addMIMODevice() never assigned m_deviceSets.back()->m_deviceAPI before
dereferencing it via setSampleMIMO(), causing the very first
POST /sdrangel/deviceset?direction=2 against a fresh headless server
to SIGSEGV in DeviceAPI::setSampleMIMO()'s vtable lookup.
DeviceSet's constructor initializes m_deviceAPI to nullptr
(sdrbase/device/deviceset.cpp:38), and the sibling helpers
addSinkDevice() (line 283) and addSourceDevice() (line 323) both
assign the new DeviceAPI* into m_deviceSets.back()->m_deviceAPI
before any later dereference. addMIMODevice() omits this assignment,
so the line that today reads:
m_mainCore->m_deviceSets.back()->m_deviceAPI->setSampleMIMO(mimo);
dereferences nullptr.
Fix by performing the same assignment as the Sink/Source paths, just
before the createSampleMIMOPluginInstance() call.
Verified against v7.25.1 in a custom headless build
(cicada-sdrangelsrv:7.25.1-mimo) with a LimeSDR-USB:
- Pre-patch: POST /sdrangel/deviceset?direction=2 -> HTTP 202,
then immediate SIGSEGV (backtrace top frame
DeviceAPI::setSampleMIMO -> MainServer::addMIMODevice).
- Post-patch: POST -> HTTP 202, container healthy, follow-up
PUT /sdrangel/deviceset/0/device with hwType=LimeSDR
direction=2 -> HTTP 202, deviceset reports the LimeSDR
correctly bound as a MIMO device.
