);
|
|
3631
|
+
}
|
|
3632
|
+
let nextIndex = existing ? existing.last_message_index + 1 : 0;
|
|
3633
|
+
const rows = [];
|
|
3634
|
+
const seqs = [];
|
|
3635
|
+
for (const span of spans) {
|
|
3636
|
+
const msg = parseJsonlLine(span.text, state);
|
|
3637
|
+
if (!msg) {
|
|
3638
|
+
seqs.push(null);
|
|
3639
|
+
continue;
|
|
3640
|
+
}
|
|
3641
|
+
rows.push({
|
|
3642
|
+
conversation_id: convId,
|
|
3643
|
+
message_index: nextIndex,
|
|
3644
|
+
byte_offset: span.byteOffset,
|
|
3645
|
+
byte_length: span.byteLength,
|
|
3646
|
+
uuid: msg.uuid ?? null,
|
|
3647
|
+
role: msg.role ?? null,
|
|
3648
|
+
ts: msg.timestamp ? Date.parse(msg.timestamp) || null : null
|
|
3649
|
+
});
|
|
3650
|
+
seqs.push(nextIndex);
|
|
3651
|
+
nextIndex++;
|
|
3652
|
+
}
|
|
3653
|
+
const tx = this.db.transaction(() => {
|
|
3654
|
+
for (const r of rows) this.stmts.insertMessageIndexRow.run(r);
|
|
3655
|
+
this.stmts.upsertFileState.run({
|
|
3656
|
+
path: filePath,
|
|
3657
|
+
identity: fileIdentity(stat3),
|
|
3658
|
+
size: stat3.size,
|
|
3659
|
+
mtime_ms: Math.round(stat3.mtimeMs),
|
|
3660
|
+
// Store the watcher's end offset verbatim — same number as entry.offset,
|
|
3661
|
+
// so the next read's contiguity check compares like-for-like (never
|
|
3662
|
+
// false-positive on a read that ended in trailing empty lines).
|
|
3663
|
+
byte_offset: endOffset,
|
|
3664
|
+
last_message_index: nextIndex - 1
|
|
3665
|
+
});
|
|
3666
|
+
});
|
|
3667
|
+
tx();
|
|
3668
|
+
return seqs;
|
|
3669
|
+
}
|
|
3670
|
+
clearIndexParseState(filePath) {
|
|
3671
|
+
this.indexParseState.delete(filePath);
|
|
3672
|
+
}
|
|
3673
|
+
/**
|
|
3674
|
+
* On-demand full backfill of the offset index for a file with no/stale
|
|
3675
|
+
* file_state (cold conversation, or after a truncation/replacement). Rebuilds
|
|
3676
|
+
* from byte 0: drops any existing rows, walks the whole file in chunks with a
|
|
3677
|
+
* running parse state, yields to the event loop every ~1000 lines so a large
|
|
3678
|
+
* file never blocks, and writes index rows + file_state.
|
|
3679
|
+
*
|
|
3680
|
+
* Single-flighted per path: concurrent callers await the same walk. The
|
|
3681
|
+
* triggering detail request is served by the scanner fallback while this runs.
