🐛 fix(subprocess): drain pipes after killing timed-out process

On Windows, process.kill() terminates the process but doesn't close
inherited pipe handles. The reader threads spawned by communicate()
stay blocked on fh.read() indefinitely, preventing the caller from
ever returning. This was observed in tox CI where the 5s timeout
fired and killed the process, yet the test still hung.

Calling communicate() after kill() joins the reader threads and
closes the pipes, as required by the Python subprocess docs.

Author: gaborbernat

src/python_discovery/_cached_py_info.py

@@ -210,6 +210,7 @@ def _run_subprocess(
             code = process.returncode
         except TimeoutExpired:
             process.kill()
+            process.communicate()
             out, err, code = "", "timed out", -1
         except OSError as os_error:
             out, err, code = "", os_error.strerror, os_error.errno

tests/test_cached_py_info.py

@@ -113,13 +113,14 @@ def test_run_subprocess_with_cookies(mocker: MockerFixture) -> None:
 
 def test_run_subprocess_timeout(mocker: MockerFixture) -> None:
     mock_process = MagicMock()
-    mock_process.communicate.side_effect = TimeoutExpired(cmd="python", timeout=30)
+    mock_process.communicate.side_effect = [TimeoutExpired(cmd="python", timeout=30), ("", "")]
     mocker.patch("python_discovery._cached_py_info.Popen", return_value=mock_process)
     failure, result = _run_subprocess(PythonInfo, sys.executable, dict(os.environ))
     assert failure is not None
     assert "timed out" in str(failure)
     assert result is None
     mock_process.kill.assert_called_once()
+    assert mock_process.communicate.call_count == 2
 
 
 def test_run_subprocess_nonzero_exit(mocker: MockerFixture) -> None: