On Tue, 8 Aug 2006, Paul Vixie wrote: > > 2.3.1
> as pekka said, the situation is more fluid than that. Understood. > > 2.3.4 > > > > Why is such a distinction needed? In other words, what is the > > advise given in this section for use of this distinction. > > so that we can refer to the distinction in 2.3.6. I guess I am somewhat confused here. 2.3.4 tells us when to set and when not to set the TC bit. 2.3.6 tells us: If any "necessary content" is silently truncated, then it is advisable that the TC bit be set I guess the confusing parts are the overlaps between "necessary content" and "additional section", eg when glue is vital for a zone to function. Because "necessary content" can be either just the authority section, or the authority section plus the additional section. But "2.1.4" says we should always treat the additional section as "possibly incomplete" if it contains glue, implying this can happen with the TC bit not being set. Shouldn't it just state that "necessary content" should not be silently truncated? Or if we add glue, why not just always set the TC bit, if we recommend that practise later on for nameservers anyway, instead of creating the state of "silent truncation"? Let me think of some extreme example, say example.com with 16 NS records ns[1..16].example.com, so lots of glue, with the cut somewhere in the additional section. Does the draft now advise to set the TC bit or not? And should nameservers always assume this is possibly incomplete and always do a TCP retry? > not at all. if any part of an rrset will not fit, then the rrset is > damaged. if the rrset is required (for example, to help a requester > determine that this is a referral rather than an empty answer) then > the TC bit would have to be set. once you're in TC-bit territory, > it doesn't matter what glue you put in or leave out. What is "required". Are 16 NS records and their glue "required". I believe you mean that to be the case (if the zoneadmin is silly enough to create it), while if I look from a resolving point of view, having less then 16 records is all I would "require", as long as there are a number of matching NS records with A* record glue. In which case, I think (but i am uncertain grounds here) that the TC bit should be set, but the nameserver might decide to try these records first, before doing a TCP retry to obtain the full records in case the original records all failed? Maybe the word "required" is a bit ambiguous here. (at least in my brain) Paul . dnsop resources:_____________________________________________________ web user interface: http://darkwing.uoregon.edu/~llynch/dnsop.html mhonarc archive: http://darkwing.uoregon.edu/~llynch/dnsop/index.html