|
|
3682
|
+
*/
|
|
3683
|
+
backfillIndex(filePath) {
|
|
3684
|
+
const inFlight = this.backfillInFlight.get(filePath);
|
|
3685
|
+
if (inFlight) return inFlight;
|
|
3686
|
+
const walk = this.runBackfill(filePath).finally(() => {
|
|
3687
|
+
this.backfillInFlight.delete(filePath);
|
|
3688
|
+
});
|
|
3689
|
+
this.backfillInFlight.set(filePath, walk);
|
|
3690
|
+
return walk;
|
|
3691
|
+
}
|
|
3692
|
+
async runBackfill(filePath) {
|
|
3693
|
+
const convId = _ConversationCache.conversationIdForFile(filePath);
|
|
3694
|
+
this.deleteFileIndex(filePath, convId);
|
|
3695
|
+
this.indexParseState.delete(filePath);
|
|
3696
|
+
if (!this.isIndexableFile(filePath)) return;
|
|
3697
|
+
const CHUNK = 256 * 1024;
|
|
3698
|
+
const YIELD_EVERY = 1e3;
|
|
3699
|
+
const state = createJsonlParseState();
|
|
3700
|
+
const fh = await openAsync(filePath, "r");
|
|
3701
|
+
let fileOffset = 0;
|
|
3702
|
+
let carry = Buffer.alloc(0);
|
|
3703
|
+
let nextIndex = 0;
|
|
3704
|
+
let linesSinceYield = 0;
|
|
3705
|
+
let lastConsumedEnd = 0;
|
|
3706
|
+
let stat3;
|
|
3707
|
+
try {
|
|
3708
|
+
stat3 = await fh.stat();
|
|
3709
|
+
const buf = Buffer.alloc(CHUNK);
|
|
3710
|
+
for (; ; ) {
|
|
3711
|
+
const { bytesRead } = await fh.read(buf, 0, CHUNK, null);
|
|
3712
|
+
if (bytesRead === 0) break;
|
|
3713
|
+
const combined = carry.length > 0 ? Buffer.concat([carry, buf.subarray(0, bytesRead)]) : buf.subarray(0, bytesRead);
|
|
3714
|
+
const { spans, consumed } = splitCompleteLines(combined, fileOffset);
|
|
3715
|
+
const rows = [];
|
|
3716
|
+
for (const span of spans) {
|
|
3717
|
+
const msg = parseJsonlLine(span.text, state);
|
|
3718
|
+
linesSinceYield++;
|
|
3719
|
+
if (msg) {
|
|
3720
|
+
rows.push({
|
|
3721
|
+
conversation_id: convId,
|
|
3722
|
+
message_index: nextIndex,
|
|
3723
|
+
byte_offset: span.byteOffset,
|
|
3724
|
+
byte_length: span.byteLength,
|
|
3725
|
+
uuid: msg.uuid ?? null,
|
|
3726
|
+
role: msg.role ?? null,
|
|
3727
|
+
ts: msg.timestamp ? Date.parse(msg.timestamp) || null : null
|
|
3728
|
+
});
|
|
3729
|
+
nextIndex++;
|
|
3730
|
+
}
|
|
3731
|
+
if (linesSinceYield >= YIELD_EVERY) {
|
|
3732
|
+
linesSinceYield = 0;
|
|
3733
|
+
await yieldToEventLoop();
|
|
3734
|
+
}
|
|
3735
|
+
}
|
|
3736
|
+
if (rows.length > 0) this.appendMessageIndexRows(rows);
|
|
3737
|
+
lastConsumedEnd = fileOffset + consumed;
|
|
3738
|
+
carry = Buffer.from(combined.subarray(consumed));
|
|
3739
|
+
fileOffset += consumed;
|
|
3740
|
+
}
|
|
3741
|
+
} finally {
|
|
3742
|
+
await fh.close();
|
|
3743
|
+
}
|
|
3744
|
+
this.upsertFileState({
|
|
3745
|
+
path: filePath,
|
|
3746
|
+
identity: fileIdentity(stat3),
|
|
3747
|
+
size: stat3.size,
|
|
3748
|
+
mtime_ms: Math.round(stat3.mtimeMs),
|
|
3749
|
+
byte_offset: lastConsumedEnd,
|
|
3750
|
+
last_message_index: nextIndex - 1
|
|
3751
|
+
});
|
|
3752
|
+
this.indexParseState.set(filePath, state);
|
|
3753
|
+
}
|
|
3754
|
+
/**
|
|
3755
|
+
* Windowed detail read straight from the offset index — the hot path.
|
|
3756
|
+
* Returns the parsed messages for message_index in [fromIndex, toIndex) plus
|
|
3757
|
+
* the total indexed count, or null when the index can't serve this file (no
|
|
3758
|
+
* file_state, identity/size mismatch = truncation/replacement, or cold index)
|
|
3759
|
+
* so the caller falls back to the scanner and enqueues a backfill.
|
|
3760
|
+
*
|
|
3761
|
+
* On a match it SQL-selects the window's byte ranges and preads exactly those
|
|
3762
|
+
* ranges from the JSONL (never the whole file), parsing only the sliced lines.
|
|
3763
|
+
* Returns messages in the same ConversationMessage shape parseJsonlLine
|
|
3764
|
+
* produces during a scan, so the payload is identical to the scanner path.
