NFD » ndn-cxx

It is not obvious that disconnecting the handler can deallocate it, thus leading to a use-after-free crash. It is even less obvious when the handler is a lambda, because wrapping it into a std::function creates a copy, therefore the copy will be deallocated regardless of the lifetime of the original lambda closure. This effectively means that lambdas cannot be used as handlers if they need to disconnect themselves.

Possible solution:

In Signal<>::disconnect, if m_isExecuting is true, do not call m_slots.erase. In Signal<>::operator() delete line 198 and, after the handler invocation, add:

if (m_currentSlot == m_slots.end())
  it = m_slots.erase(it);
else
  ++it;

This should keep the closure alive until after the handler has finished executing.