fix(server/debug): robust debugpy attach with bounded wait and tests update

server/src/_debug_server.py

@@ -6,7 +6,7 @@ import os
 import pathlib
 import runpy
 import sys
-import time
+import threading
 
 
 def update_sys_path(path_to_add: str) -> None:
@@ -27,21 +27,35 @@ def _debug_endpoint() -> tuple[str, int]:
     return host, port
 
 
-def _connect_debugger(debugpy, host: str, port: int, timeout: float = 15.0) -> None:
-    deadline = time.monotonic() + timeout
-    last_error: OSError | None = None
-    while time.monotonic() < deadline:
+def _connect_debugger(debugpy, host: str, port: int, timeout: float = 30.0) -> None:
+    errors: list[BaseException] = []
+
+    def attach() -> None:
+        # debugpy.connect() cannot be retried: after a refused connection a second
+        # call terminates the process silently. Connect exactly once.
         try:
             debugpy.connect((host, port))
             debugpy.wait_for_client()
-            return
-        except (ConnectionRefusedError, OSError) as error:
-            last_error = error
-            time.sleep(0.25)
+        except BaseException as error:  # pylint: disable=broad-exception-caught
+            errors.append(error)
 
-    raise RuntimeError(
-        f"Could not connect debugpy to {host}:{port} within {timeout:.0f} seconds"
-    ) from last_error
+    # A stale debugpy adapter from an earlier debug session can still own the port.
+    # It accepts the connection but never attaches, so bound the wait.
+    waiter = threading.Thread(target=attach, daemon=True)
+    waiter.start()
+    waiter.join(timeout)
+    if waiter.is_alive():
+        raise RuntimeError(
+            f"Connected to {host}:{port}, but no VS Code debug session attached within "
+            f"{timeout:.0f} seconds. A stale debugpy adapter probably still owns the "
+            f"port; stop it (e.g. 'fuser -k {port}/tcp') and restart debugging."
+        )
+    if errors:
+        raise RuntimeError(
+            f"No debugpy listener on {host}:{port}. Start the launch configuration "
+            "'Python debug server (hidden)' (e.g. via a 'Debug Extension and Python' "
+            "compound) before the language server."
+        ) from errors[0]
 
 
 def main() -> None:
@@ -61,7 +75,12 @@ def main() -> None:
 
     host, port = _debug_endpoint()
     print(f"debugpy: waiting for VS Code at {host}:{port}", file=sys.stderr)
-    _connect_debugger(debugpy, host, port)
+    try:
+        _connect_debugger(debugpy, host, port)
+    except RuntimeError as error:
+        print(f"debugpy: {error}", file=sys.stderr, flush=True)
+        # debugpy's background threads can keep the interpreter alive; exit hard.
+        os._exit(1)
     print("debugpy: VS Code attached; starting language server", file=sys.stderr)
 
     server_path = os.fspath(pathlib.Path(__file__).parent / "lsp_server.py")

server/tests/python_tests/test_debug_server.py

@@ -1,4 +1,5 @@
 import sys
+import threading
 from pathlib import Path
 
 import pytest
@@ -27,27 +28,46 @@ def test_debug_endpoint_rejects_invalid_port(monkeypatch, port):
         _debug_server._debug_endpoint()
 
 
-def test_connect_debugger_retries_until_adapter_is_ready(monkeypatch):
-    class FakeDebugpy:
-        def __init__(self):
-            self.connect_calls = 0
-            self.wait_calls = 0
+class FakeDebugpy:
+    def __init__(self, refuse=False, block_wait=False):
+        self.refuse = refuse
+        self.block_wait = block_wait
+        self.connect_calls = 0
+        self.wait_calls = 0
 
-        def connect(self, endpoint):
-            assert endpoint == ("127.0.0.1", 5678)
-            self.connect_calls += 1
-            if self.connect_calls < 3:
-                raise ConnectionRefusedError("listener is starting")
+    def connect(self, endpoint):
+        assert endpoint == ("127.0.0.1", 5678)
+        self.connect_calls += 1
+        if self.refuse:
+            raise ConnectionRefusedError("no listener")
 
-        def wait_for_client(self):
-            self.wait_calls += 1
+    def wait_for_client(self):
+        self.wait_calls += 1
+        if self.block_wait:
+            threading.Event().wait()
 
+
+def test_connect_debugger_connects_once_and_waits_for_client():
     fake_debugpy = FakeDebugpy()
-    monkeypatch.setattr(_debug_server.time, "sleep", lambda _seconds: None)
 
-    _debug_server._connect_debugger(
-        fake_debugpy, "127.0.0.1", 5678, timeout=1.0
-    )
+    _debug_server._connect_debugger(fake_debugpy, "127.0.0.1", 5678, timeout=1.0)
 
-    assert fake_debugpy.connect_calls == 3
+    assert fake_debugpy.connect_calls == 1
     assert fake_debugpy.wait_calls == 1
+
+
+def test_connect_debugger_does_not_retry_refused_connection():
+    # debugpy.connect() cannot be called a second time after a refused connection.
+    fake_debugpy = FakeDebugpy(refuse=True)
+
+    with pytest.raises(RuntimeError, match="No debugpy listener"):
+        _debug_server._connect_debugger(fake_debugpy, "127.0.0.1", 5678, timeout=1.0)
+
+    assert fake_debugpy.connect_calls == 1
+
+
+def test_connect_debugger_times_out_on_stale_adapter():
+    fake_debugpy = FakeDebugpy(block_wait=True)
+
+    with pytest.raises(RuntimeError, match="stale debugpy adapter"):
+        _debug_server._connect_debugger(fake_debugpy, "127.0.0.1", 5678, timeout=0.1)