[BUG] 2.1.220 regression: severe scroll lag in Je

[jiaxinxing]

[BUG] 2.1.220 regression: severe scroll lag in JetBrains IDEA plugin terminal (absent in 2.1.219)

Summary

A clear performance regression in the TUI rendering / scroll behavior was introduced in 2.1.220. Scrolling the conversation context becomes extremely laggy after only a couple of turns, and the lag also affects the host IDE (JetBrains IDEA). The previous version 2.1.219 did not exhibit this even after tens of hours of continuous chatting in an open session.

Version

• Claude Code: 2.1.220 (confirmed via claude --version)
• Regression introduced in: 2.1.220 (verified by A/B against 2.1.219)
• Previous working version: 2.1.219

Environment

• OS: Windows 11 Enterprise (10.0.26200)
• Shell/terminal: Git Bash launched via the Claude Code JetBrains (IDEA) plugin's integrated terminal
• Model: a third-party model (glm-5.2) — likely unrelated to the rendering issue, noted for completeness
• MCP servers in use: codegraph (local SQLite index + file watcher)

Reproduction

1. Launch Claude Code 2.1.220 through the IDEA plugin (Git Bash terminal widget).
2. Have a normal conversation for 2–3 turns (context is still tiny — far from any auto-compact threshold).
3. Scroll up through the context content.
4. Observe: scrolling is severely laggy / stuttering; the IDEA host UI also becomes sluggish while the Claude tab is focused.

A/B evidence (same machine, same usage pattern)

Version	Session length	Scroll behavior	Host IDE impact
2.1.219	tens of hours, continuously open	Occasionally garbled content on scroll, but not laggy	None
2.1.220	2 questions	Severely laggy	IDEA also lags

This is a clean before/after on identical hardware and usage, so the regression is attributable to 2.1.220.

What I ruled out

• Not context size: only 2 turns of conversation; far below any context window threshold.
• Not CPU saturation: sampled process CPU twice with a 3-second gap; the Claude process delta was ~1.1s, other node processes (codegraph, two stray @openai/codex instances) had delta = 0 (idle) at the time of the lag. No 100% spin.
• Not MCP / file-watcher real-time CPU: codegraph and codex processes were silent during the lag.

Suspected area

Because the symptom appears when scrolling/repainting the conversation buffer and only with a tiny context, and because 2.1.219 on the same setup was fluid (its only issue was occasional garbled content on scroll), the regression most likely lives in the TUI/Ink rendering or the alternate-screen repaint path that changed between 2.1.219 and 2.1.220. The JetBrains terminal widget is the weakest renderer in the chain and amplifies the regression. The changelog entry for 2.1.220 is only "Bug fixes and reliability improvements" with no detail, so I cannot pin a specific commit.

Related issues found

• Idle CLI sessions burn 100%+ CPU each in recurring ~1.1h episodes (~11 CPU-hours per episode); originally attributed to sleep/resume — refuted by instrumentation #81353 — idle CLI sessions burning 100%+ CPU on 2.1.220 (Windows, Bun binary). Different symptom (CPU spin, not scroll lag), but confirms 2.1.220 shipped with performance regressions on Windows.
• [BUG] Claude Code tab severely laggy in Desktop app while Chat tab is fluid. Windows MSIX, CLI unaffected #82350 — severely laggy Claude Code tab in Desktop app; CLI noted as unaffected. Mentions lag but different surface (Desktop, not CLI/IDEA).

Ask

1. What TUI/Ink rendering changes landed in 2.1.220 vs 2.1.219 that affect scroll/repaint on Windows?
2. Is there a known interaction with the JetBrains plugin terminal widget that regressed in 2.1.220?
3. Any recommended workaround besides downgrading to 2.1.219?

Workaround

Downgrade:

npm install -g @anthropic-ai/claude-code@2.1.219

Disable auto-update to prevent re-upgrade:

export DISABLE_AUTOUPDATER=1