Copilot Chat typing stutter - suggestModel.ts consuming 40% of scripting time

[tharveysa]

Trace-20251223T104426.zip

Issue Type: Performance

VS Code Version

1.107.1 (994fd12, x64)

OS Version

Windows 11

Description

Typing in Copilot Chat panel causes significant UI stuttering/lag. Performance profiling shows suggestModel.ts:338 consuming 40% of scripting self-time during chat input, even though suggestions aren't being displayed in the chat input.

Steps to Reproduce

1. Open VS Code
2. Open Copilot Chat panel (Ctrl+Alt+I)
3. Start typing a message
4. Observe stuttering/lag while typing

Performance Profile Findings

Captured a ~37 second trace while reproducing. Bottom-up analysis sorted by Self Time:

Self Time	%	Function	Source
783.6 ms	40.1%	y	suggestModel.ts:338:10
342.3 ms	17.5%	encode	-
178.2 ms	9.1%	send	ipc.electron.ts:23:2
147.4 ms	7.5%	send	VM4 sandbox_bundle
98.9 ms	5.1%	concat	buffer.ts:849
76.5 ms	3.9%	Major GC	Garbage collection

The suggest model appears to be processing every keystroke in the chat input, triggering heavy IPC and encoding operations, causing the main thread to block.

Workaround

Adding these settings reduces the stutter:

{
  "editor.quickSuggestionsDelay": 500,
  "editor.suggest.delay": 500,
  "editor.inlineSuggest.suppressSuggestions": true
}

Expected Behavior

Typing in Copilot Chat should not trigger the editor's suggestion model processing, or should be debounced/throttled to prevent UI blocking.

Additional Context

• Performance trace attached

[tharveysa]

Trace-20251223T111351.zip

Follow-up: Workaround ineffective, issue persists

The suggested workaround settings did not resolve the issue. Captured a second trace after applying:

• editor.quickSuggestionsDelay: 500
• editor.suggest.delay: 500
• editor.inlineSuggest.suppressSuggestions: true

Second trace Bottom-up analysis (Self Time):

Self Time	%	Total Time	%	Function	Source
2,134.1 ms	33.4%	3,338.8 ms	52.2%	y	suggestModel.ts:338:10
1,113.1 ms	17.4%	1,198.4 ms	18.7%	encode	-
545.3 ms	8.5%	548.3 ms	8.6%	stringify	rpcProtocol.ts:81:10
478.7 ms	7.5%	1,011.8 ms	15.8%	send	ipc.electron.ts:23:2
441.4 ms	6.9%	441.4 ms	6.9%	objectHash	hash.ts:64:10
368.5 ms	5.8%	530.7 ms	8.3%	send	VM662 sandbox_bundle

The suggestModel.ts now shows 52% total time (up from 63% in first trace), and the editor.suggest.* settings appear to have no effect. The chat input seems to trigger suggestions through a code path that bypasses these settings.

Additional culprits visible in this trace:

• Heavy RPC serialization (stringify, encode, objectHash)
• IPC overhead to extension host

Second trace attached.