|
|
3765
|
+
*/
|
|
3766
|
+
readMessageWindow(filePath, fromIndex, toIndex) {
|
|
3767
|
+
const fileState = this.getFileState(filePath);
|
|
3768
|
+
if (!fileState) return null;
|
|
3769
|
+
let stat3;
|
|
3770
|
+
try {
|
|
3771
|
+
stat3 = statSync2(filePath);
|
|
3772
|
+
} catch {
|
|
3773
|
+
return null;
|
|
3774
|
+
}
|
|
3775
|
+
if (fileIdentity(stat3) !== fileState.identity || stat3.size !== fileState.byte_offset) {
|
|
3776
|
+
return null;
|
|
3777
|
+
}
|
|
3778
|
+
if (fileState.last_message_index < 0 || !this.isIndexableFile(filePath)) {
|
|
3779
|
+
return null;
|
|
3780
|
+
}
|
|
3781
|
+
const total = fileState.last_message_index + 1;
|
|
3782
|
+
const from = Math.max(0, fromIndex);
|
|
3783
|
+
const to = Math.min(toIndex, total);
|
|
3784
|
+
if (to <= from) return { messages: [], total, fromIndex: from };
|
|
3785
|
+
const rows = this.getMessageIndexWindow(
|
|
3786
|
+
_ConversationCache.conversationIdForFile(filePath),
|
|
3787
|
+
from,
|
|
3788
|
+
to
|
|
3789
|
+
);
|
|
3790
|
+
if (rows.length === 0) return { messages: [], total, fromIndex: from };
|
|
3791
|
+
const messages = [];
|
|
3792
|
+
const fd = openSync2(filePath, "r");
|
|
3793
|
+
try {
|
|
3794
|
+
const state = createJsonlParseState();
|
|
3795
|
+
for (const row of rows) {
|
|
3796
|
+
const buf = Buffer.alloc(row.byte_length);
|
|
3797
|
+
readSync2(fd, buf, 0, row.byte_length, row.byte_offset);
|
|
3798
|
+
const msg = parseJsonlLine(buf.toString("utf-8"), state);
|
|
3799
|
+
if (msg) messages.push(msg);
|
|
3800
|
+
}
|
|
3801
|
+
} finally {
|
|
3802
|
+
closeSync2(fd);
|
|
3803
|
+
}
|
|
3804
|
+
return { messages, total, fromIndex: from };
|
|
3805
|
+
}
|
|
3462
3806
|
agentEntrypointsKey() {
|
|
3463
3807
|
return [...this.agentEntrypoints].sort().join(",");
|
|
3464
3808
|
}
|
|
@@ -4359,12 +4703,14 @@ var ConversationWatcher = class {
|
|
|
4359
4703
|
directories = /* @__PURE__ */ new Map();
|
|
4360
4704
|
onNewLine;
|
|
4361
4705
|
onNewLines;
|
|
4706
|
+
onNewLineSpans;
|
|
4362
4707
|
onConversationChanged;
|
|
4363
4708
|
onFileDeleted;
|
|
4364
4709
|
onError;
|
|
4365
4710
|
constructor(events = {}) {
|
|
4366
4711
|
this.onNewLine = events.onNewLine;
|
|
4367
4712
|
this.onNewLines = events.onNewLines;
|
|
4713
|
+
this.onNewLineSpans = events.onNewLineSpans;
|
|
4368
4714
|
this.onConversationChanged = events.onConversationChanged;
|
|
4369
4715
|
this.onFileDeleted = events.onFileDeleted;
|
|
4370
4716
|
this.onError = events.onError;
|
|
@@ -4468,9 +4814,13 @@ var ConversationWatcher = class {
|
|
|
4468
4814
|
} finally {
|
|
4469
4815
|
await fh.close();
|
|
4470
4816
|
}
|
|
4471
|
-
|
|
4817
|
+
const { spans, consumed } = splitCompleteLines(buf, readFrom);
|
|
4818
|
+
entry.offset = readFrom + consumed;
|
|
4472
4819
|
if (!this.files.has(filePath)) return;
|
|
4473
|
-
const lines =
|
|
4820
|
+
const lines = spans.map((s) => s.text);
|
|
4821
|
+
if (spans.length > 0) {
|
|
4822
|
+
this.onNewLineSpans?.(filePath, spans, readFrom, entry.offset);
|
|
4823
|
+
}
|
|
4474
4824
|
if (this.onNewLines) {
|
|
4475
4825
|
this.onNewLines(filePath, lines);
|
|
4476
4826
|
} else {
|
|
@@ -4939,13 +5289,13 @@ function sanitizeFilename(name) {
|
|
|
4939
5289
|
}
|
|
4940
5290
|
|
|
4941
5291
|
// src/utils/conversationEtag.ts
|
|
4942
|
-
import { createHash } from "crypto";
|
|
5292
|
+
import { createHash as createHash2 } from "crypto";
|
|
4943
5293
|
function computeConversationEtag({
|
|
4944
5294
|
filePath,
|
|
4945
5295
|
messageCount,
|
|
4946
5296
|
timestamp
|
|
4947
5297
|
}) {
|
|
4948
|
-
const digest =
|
|
5298
|
+
const digest = createHash2("sha1").update(`${filePath}:${messageCount}:${timestamp}`).digest("hex").slice(0, 16);
|
|
4949
5299
|
return `"${digest}"`;
|
|
4950
5300
|
}
|
|
4951
5301
|
|
|
@@ -5110,6 +5460,10 @@ var StreamerServer = class {
|
|
|
5110
5460
|
fileWatcher;
|
|
5111
5461
|
sessionFileMap = /* @__PURE__ */ new Map();
|
|
5112
5462
|
// sessionId → JSONL filePath
|
|
5463
|
+
// Per-file seq assignments from the most recent onNewLineSpans (offset index),
|
|
5464
|
+
// handed to the immediately-following onNewLines so it can stamp WS `seq` on
|
|
5465
|
+
// the matching conversation_events entries. Same read → same lines order.
