ChatGPT desktop app freezes when switching to a large image-heavy thread (ResizeObserver loop)

Open 💬 2 comments Opened Jul 17, 2026 by super-administrator

What version of the Codex App are you using (From “About Codex” dialog)?

26.715.21425 (5488)

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Darwin 25.4.0 arm64 arm

What issue are you seeing?

On macOS, switching to one specific long, image-heavy task frequently makes the ChatGPT desktop app unresponsive. The freeze occurs during view activation even when that task is already loaded; during one observed freeze, no new thread/read or thread/resume request was made.

During diagnostics:

  • The app logs repeatedly reported: ResizeObserver loop completed with undelivered notifications.
  • Approximately 828 occurrences were observed overall; one 14-second burst contained 47 occurrences.
  • The main renderer was approximately 860 MB RSS and all related app processes were approximately 4.2 GB RSS in one snapshot.
  • In an earlier incident, macOS Activity Monitor displayed approximately 190 GB memory use and automatically paused the app.
  • GPU/renderer activity spikes while switching to the affected task.

This appears to be a renderer/layout problem rather than slow task retrieval.

What steps can reproduce the bug?

  1. Open the ChatGPT desktop app on macOS.
  2. Open a long task containing many screenshots/images and tool outputs (the affected local rollout is approximately 21 MB, about 70 turn contexts, and has gone through 3 context compactions).
  3. Navigate to another task.
  4. Switch back to the affected task.
  5. The app becomes unresponsive during task view activation. Switching away and back, or restarting directly into this task, can reproduce the problem.

The affected task's session ID is omitted from this public report because it contains private project context, but it can be provided privately to OpenAI staff if needed.

What is the expected behavior?

Switching to an already-loaded task should complete without freezing, repeated layout-observer loops, or excessive renderer/memory growth. Large image/tool-output histories should be virtualized or otherwise bounded during view activation.

Additional information

App version/build: 26.715.21425 (5488)
Display name: ChatGPT
macOS bundle identifier: com.openai.codex
Architecture: Apple Silicon

Possibly related to #33582, #18693, and #22091, but this case adds a narrower trigger and renderer evidence: the affected rollout is only about 21 MB, task retrieval is not the bottleneck, switching to the already-loaded task triggers the freeze, and it correlates with repeated ResizeObserver layout-loop errors.

Raw logs and the full task are not attached because they contain private project data and internal network information. Sanitized diagnostic excerpts can be provided privately.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