Vite core member finds three methodology flaws in OJ's memory benchmarks vs Vite
cnakazawa · x · 2026-09-20
Vite core team member sapphi-red filed an issue on lovablelabs/oj arguing its README benchmarks comparing against Vite are methodologically flawed and may make the memory comparison misleading:
- Only the parent process is measured: the script uses ps -o rss= -p $PID, missing oj's separately spawned Node process — roughly 110MB uncounted at 1k components and 150MB at 10k on the author's machine.
- Memory measured before GC: JS doesn't eagerly release memory, so point-in-time RSS includes no-longer-needed memory; forcing GC before measuring cut Vite's reported usage by 30%, though forced GC suits retained-memory measurement rather than normal runtime.
- RSS alone is misleading: RSS includes reclaimable clean and file-backed pages; mimalloc (Rolldown's allocator) retains fragmented allocator pages after dependency optimization, keeping RSS high even when much is reclaimable.
The author says the goal was finding wins to bring back to Vite/Rolldown.
More from coding & agent
- Swarms Cloud changelog: Auto Agent Builder, MCP, and full agent execution observability — KyeGomezB · 2026-09-20
- Ex-OpenAI researcher: iterate on the plan document with multiple frontier models before writing code — doodlestein · 2026-09-20
- Using Jev as a 75ms 'subconscious' filter to cut agent context bloat for under half a cent a day — Obvious_Unicorn · 2026-09-20
- Scoble says an AI agent wrote every word of an entire book — Scobleizer · 2026-09-20
- "Look at your data": dev mocks reflex to spin up another agent — chrisalbon · 2026-09-20
- Evaluating 7 Models Across Claude Code, Codex, and Pi: Harness Choice Drives Cost, Not Success — CShorten30 · 2026-09-20