|
|
5466
|
+
pendingLineSeqs = /* @__PURE__ */ new Map();
|
|
5113
5467
|
pendingQuestions = /* @__PURE__ */ new Map();
|
|
5114
5468
|
// Content key of the AskUserQuestion currently broadcast for a session (from
|
|
5115
5469
|
// either the rendered screen or JSONL), used to de-dupe the two paths: when
|
|
@@ -5255,6 +5609,42 @@ var StreamerServer = class {
|
|
|
5255
5609
|
this.sessionStore = new SessionStore();
|
|
5256
5610
|
this.wsHub = new WSHub();
|
|
5257
5611
|
this.fileWatcher = new ConversationWatcher({
|
|
5612
|
+
onNewLineSpans: (filePath, spans, readFrom, endOffset) => {
|
|
5613
|
+
if (!this.cache) return;
|
|
5614
|
+
const cache = this.cache;
|
|
5615
|
+
this.pendingLineSeqs.delete(filePath);
|
|
5616
|
+
try {
|
|
5617
|
+
const seqs = cache.extendMessageIndex(
|
|
5618
|
+
filePath,
|
|
5619
|
+
spans,
|
|
5620
|
+
statSync5(filePath),
|
|
5621
|
+
readFrom,
|
|
5622
|
+
endOffset
|
|
5623
|
+
);
|
|
5624
|
+
if (seqs === null) {
|
|
5625
|
+
cache.deleteFileIndex(filePath, ConversationCache.conversationIdForFile(filePath));
|
|
5626
|
+
cache.clearIndexParseState(filePath);
|
|
5627
|
+
this.trackCacheWrite(
|
|
5628
|
+
cache.backfillIndex(filePath).catch((err) => {
|
|
5629
|
+
this.log.warn("offset-index.backfill_failed", {
|
|
5630
|
+
event: "offset_index.backfill_failed",
|
|
5631
|
+
filePath,
|
|
5632
|
+
trigger: "noncontiguous-append",
|
|
5633
|
+
err
|
|
5634
|
+
});
|
|
5635
|
+
})
|
|
5636
|
+
);
|
|
5637
|
+
return;
|
|
5638
|
+
}
|
|
5639
|
+
this.pendingLineSeqs.set(filePath, seqs);
|
|
5640
|
+
} catch (err) {
|
|
5641
|
+
this.log.warn("offset-index.extend_failed", {
|
|
5642
|
+
event: "offset_index.extend_failed",
|
|
5643
|
+
filePath,
|
|
5644
|
+
err
|
|
5645
|
+
});
|
|
5646
|
+
}
|
|
5647
|
+
},
|
|
5258
5648
|
onNewLines: (filePath, lines) => {
|
|
5259
5649
|
this.cache?.updateFromLines(filePath, lines);
|
|
5260
5650
|
for (const [sessionId, watchedPath] of this.sessionFileMap) {
|
|
@@ -5281,13 +5671,20 @@ var StreamerServer = class {
|
|
|
5281
5671
|
this.pendingQuestionKey.set(sessionId, key);
|
|
5282
5672
|
if (broadcast) this.wsHub.broadcast(m);
|
|
5283
5673
|
}
|
|
5284
|
-
this.
