review: quote JSON log fields, reset sniff on reattach, nits

Author: rpurdel

src/transcriberproxy.ts

@@ -45,7 +45,7 @@ export class TranscriberProxy extends EventEmitter {
 	private audioPacketCount = 0;
 	private interimTranscriptionCount = 0;
 	private finalTranscriptionCount = 0;
-	private firstFrameLoggedTags: Set<string> = new Set();
+	private firstFrameLoggedTags = new Set<string>();
 
 	constructor(ws: WebSocket, options: TranscriberProxyOptions) {
 		super({ captureRejections: true });
@@ -284,11 +284,15 @@ export class TranscriberProxy extends EventEmitter {
 			if (hasAudio) {
 				this.audioPacketCount++;
 				if (!this.firstFrameLoggedTags.has(tag)) {
+					// 64 base64 chars decode to at most 48 bytes; we only emit the first 16.
 					const head = Buffer.from(payloadB64.slice(0, 64), 'base64');
-					const headHex = head.subarray(0, Math.min(16, head.length)).toString('hex');
-					const mediaSnapshot = { ...parsedMessage.media, payload: `<b64:${payloadB64.length} chars, first ${headHex.length / 2} decoded bytes=${headHex}>` };
+					const headByteCount = Math.min(16, head.length);
+					const headHex = head.subarray(0, headByteCount).toString('hex');
+					const mediaSnapshot = { ...parsedMessage.media, payload: `<b64:${payloadB64.length} chars, first ${headByteCount} decoded bytes=${headHex}>` };
+					// JSON-valued fields are quoted so that downstream logfmt-style parsers
+					// don't misinterpret spaces inside the JSON payload (e.g. inside `tag`).
 					logger.info(
-						`First client frame sniff: sessionId=${this.sessionId} tag=${tag} provider=${this.options.provider ?? 'default'} urlEncoding=${this.options.encoding ?? 'opus'} startFormat=${JSON.stringify(connection.getInputFormat())} media=${JSON.stringify(mediaSnapshot)}`,
+						`First client frame sniff: sessionId=${this.sessionId} tag=${tag} provider=${this.options.provider ?? 'default'} urlEncoding=${this.options.encoding ?? 'opus'} startFormat='${JSON.stringify(connection.getInputFormat())}' media='${JSON.stringify(mediaSnapshot)}'`,
 					);
 					this.firstFrameLoggedTags.add(tag);
 				}
@@ -368,6 +372,11 @@ export class TranscriberProxy extends EventEmitter {
 		// Re-setup listeners on new WebSocket
 		this.setupWebSocketListeners();
 
+		// Treat a reattach as a new connection for diagnostic purposes: the client
+		// may negotiate a different audio format on reconnect, so fire the
+		// first-frame sniff again on the first real audio packet per tag.
+		this.firstFrameLoggedTags.clear();
+
 		// Reset chunk tracking on all connections so frames from the new client
 		// aren't discarded as "reordered" (chunk numbers restart from 0)
 		this.outgoingConnections.forEach((connection, tag) => {

test/unit/TranscriberProxy.test.ts

@@ -53,6 +53,7 @@ vi.mock('../../src/OutgoingConnection', () => ({
 		this.addTranscriptContext = vi.fn();
 		this.updateInputFormat = vi.fn();
 		this.getInputFormat = vi.fn(() => inputFormat ?? { encoding: 'opus' });
+		this.resetChunkTracking = vi.fn();
 		this.close = vi.fn();
 		this.onInterimTranscription = undefined;
 		this.onCompleteTranscription = undefined;
@@ -599,7 +600,7 @@ describe('TranscriberProxy', () => {
 			const msg = sniffCalls[0][0] as string;
 			expect(msg).toContain('tag=tag1');
 			expect(msg).toContain('urlEncoding=opus');
-			expect(msg).toContain('startFormat={"encoding":"opus"}');
+			expect(msg).toContain(`startFormat='{"encoding":"opus"}'`);
 			expect(msg).toContain('4f676753'); // 'OggS' in hex
 			expect(msg).toContain(`<b64:${OGG_PAYLOAD.length} chars, first 4 decoded bytes=4f676753>`);
 		});
@@ -647,6 +648,21 @@ describe('TranscriberProxy', () => {
 			expect(endCall?.[0]).toContain('audioPackets=1');
 		});
 
+		it('fires the first-frame sniff again after a WebSocket reattach', () => {
+			const proxy = new TranscriberProxy(mockWebSocket, options);
+			proxy.handleStartEvent({ event: 'start', start: { tag: 'tag1', mediaFormat: { encoding: 'opus' } } });
+
+			proxy.handleMediaEvent({ event: 'media', media: { tag: 'tag1', payload: OGG_PAYLOAD, chunk: 0, timestamp: 0 } });
+			proxy.handleMediaEvent({ event: 'media', media: { tag: 'tag1', payload: OGG_PAYLOAD, chunk: 1, timestamp: 0 } });
+
+			vi.mocked(logger.info).mockClear();
+			proxy.reattachWebSocket({ addEventListener: vi.fn(), send: vi.fn(), close: vi.fn() } as any);
+
+			proxy.handleMediaEvent({ event: 'media', media: { tag: 'tag1', payload: OGG_PAYLOAD, chunk: 0, timestamp: 0 } });
+			const sniffCalls = vi.mocked(logger.info).mock.calls.filter(([msg]) => typeof msg === 'string' && msg.startsWith('First client frame sniff:'));
+			expect(sniffCalls).toHaveLength(1);
+		});
+
 		it('emits a session-end summary with audioPackets, interims, finals, and provider', () => {
 			const proxy = new TranscriberProxy(mockWebSocket, { ...options, provider: 'deepgram' });
 			proxy.handleStartEvent({ event: 'start', start: { tag: 'tag1', mediaFormat: { encoding: 'opus' } } });