[Bug] postProvision.sh fails on cleanup step — exit code 1 despite successful provisioning

[sihbher]

Summary

azd provision reports a failed post-provision hook (exit code: 1) due to a rm -rf cleanup failure on the temporary Python
virtual environment. All actual Azure resource provisioning and configuration steps complete successfully before this error
occurs.

Root Cause

The script scripts/postProvision.sh uses set -euo pipefail (correct for provisioning logic), but the cleanup block at the end
has no error handling:

deactivate
rm -rf config/.venv_temp

On macOS with Python 3.14, rm -rf fails with "Directory not empty" when .pyc files or file handles are still held open briefly
after python -m invocations. Under set -e, this causes the entire script to exit with failure.

Additionally, deactivate may reference undefined shell variables ($_OLD_VIRTUAL_PS1) under set -u in non-interactive shells.

Steps to Reproduce

1. Run azd init -t azure/gpt-rag on macOS with Python
 3.14
2. Run azd provision
3. All Azure resources provision successfully
4. Observe the cleanup failure:

🧹 Cleaning Python environment up…
rm: config/.venv_temp/lib/python3.14: Directory not empty
ERROR: failed running post hooks: 'postprovision' hook failed with exit code: '1'

Expected Behavior

The hook should report success when all provisioning steps complete. Cleanup failures should not be fatal.

Proposed Solution

Register cleanup as a trap EXIT handler so it always runs and never propagates its exit code:

cleanup() {
  deactivate 2>/dev/null || true
  rm -rf config/.venv_temp 2>/dev/null || true
}
trap cleanup EXIT

Note: The equivalent postProvision.ps1 already has its cleanup commented out with retry logic, confirming this class of failure
was already identified on Windows.

Affected Files

- scripts/postProvision.sh — cleanup block at the end of the file