macOS desktop app ≥26.730 crash-loops on launch on 16 GB Apple Silicon Mac (V8 crash ~6–15 s after launch); same builds fine on 48 GB machine

Resolved 💬 3 comments Opened Aug 7, 2026 by pkruger1977 Closed Aug 7, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

Summary

Every build from 26.730.61639 onward crash-loops on launch on one of my Macs, while the identical builds run fine on another. 26.727.51351 is the last build that works on the affected machine.

  • 26.730.61639 (Sparkle 6234): crashes ~6 s after launch — V8 "JavaScript heap out of memory", faulting thread CrBrowserMain.
  • 26.803.41515 (Sparkle 6321): crashes ~6–15 s after launch — EXC_BREAKPOINT (SIGTRAP) in V8, faulting thread V8Worker.

Both crash reports attached.

Environment comparison (why I believe it's memory-related)

| | Affected machine | Working machine |
|---|---|---|
| Model | MacBook (Apple Silicon) | Mac mini 2024 (M4 Pro) |
| RAM | 16 GB | 48 GB |
| macOS | 26.2 → updated to 26.6.1 (25G76) | 26.3.1(a) |
| Result | Crashes on 26.730 and 26.803 | Both builds run fine |

Ruled out

  • User-level state: launching with a fresh CODEX_HOME and fresh Application Support still crashes.
  • Corrupt download / Sparkle delta: performed a clean direct install of 26.803 from the full appcast zip (ChatGPT-darwin-arm64-26.803.41515.zip, size verified, codesign -v --deep passes) — still crashes.
  • macOS version: updated the affected machine from 26.2 to 26.6.1 — newer than the working machine's OS — and 26.803 still crashes with the identical signature.

The only remaining differences are hardware — most notably 16 GB vs 48 GB RAM. Since V8 sizes its default heap relative to system memory, a launch-time allocation that fits within the larger heap cap on a 48 GB machine could exhaust the smaller cap on a 16 GB machine, which matches the 26.730 heap-OOM signature.

Crash signatures

26.730.61639 — faulting thread CrBrowserMain:

V8 JavaScript heap out of memory
(full .ips attached)

26.803.41515 — faulting thread V8Worker:

EXC_BREAKPOINT (SIGTRAP)
frames in Codex Framework: v8::CppHeap::CollectGarbageInYoungGeneration…,
v8::internal::compiler::CompilationDependencies::DependOnElementsKind, …
(full .ips attached)

Steps to reproduce (affected machine)

  1. Install 26.730 or later on a 16 GB Apple Silicon Mac (macOS 26.2 or 26.6.1).
  2. Launch the app. No interaction needed.
  3. App quits ~6–15 s after launch; relaunches via crash recovery and dies again (crash loop).

Workaround

Pinned to 26.727.51351: chflags uchg on the bundle + SUFeedURL pointed at localhost to block Sparkle. Works, but means no updates.

Happy to provide more diagnostics, run instrumented builds, or test a fix.

---
Full .ips crash reports for both builds available — will attach in a comment.

View original on GitHub ↗

3 Comments

github-actions[bot] contributor · 20 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #36523

Powered by Codex Action

pkruger1977 · 20 days ago

Closing as duplicate of #36523 — same root cause (external-agent-import parsing ~/Library/Application Support/Claude into the V8 heap at startup). My confirming data (19 GB / 79,740 files → deterministic crash on every launch; identical builds fine on a machine without Claude data) posted there: https://github.com/openai/codex/issues/36523#issuecomment-5221739702

CrossPeakZane · 15 days ago

What does closed as not planned mean? Is OpenAI ok with Mac crashing right away? I appreciate any clarification.