I have noticed some problem in the cppdb branch (at least) in
the inband code.
when the usrp instances are destructed, the vector holding the
instances is destructed as well, but there seems to be some problem
destruction in the right order. Symptoms are: a) deadlock in the inband
examples, b) gdb stopping in siglongjmp, c) valgrind reporting accesses
already freed memory.
The attached small patch fixes this problem.