Summary
The infinite React/Ink render loop reported and closed in #2802 (v1.0.31) has regressed. On recent releases the main view intermittently freezes: the submitted prompt disappears, the UI shows "Working...", and no output is rendered. Exiting and running /resume dumps the previously-invisible buffered output and the session often works on the next try. It is intermittent - with the same binary, one session works and the next freezes within the same minute.
The underlying error in the logs is identical to #2802:
[ERROR] Maximum update depth exceeded. This can happen when a component
calls setState inside useEffect, but useEffect either doesn't have a
dependency array, or one of the dependencies changes on every render.
Version correlation (strong evidence it's a version regression)
Across all local ~/.copilot/logs/process-*.log files:
- Zero
Maximum update depth occurrences on v1.0.67 and earlier (ran 3-4 concurrent sessions daily for days with no wedge).
- First-ever occurrence:
2026-07-20T18:24:58.714Z - ~1 second after an auto-update completed:
2026-07-20T18:24:44.133Z [INFO] Cleaned up 1 old package version(s): 1.0.67
2026-07-20T18:24:57.593Z [INFO] Successfully updated binary
2026-07-20T18:24:58.714Z [ERROR] Maximum update depth exceeded. ...
That update landed on v1.0.72.
- Still reproduces on v1.0.73.
- 15 separate session logs carry the loop, all on/after that update.
- Rolling the global npm binary back to v1.0.67 eliminates it (confirmed stable).
(I can't pinpoint whether 1.0.68-1.0.71 were affected - they only ran briefly during the auto-update climb - but 1.0.67 is clean and 1.0.72/1.0.73 are not.)
Things that are NOT the cause (ruled out)
- Not multi-session contention - this workflow runs 3-4 CLI sessions in the VS Code integrated terminal daily and was fine for months on 1.0.67.
- The IDE banner
Conflict: A connection for this session already exists prints on every launch (working or frozen) - benign, not a predictor.
MCP Servers reloaded: 1 server connected is not a reliable success marker - sessions froze even after it printed.
- No reliable UI tell for working-vs-frozen; it behaves like a race in the IDE-reconnect / MCP-reload handshake, which matches the intermittency and the self-heal on a fresh
/resume.
Environment
Workaround
setx COPILOT_AUTO_UPDATE false # pin auto-update off
npm install -g @github/copilot@1.0.67 # roll back to last-good
# then fully exit all CLI sessions and restart VS Code
Related
Summary
The infinite React/Ink render loop reported and closed in #2802 (v1.0.31) has regressed. On recent releases the main view intermittently freezes: the submitted prompt disappears, the UI shows "Working...", and no output is rendered. Exiting and running
/resumedumps the previously-invisible buffered output and the session often works on the next try. It is intermittent - with the same binary, one session works and the next freezes within the same minute.The underlying error in the logs is identical to #2802:
Version correlation (strong evidence it's a version regression)
Across all local
~/.copilot/logs/process-*.logfiles:Maximum update depthoccurrences on v1.0.67 and earlier (ran 3-4 concurrent sessions daily for days with no wedge).2026-07-20T18:24:58.714Z- ~1 second after an auto-update completed:(I can't pinpoint whether 1.0.68-1.0.71 were affected - they only ran briefly during the auto-update climb - but 1.0.67 is clean and 1.0.72/1.0.73 are not.)
Things that are NOT the cause (ruled out)
Conflict: A connection for this session already existsprints on every launch (working or frozen) - benign, not a predictor.MCP Servers reloaded: 1 server connectedis not a reliable success marker - sessions froze even after it printed./resume.Environment
Workaround
Related
write EIO/EPIPEstdout cascade - different mechanism).