From ee3bf285877efb7c4ab47bfed2230007c25ca365 Mon Sep 17 00:00:00 2001
From: Nic Polumeyv <162764842+Nic-Polumeyv@users.noreply.github.com>
Date: Thu, 23 Jul 2026 12:41:36 -0400
Subject: [PATCH] perf: parse large streamed frames in linear time

---
 .changeset/calm-streams-flow.md               |  5 ++
 packages/kit/src/runtime/client/stream.js     | 47 ++++++------
 .../kit/src/runtime/client/stream.spec.js     | 71 +++++++++++++++++++
 3 files changed, 103 insertions(+), 20 deletions(-)
 create mode 100644 .changeset/calm-streams-flow.md
 create mode 100644 packages/kit/src/runtime/client/stream.spec.js

diff --git a/.changeset/calm-streams-flow.md b/.changeset/calm-streams-flow.md
new file mode 100644
index 000000000000..86eb2dfe9493
--- /dev/null
+++ b/.changeset/calm-streams-flow.md
@@ -0,0 +1,5 @@
+---
+'@sveltejs/kit': patch
+---
+
+perf: parse large streamed frames in linear time
diff --git a/packages/kit/src/runtime/client/stream.js b/packages/kit/src/runtime/client/stream.js
index 8256f0609632..da36c7e9fec7 100644
--- a/packages/kit/src/runtime/client/stream.js
+++ b/packages/kit/src/runtime/client/stream.js
@@ -8,32 +8,39 @@
  */
 export async function* read_stream(reader, delimiter, options) {
 	let done = false;
-	let buffer = '';
+	/** @type {string[]} */
+	let parts = [];
+	let rest = '';
 	const decoder = new TextDecoder(undefined, options);
+	const carry = delimiter.length - 1;
 
-	while (true) {
-		let split = buffer.indexOf(delimiter);
-		while (split !== -1) {
-			yield buffer.slice(0, split);
-			buffer = buffer.slice(split + delimiter.length);
-			split = buffer.indexOf(delimiter);
-		}
+	while (!done) {
+		const chunk = await reader.read();
+		done = chunk.done;
+		let text = rest;
+		if (chunk.value) text += decoder.decode(chunk.value, { stream: true });
+		if (done) text += decoder.decode();
 
-		if (done) {
-			if (buffer) {
-				yield buffer;
+		let start = 0;
+		let split = text.indexOf(delimiter);
+		while (split !== -1) {
+			if (parts.length > 0) {
+				parts.push(text.slice(start, split));
+				yield parts.join('');
+				parts = [];
+			} else {
+				yield text.slice(start, split);
 			}
-			return;
+			start = split + delimiter.length;
+			split = text.indexOf(delimiter, start);
 		}
 
-		const chunk = await reader.read();
-		done = chunk.done;
-		if (chunk.value) {
-			buffer += decoder.decode(chunk.value, { stream: true });
-		}
+		const keep = Math.max(start, text.length - carry);
+		if (keep > start) parts.push(text.slice(start, keep));
+		rest = text.slice(keep);
+	}
 
-		if (done) {
-			buffer += decoder.decode();
-		}
+	if (parts.length > 0 || rest) {
+		yield parts.join('') + rest;
 	}
 }
diff --git a/packages/kit/src/runtime/client/stream.spec.js b/packages/kit/src/runtime/client/stream.spec.js
new file mode 100644
index 000000000000..5e6d03c3a342
--- /dev/null
+++ b/packages/kit/src/runtime/client/stream.spec.js
@@ -0,0 +1,71 @@
+import { expect, test } from 'vitest';
+import { read_stream } from './stream.js';
+
+const encoder = new TextEncoder();
+
+/**
+ * @param {Uint8Array[]} chunks
+ * @param {string} delimiter
+ * @param {TextDecoderOptions} [options]
+ * @returns {Promise<string[]>}
+ */
+async function collect(chunks, delimiter, options) {
+	const values = [];
+	const reader = new ReadableStream({
+		start(controller) {
+			for (const chunk of chunks) {
+				controller.enqueue(chunk);
+			}
+			controller.close();
+		}
+	}).getReader();
+
+	for await (const value of read_stream(reader, delimiter, options)) {
+		values.push(value);
+	}
+
+	return values;
+}
+
+test('parses a delimiter split across chunks', async () => {
+	const chunks = [encoder.encode('a\n'), encoder.encode('\nb')];
+
+	await expect(collect(chunks, '\n\n')).resolves.toEqual(['a', 'b']);
+});
+
+test('parses UTF-8 code points split across chunks', async () => {
+	const bytes = encoder.encode('a\u00e9\n');
+	const chunks = Array.from(bytes, (byte) => new Uint8Array([byte]));
+
+	await expect(collect(chunks, '\n')).resolves.toEqual(['a\u00e9']);
+});
+
+test('parses a frame much larger than a chunk', async () => {
+	const size = 100 * 1024;
+	const chunks = Array.from({ length: size / 1024 }, () => encoder.encode('x'.repeat(1024)));
+	chunks.push(encoder.encode('\n'));
+
+	const values = await collect(chunks, '\n');
+	expect(values).toHaveLength(1);
+	expect(values[0]).toHaveLength(size);
+	expect(values[0]?.slice(0, 5)).toBe('xxxxx');
+	expect(values[0]?.slice(-5)).toBe('xxxxx');
+});
+
+test('parses short frames after a long frame', async () => {
+	const prefix = 'x'.repeat(100 * 1024);
+	const chunks = [encoder.encode(prefix), encoder.encode('end\nshort1\nshort2\n')];
+
+	await expect(collect(chunks, '\n')).resolves.toEqual([`${prefix}end`, 'short1', 'short2']);
+});
+
+test('parses a trailing frame without yielding trailing empty frames', async () => {
+	await expect(collect([encoder.encode('trailing')], '\n')).resolves.toEqual(['trailing']);
+	await expect(collect([encoder.encode('complete\n')], '\n')).resolves.toEqual(['complete']);
+});
+
+test('rejects malformed UTF-8', async () => {
+	const chunks = [encoder.encode('valid'), new Uint8Array([0xff]), encoder.encode('\n')];
+
+	await expect(collect(chunks, '\n', { fatal: true })).rejects.toThrow(TypeError);
+});