perf(ccusage): scan Bun block JSONL bytes

Extend the Bun JSONL byte-scan path to block loading. Usage rows are still decoded and parsed through the existing fast parser, while non-usage block timestamps are extracted directly from bytes so block loading avoids decoding every JSONL line on Bun. Node keeps the existing text/file-handle path.

On real local Claude data with LOG_LEVEL=0 and COLUMNS=200, the combined Bun byte-scan path improved the pre-byte baseline from daily JSON 419.9ms ± 6.2ms to 362.7ms ± 7.6ms, session JSON 450.2ms ± 20.9ms to 378.0ms ± 9.3ms, and blocks JSON 551.4ms ± 9.2ms to 462.2ms ± 13.8ms. Table output saw the same shape: daily 431.2ms ± 21.5ms to 368.1ms ± 6.9ms, session 442.7ms ± 5.7ms to 379.8ms ± 7.7ms, and blocks 552.9ms ± 12.9ms to 462.8ms ± 12.3ms.

Current Rust #977 reference from the same host remains faster: JS daily 368.2ms ± 15.0ms vs Rust 292.9ms ± 4.1ms, JS session 376.4ms ± 10.9ms vs Rust 298.3ms ± 5.5ms, and JS blocks 449.1ms ± 8.0ms vs Rust 297.7ms ± 5.4ms.

Validation: pnpm run format; pnpm typecheck; pnpm --filter ccusage exec vitest run src/data-loader.ts --testNamePattern loadSessionBlockData\|loadDailyUsageData\|loadSessionData; pnpm --filter ccusage run build; pnpm run test; Bun/Node JSON parity for daily/session/blocks/monthly/weekly; table parity for daily/session/blocks.

Author: ryoppippi

apps/ccusage/src/data-loader.ts

@@ -84,6 +84,7 @@ const REQUEST_ID_MARKER = '"requestId":"';
 const SESSION_ID_MARKER = '"sessionId":"';
 const SPEED_MARKER = '"speed":"';
 const TIMESTAMP_MARKER = '"timestamp":"';
+const TIMESTAMP_MARKER_BUFFER = Buffer.from(TIMESTAMP_MARKER);
 const VERSION_MARKER = '"version":"';
 const VERSION_PATTERN = /^\d+\.\d+\.\d+/;
 function parseTwoDigits(value: string, offset: number): number {
@@ -233,6 +234,28 @@ function getTimestampFromLine(line: string): Date | null {
 	}
 }
 
+function getTimestampFromBytes(content: Buffer, lineStart: number, lineEnd: number): Date | null {
+	const timestampStart = content.indexOf(TIMESTAMP_MARKER_BUFFER, lineStart);
+	if (timestampStart === -1 || timestampStart >= lineEnd) {
+		return null;
+	}
+
+	const valueStart = timestampStart + TIMESTAMP_MARKER_BUFFER.length;
+	const valueEnd = content.indexOf(34, valueStart);
+	if (valueEnd === -1 || valueEnd > lineEnd) {
+		return null;
+	}
+
+	const timestamp = content.subarray(valueStart, valueEnd).toString('utf8');
+	const parsedTimestampMs = parseIsoTimestampMs(timestamp);
+	if (!Number.isNaN(parsedTimestampMs)) {
+		return new Date(parsedTimestampMs);
+	}
+
+	const date = new Date(timestamp);
+	return Number.isNaN(date.getTime()) ? null : date;
+}
+
 /**
  * Get Claude data directories to search for usage data
  * When CLAUDE_CONFIG_DIR is set: uses only those paths
@@ -2008,7 +2031,7 @@ async function collectBlockFileResult(
 		}
 	};
 
-	await processJSONLFileByLine(file, (line) => {
+	const processLine = (line: string): void => {
 		try {
 			if (!line.includes(USAGE_LINE_MARKER)) {
 				const lineTimestamp = getTimestampFromLine(line);
@@ -2083,7 +2106,35 @@ async function collectBlockFileResult(
 				`Skipping invalid JSON line in 5-hour blocks: ${error instanceof Error ? error.message : String(error)}`,
 			);
 		}
-	});
+	};
+
+	const bytes = await readBufferedJSONLBytes(file);
+	if (bytes != null) {
+		const content = Buffer.isBuffer(bytes)
+			? bytes
+			: Buffer.from(bytes.buffer, bytes.byteOffset, bytes.byteLength);
+		let lineStart = 0;
+		let markerIndex = content.indexOf(USAGE_LINE_MARKER_BUFFER);
+		while (lineStart < content.length) {
+			let lineEnd = content.indexOf(10, lineStart);
+			if (lineEnd === -1) {
+				lineEnd = content.length;
+			}
+			const decodeEnd = lineEnd > lineStart && content[lineEnd - 1] === 13 ? lineEnd - 1 : lineEnd;
+			if (markerIndex !== -1 && markerIndex < decodeEnd) {
+				processLine(content.subarray(lineStart, decodeEnd).toString('utf8'));
+				markerIndex = content.indexOf(USAGE_LINE_MARKER_BUFFER, lineEnd + 1);
+			} else {
+				const lineTimestamp = getTimestampFromBytes(content, lineStart, decodeEnd);
+				if (lineTimestamp != null) {
+					setEarliestTimestamp(lineTimestamp, lineTimestamp.getTime());
+				}
+			}
+			lineStart = lineEnd + 1;
+		}
+	} else {
+		await processJSONLFileByLine(file, processLine);
+	}
 
 	return {
 		file,