From 67aac4d863db4e1b5a462672d4cb6bdf7a3196ef Mon Sep 17 00:00:00 2001
From: Teknium <127238744+teknium1@users.noreply.github.com>
Date: Sat, 22 Aug 2026 11:55:19 -0700
Subject: [PATCH] fix(gateway): control-socket fallback survives deep TMPDIR;
 tests bind-location-aware

CI runners put pytest tmp roots past sun_path, which routed every test
home through the fallback: two tests assumed in-home binding. Tests now
assert against the resolved bind location, and the fallback itself
prefers /tmp when tempfile.gettempdir() is too deep to fit sun_path.
---
 gateway/control_socket.py            | 17 +++++++++++++++--
 tests/gateway/test_control_socket.py | 28 +++++++++++++++++++++++-----
 2 files changed, 38 insertions(+), 7 deletions(-)

diff --git a/gateway/control_socket.py b/gateway/control_socket.py
index 9f07faa4d3..3962de58a5 100644
--- a/gateway/control_socket.py
+++ b/gateway/control_socket.py
@@ -96,8 +96,21 @@ def _default_socket_path(home: Path) -> Path:
 
 
 def _fallback_socket_path(home: Path) -> Path:
-    """Short temp-dir path for homes whose direct socket path exceeds sun_path."""
-    return Path(tempfile.gettempdir()) / f"hermes-gw-{_home_hash(home)}.sock"
+    """Short temp-dir path for homes whose direct socket path exceeds sun_path.
+
+    Prefers ``tempfile.gettempdir()``; when even that yields a too-long path
+    (deep $TMPDIR), falls back to ``/tmp`` on POSIX. If nothing fits, the
+    tempdir candidate is returned anyway — bind will fail non-fatally and
+    consumers use the scan layer.
+    """
+    name = f"hermes-gw-{_home_hash(home)}.sock"
+    candidates = [Path(tempfile.gettempdir()) / name]
+    if not _IS_WINDOWS:
+        candidates.append(Path("/tmp") / name)
+    for candidate in candidates:
+        if len(str(candidate).encode("utf-8")) <= _MAX_UNIX_PATH:
+            return candidate
+    return candidates[0]
 
 
 def resolve_server_socket_path(home: Path) -> tuple[Path, Optional[Path]]:
diff --git a/tests/gateway/test_control_socket.py b/tests/gateway/test_control_socket.py
index b1cbe156db..2e8a7bb1df 100644
--- a/tests/gateway/test_control_socket.py
+++ b/tests/gateway/test_control_socket.py
@@ -44,10 +44,24 @@ def _serve(home: Path, handlers=None):
 # Path resolution
 # ---------------------------------------------------------------------------
 
-def test_short_home_binds_in_home(home: Path):
-    bind, pointer = resolve_server_socket_path(home)
-    assert bind == home / "gateway.sock"
-    assert pointer is None
+def test_short_home_binds_in_home(tmp_path: Path):
+    # A home short enough for sun_path binds in-home with no pointer.
+    # tmp_path can exceed the limit on CI runners, so build one in the
+    # system temp root directly.
+    import tempfile
+
+    short_root = Path(tempfile.mkdtemp(prefix="hgw-", dir="/tmp"))
+    try:
+        short_home = short_root / ".hermes"
+        short_home.mkdir()
+        assert len(str(short_home / "gateway.sock").encode()) <= 100
+        bind, pointer = resolve_server_socket_path(short_home)
+        assert bind == short_home / "gateway.sock"
+        assert pointer is None
+    finally:
+        import shutil
+
+        shutil.rmtree(short_root, ignore_errors=True)
 
 
 def test_long_home_uses_pointer_fallback(tmp_path: Path):
@@ -163,7 +177,11 @@ def test_stop_removes_socket_and_pointer(home: Path):
 
 
 def test_stale_socket_file_is_replaced_on_bind(home: Path):
-    (home / "gateway.sock").touch()  # crashed predecessor's leftover
+    # Plant the stale file at wherever the server will actually bind
+    # (in-home OR the temp-dir fallback, depending on path length).
+    bind, _ = resolve_server_socket_path(home)
+    bind.parent.mkdir(parents=True, exist_ok=True)
+    bind.touch()  # crashed predecessor's leftover
 
     async def scenario():
         server = GatewayControlServer(