The test_ensure_python_fails_loudly_when_no_interpreter_can_import_agent
test was passing locally but failing on CI runners because:
1. CI runners don't have REPO_ROOT/.venv/bin/python on the filesystem
2. The function path on missing venv calls venv.EnvBuilder(with_pip=True).create()
3. That internally calls subprocess.check_output() — a different code path
than the monkey-patched bootstrap.subprocess.run, which only stubs run().
4. CI fails with: AttributeError: NoneType has no attribute stdout
The behavior under test is "what happens when no interpreter can import
both WebUI deps and the agent" — NOT the venv-creation path. So we sidestep
EnvBuilder by setting REPO_ROOT to tmp_path with a pre-existing
.venv/bin/python file. The venv-existence check passes, EnvBuilder is
skipped, the stubbed _python_can_run_webui_and_agent returns False on the
final check, and the expected RuntimeError fires.
Co-authored-by: ccqqlo <ccqqlo@users.noreply.github.com>
Hermes Bot