Leaving aside the dependancy bug (which is not dnet-common's issue at all) - there's three distinct problems here, both of which definitely are :-
1/ The MAC address assigned in the default case is not unique (imagine the effect of a stock-with-defaults install of this package on a 1000-way compute cluster!) 2/ Burned-in-addresses are, for better or worse, increasingly being incorporated into link-layer authentication/ authorization schemes (and even, in certain cases, even encryption schemes) - particularly in the wireless space 3/ 802.1d uses address uniqueness to decide which bridge-port a given host is reachable-via (a certain hilarity will result if two stock-with-defaults install of this package are connected to networks either side of a learning-capable bridge - such as those commonly found in EVERY modern Ethernet switch fabric) Please - dnet-common needs to default to not changing anything at all (ie, please DON'T default to safety-off when handling shotguns loaded with live ammunition and aimed at feet); and if and/ or when it does - the default should not be static assignment (ie, the _whole world_ defaulting to the same DECNet address) but rather introduce at least some kind of fuzz (ie to reduce the probability of collisions within a given broadcast domain).