AWS CLI v2 aborts under MallocStackLogging on macOS 26

Minimal reproduction: the AWS CLI v2 aborts with SIGABRT (exit code 134) on macOS 26 (Tahoe) whenever the MallocStackLogging environment variable is set. It fires even for aws --version — no AWS account, credentials, or network call required.

Symptom

$ MallocStackLogging=1 aws --version
Assertion failed: (previous_refcount >= refs && "Detected possible double free
or buffer overrun of heap allocation. Consider reproducing with ASan enabled."),
function uniquing_table_node_release_internal, file uniquing_table_mutator.c, line 1291.
Abort trap: 6
$ echo $?
134

Without the variable, the same command succeeds (exit 0).

Reproduce

• CI: Actions tab → run the aws-cli-mallocstacklogging-macos26 workflow (runs free on a public repo's standard macos-26 runner).

• Locally, on any macOS 26 machine:

  aws --version                       # exit 0
  MallocStackLogging=1 aws --version  # Abort trap: 6, exit 134

Notes

• The faulting frame is in Apple's libmalloc stack-logging (uniquing_table_mutator.c / uniquing_table_node_release_internal), surfaced through the AWS CLI's bundled Python runtime.
• MallocStackLogging is a standard Apple memory-debugging flag; tooling that sets it job-wide (e.g. for crash diagnostics) makes every subsequent aws invocation abort.
• Observed on GitHub-hosted macos-26 runners: aws-cli/2.34.x, Python 3.14, macOS 26.4.