Loading
Give the valgrind memory test a longer context timeout
Nearly all of a device server's shutdown happens after its last Python statement, in Py_FinalizeEx and the C++ static destructors. Valgrind inflates that well beyond the test context's default process timeout, so the context declared the server stuck before it had finished exiting, and the leak checks never ran. The margin was already almost exhausted, and Python 3.15 — whose finalization costs roughly a quarter more — turned a latent flake into a deterministic failure. The valgrind pass of the memory test now uses an explicit, generous timeout; the plain RSS and block-count passes keep the default.