macOS libsystem_pthread POSIX thread cancellation bug

[dimpase]

Output of brew config

HOMEBREW_VERSION: 7.0.9-163-g3849568
ORIGIN: https://github.com/Homebrew/brew
HEAD: 38495687df01c86c02349292cb58603c1130b2b7
Last commit: 6 hours ago
Branch: main
Core tap HEAD: 652f016237dc92ce3f1faa9e3af85017657fce75
Core tap last commit: 3 hours ago
Core cask tap: N/A
HOMEBREW_PREFIX: /opt/homebrew
Homebrew Ruby: 4.0.7 => /opt/homebrew/Library/Homebrew/vendor/portable-ruby/4.0.7_1/bin/ruby
CPU: 14-core 64-bit arm_brava
Clang: 21.0.0 build 2100
Git: 2.54.0 => /Library/Developer/CommandLineTools/usr/bin/git
Curl: 8.7.1 => /usr/bin/curl
macOS: 27.0.1-arm64
CLT: 27.0.0.0.1788430756
Xcode: 27.0
Metal Toolchain: N/A
Rosetta 2: false

Output of brew doctor

Warning: Some installed formulae are deprecated or disabled.
You should find replacements for the following formulae:

  expect
  monkeysphere
  tcl-tk@8

Warning: You have unlinked kegs in your Cellar.
Leaving kegs unlinked can lead to build-trouble and cause formulae that depend on
those kegs to fail to run properly once built.

Run `brew link` on these:
  r
  numpy
  scipy

Description of issue

various long computations using POSIX threads eventually lead to zombie processes clogging up CPUs and RAM.
It's not limited to Darwin-27, we also see the same on macos-15-intel, and on other arm64 versions of macOS.

We traced it down to threads just not getting cancelled, in a random way (in the test example below about one per 1500).
With the C code attached, and Python script attached, and run as follows, we see a rougue thread like this:

% ./reproduce.py 
1000 runs completed
Run 1338 timed out; sampling PID 12201
{
  "started": "2026-10-17T21:56:34.320583-05:00",
  "build_command": [
    "cc",
    "-O2",
    "-g",
    "-Wall",
    "-Wextra",
    "-pthread",
    "/private/tmp/bill-pthread-checked.c",
    "-o",
    "/private/tmp/results/bill-pthread-checked"
  ],
  "requested_iterations": 10000,
  "timeout_seconds": 5,
  "confirmation_seconds": 10,
  "completed": 1337,
  "outcome": "persistent-hang",
  "iteration": 1338,
  "pid": 12201,
  "observed_at": "2026-10-17T21:56:42.629657-05:00",
  "initial_sample_status": 0,
  "confirmation_sample_status": 0,
  "returncode": -9,
  "stdout": "",
  "stderr": "",
  "elapsed_seconds": 20.53075637499569
}

whereas on Linux no runs are timed out.

We filed this with Apple as FB25195692. Hopefully someone with contacts at Apple can nudge them to look into it (I imagine it may even have an exploitable, via denial of service, potential).
The C code showing this
has been out on the next for almost a year by now.

bill-pthread-checked.c
reproduce.py

[p-linnane]

This appears to be a macOS bug, and your reproducer doesn’t involve Homebrew.

[dimpase]

Homebrew is often used for running long computational jobs using pthreads, at least in my community. And it also can affect macOS CI runners, something used by Homebrew in an essential way. Thus it's a bug affecting Homebrew users harder than the rest.

[p-linnane]

Sure, but my point is that there isn't anything Homebrew can do about this.

[dimpase]

you say somewhere that you do have upstream contacts at Apple, no?

[p-linnane]

Asking an open-source project to escalate your issue through personal contacts isn’t an appropriate substitute for Apple’s bug-reporting process. I’m closing this discussion and suggest you follow up through your existing report to Apple.