AI & Agent Dev Bug Sandbox Back to Radar

Empty `UseServerInsertedHTML` Results Still Trigger A React Render For Each Streamed Chunk

In Next.js App Router, when a `useServerInsertedHTML` callback returns an empty Fragment (or null, empty array), the server still performs an additional React render for each streamed chunk because the callback result is always wrapped in a keyed Fragment, preventing the existing empty-result optimization in `getServerInsertedHTML()`. This leads to significant performance overhead in streaming scenarios with many Suspense boundaries.

highConfidence 92%Next.js Affected V16.3.6 Affected V16.4.0-Canary.48 Affected V16.4.0-Canary.50

Origin Analysis

The function `renderServerInsertedHTML()` wraps every callback result in a keyed React Fragment without checking if the result is actually empty (null, undefined, empty array, empty Fragment, etc.). This makes the collected array non-empty, so `getServerInsertedHTML()` never takes its fast path for empty results, causing a Fizz render and extra event-loop turns for each streamed chunk even when no server-inserted HTML is produced.

1. Clone the reproduction repo: `git clone --branch repro/server-inserted-html-skip --single-branch https://github.com/iamibi/next.js.git` 2. Install dependencies and build: `cd next.js && npm install && npx next build` 3. Start the server with OTel verbose: `NEXT_OTEL_VERBOSE=1 npx next start` 4. Request routes with and without the registry and different boundary counts: `curl -s -o /dev/null 'http://localhost:3000/without-registry?boundaries=5'` `curl -s -o /dev/null 'http://localhost:3000/with-registry?boundaries=5'` `curl -s -o /dev/null 'http://localhost:3000/with-registry?boundaries=20'` 5. Observe span counts: without registry produces 1 server_inserted_html_render; with registry produces 14 for 5 boundaries and 44 for 20. The HTML still contains the style exactly once.

Fixing Code Block

0x7F 45 4C 46 02 01 01 000x90 12 AB 04 C1 7E 33 08
0xA4 1C 90 3E 77 0B D2 580x11 F6 6A 2C 80 19 DE 44
0x00 00 00 00 03 00 3E 000x01 00 B7 4D 55 90 0C 61
0x6C 69 62 63 2E 73 6F 2E0x36 00 8E 17 42 AA 05 F3
0xDE C0 DE 01 8B 45 FC 890xC7 5D 20 6F 91 03 E4 78
0x13 37 00 4F 9A 2D E1 700x08 B4 5C 22 71 D0 4A 16
0x48 8D 3D 22 01 00 00 E80x91 FF 0A 63 2E 57 B8 04
0x5A 91 C3 0E 44 70 1B 880x02 6F 39 C4 15 80 ED 27
0xFF 15 40 20 60 00 48 890xC3 90 7A 01 4C 28 B1 56
0x0F 1F 44 00 00 55 48 890xE5 C3 18 6D 92 0B 33 A7

Edge Case Audit

The filtering must be strictly limited to statically empty nodes to avoid removing components that might have side effects or conditionally render content. Ensure that the check for empty Fragment only considers direct children and does not attempt to evaluate dynamic content. If any regression occurs (e.g., missing styles or unexpected rendering), revert this change by removing the filter. This fix should be tested with various CSS-in-JS libraries and edge cases where callbacks return nested fragments or arrays. The change should be safe for concurrency and streaming, but monitor memory and performance to confirm no new overhead is introduced.