i think this applies to programming issues as well. the IETF used to strongly emphasize 'soft state' protocols. Not that there was no state, but it was always a

i think this applies to programming issues as well. the IETF used to strongly emphasize 'soft state' protocols. Not that there was no state, but it was always associated with a timeout and went away on its own. A best effort cache - although alot of these protocols (i.e pim) depend on having some non-zero lifetime to work at all. think about the additional bookkeeping and correctness issues associated with the destruction/cleanup path. alot of the time we maintain references and search structures so that we can 'properly clean up', when in reality we just want to tear down the whole world. construction and destruction do have a certain pleasant symmetry, but since the destruction path is often less tested and error prone, and may double the work at hand, isn't it worth considering implicit destruction?