|
|
5674
|
+
const seqs = this.pendingLineSeqs.get(filePath);
|
|
5675
|
+
this.wsHub.broadcast({
|
|
5676
|
+
type: "conversation_events",
|
|
5677
|
+
sessionId,
|
|
5678
|
+
lines,
|
|
5679
|
+
...seqs && seqs.length === lines.length ? { seqs } : {}
|
|
5680
|
+
});
|
|
5285
5681
|
for (const line of lines) {
|
|
5286
5682
|
this.wsHub.broadcast({ type: "conversation_event", sessionId, line });
|
|
5287
5683
|
}
|
|
5288
5684
|
break;
|
|
5289
5685
|
}
|
|
5290
5686
|
}
|
|
5687
|
+
this.pendingLineSeqs.delete(filePath);
|
|
5291
5688
|
},
|
|
5292
5689
|
onConversationChanged: (filePath) => {
|
|
5293
5690
|
this.fileWatcher.poke(filePath);
|
|
@@ -6371,9 +6768,13 @@ var StreamerServer = class {
|
|
|
6371
6768
|
return;
|
|
6372
6769
|
}
|
|
6373
6770
|
const etagSource = conversation;
|
|
6771
|
+
const indexedCount = etagSource.filePath && this.cache ? this.cache.getIndexedMessageCount(
|
|
6772
|
+
ConversationCache.conversationIdForFile(etagSource.filePath)
|
|
6773
|
+
) : 0;
|
|
6774
|
+
const etagMessageCount = Math.max(etagSource.messageCount, indexedCount);
|
|
6374
6775
|
const etag = computeConversationEtag({
|
|
6375
6776
|
filePath: etagSource.filePath,
|
|
6376
|
-
messageCount:
|
|
6777
|
+
messageCount: etagMessageCount,
|
|
6377
6778
|
timestamp: etagSource.timestamp
|
|
6378
6779
|
});
|
|
6379
6780
|
const isFirstPage = !url.searchParams.has("before_index") && !url.searchParams.has("anchor_index") && !url.searchParams.has("after_index");
|
|
@@ -6390,12 +6791,14 @@ var StreamerServer = class {
|
|
|
6390
6791
|
let slice = filtered;
|
|
6391
6792
|
let fromIdx = 0;
|
|
6392
6793
|
let messagePagination;
|
|
6794
|
+
let indexTotal = null;
|
|
6393
6795
|
if (usePaging) {
|
|
6394
6796
|
const limit = Math.min(Math.max(intParam(url, "msg_limit", 80), 1), 500);
|
|
6395
6797
|
let beforeIndex = total;
|
|
6396
6798
|
let scanLimit = limit;
|
|
6397
6799
|
let anchorIndex = null;
|
|
6398
6800
|
let newerPaging = false;
|
|
6801
|
+
let usedAfterIndex = false;
|
|
6399
6802
|
if (url.searchParams.has("before_index")) {
|
|
6400
6803
|
beforeIndex = intParam(url, "before_index", total);
|
|
6401
6804
|
beforeIndex = Math.min(Math.max(beforeIndex, 0), total);
|
|
@@ -6404,6 +6807,7 @@ var StreamerServer = class {
|
|
|
6404
6807
|
beforeIndex = Math.min(total, from + limit);
|
|
6405
6808
|
scanLimit = beforeIndex - from;
|
|
6406
6809
|
newerPaging = true;
|
|
6810
|
+
usedAfterIndex = true;
|
|
6407
6811
|
} else if (hasAnchor) {
|
|
6408
6812
|
anchorIndex = Math.min(
|
|
6409
6813
|
Math.max(intParam(url, "anchor_index", 0), 0),
|
|
@@ -6413,9 +6817,34 @@ var StreamerServer = class {
|
|
|
6413
6817
|
beforeIndex = Math.min(total, from + limit);
|
|
6414
6818
|
newerPaging = true;
|
|
6415
6819
|
}
|
|
6416
|
-
const
|
|
6417
|
-
const
|
|
6418
|
-
|
|
6820
|
+
const isTailRequest = !url.searchParams.has("before_index") && !hasAfter && !hasAnchor;
|
|
6821
|
+
const indexFilePath = conversation.filePath;
|
|
6822
|
+
if (isTailRequest && indexFilePath && this.cache) {
|
|
6823
|
+
const indexed = this.cache.getIndexedMessageCount(
|
|
6824
|
+
