At Thu, 23 Feb 2017 13:16:46 +0900, Lorenzo Colitti <lorenzo@xxxxxxxxxx> wrote: > Help he understand, then. There is widely-deployed code that assumes that > the interface ID is 64 and does not work on anything other than 64 bit > prefix lengths. Out of curiosity: which code is it, and exactly what does its assumption mean? Does it mean, for example, it allows manual configuration of an address but requires its IID length be 64 bits? (If it means something like this, I'd also wonder how its "IID" matters in the first place - does that implementation use the lower 64 bits for some specific purpose?) I thought general-purpose OSes are actually quite flexible about the length of "IIDs" (in many cases it's actually derived from the length of on-link prefix corresponding to the address). I know BSD variants are intentionally flexible on this. I also know at least some (if not all) kernel versions of Linux are flexible. The only example of "widely-deployed code" with such an assumption that I know of is ISC DHCPv6 client (though I'm not sure if its latest version still has that assumption) as described in Section 4.4 of RFC7421. But I guess you're referring to something else. If you can be more specific it might help provide some clarity for the discussion, although I'm not so optimistic that it will automagically resolve the conflicting views we are seeing. -- JINMEI, Tatuya