If you're going to control a relay with a transistor and will not include a flywheel diode, you're going to have a bad time, it's a known fact. But what about relay-relay circuits? I have mixed feelings about this:
When "driving" relay A is switching off another relay B, the inductive kickback of coil B would produce a spark in relay A contacts, thus, potentially reducing the overall life of relay A. So, we just include a flywheel diode on relay B and none on relay A. Well...what if A is also driven by another relay? (i.e. in a relay-based CPU? :))
So, if all relays have flywheel diodes, then we have a problem on "driving" or "controlling" relay A again: according to this article: http://www.te.com/commerce/DocumentDelivery/DDEController?Action=srchrtrv&DocNm=13C3264_AppNote&DocType=CS&DocLang=EN adding a flywheel diode would modify de-energizing profile of the coil, which modifies how the contact spring reacts to a switch off, which damages spring and reduces life of a relay.
Any comments? Drop them, use them, or use diode+zener? I think I may have to just make a pair of inverter loops (with and without diodes) and see which one dies first :)