ConversationCache.conversationIdForFile(indexFilePath)
|
|
6825
|
+
);
|
|
6826
|
+
if (indexed > beforeIndex) {
|
|
6827
|
+
beforeIndex = indexed;
|
|
6828
|
+
}
|
|
6829
|
+
}
|
|
6830
|
+
const windowStart = Math.max(0, beforeIndex - scanLimit);
|
|
6831
|
+
const indexWindow = scanLimit > 0 && !hasAnchor && indexFilePath && this.cache ? this.cache.readMessageWindow(indexFilePath, windowStart, beforeIndex) : null;
|
|
6832
|
+
if (!indexWindow && indexFilePath && this.cache && !hasAnchor) {
|
|
6833
|
+
this.trackCacheWrite(
|
|
6834
|
+
this.cache.backfillIndex(indexFilePath).catch((err) => {
|
|
6835
|
+
this.log.warn("offset-index.backfill_failed", {
|
|
6836
|
+
event: "offset_index.backfill_failed",
|
|
6837
|
+
conversationId: id,
|
|
6838
|
+
filePath: indexFilePath,
|
|
6839
|
+
err
|
|
6840
|
+
});
|
|
6841
|
+
})
|
|
6842
|
+
);
|
|
6843
|
+
}
|
|
6844
|
+
const pagedScanner = !indexWindow && this.scannerReady ? await this.getScanner(true) : null;
|
|
6845
|
+
const page = indexWindow ?? (scanLimit > 0 && pagedScanner && typeof pagedScanner.getConversationPage === "function" ? await pagedScanner.getConversationPage(id, { beforeIndex, limit: scanLimit }) : null);
|
|
6846
|
+
if (indexWindow) indexTotal = indexWindow.total;
|
|
6847
|
+
const start = page?.fromIndex ?? windowStart;
|
|
6419
6848
|
slice = page?.messages ?? filtered.slice(start, beforeIndex);
|
|
6420
6849
|
fromIdx = start;
|
|
6421
6850
|
const effectiveTotal = page?.total ?? total;
|
|
@@ -6431,6 +6860,9 @@ var StreamerServer = class {
|
|
|
6431
6860
|
messagePagination.has_more_newer = beforeIndex < effectiveTotal;
|
|
6432
6861
|
messagePagination.next_after_index = beforeIndex < effectiveTotal ? beforeIndex : null;
|
|
6433
6862
|
}
|
|
6863
|
+
if (usedAfterIndex) {
|
|
6864
|
+
messagePagination.etag = etag;
|
|
6865
|
+
}
|
|
6434
6866
|
}
|
|
6435
6867
|
const messagesPayload = slice.map((m, localIdx) => {
|
|
6436
6868
|
const content = [];
|
|
@@ -6472,6 +6904,8 @@ var StreamerServer = class {
|
|
|
6472
6904
|
const cachedConvMeta = this.cache?.getMetaById(id);
|
|
6473
6905
|
const convProvider = coerceProviderForRunner(conv.provider ?? cachedConvMeta?.provider);
|
|
6474
6906
|
const availability = classifyResumability(conv.projectPath);
|
|
6907
|
+
const metaMessageCount = indexTotal != null && indexTotal > conv.messageCount ? indexTotal : conv.messageCount;
|
|
6908
|
+
const metaLastUpdatedAt = indexTotal != null && indexTotal > conv.messageCount ? slice.at(-1)?.timestamp ?? conv.timestamp : conv.timestamp;
|
|
6475
6909
|
const body = {
|
|
6476
6910
|
meta: {
|
|
6477
6911
|
id,
|
|
@@ -6479,8 +6913,8 @@ var StreamerServer = class {
|
|
|
6479
6913
|
project_name: conv.projectName,
|
|
6480
6914
|
project_path: conv.projectPath,
|
|
6481
6915
|
file_path: conv.filePath,
|
|
6482
|
-
last_updated_at:
|
|
6483
|
-
message_count:
|
|
6916
|
+
last_updated_at: metaLastUpdatedAt,
|
|
6917
|
+
message_count: metaMessageCount,
|
|
6484
6918
|
last_prompt: conv.lastPrompt ?? void 0,
|
|
6485
6919
|
provider: convProvider,
|
|
6486
6920
|
resumable: isProviderResumable(convProvider, availability.resumable),
|