Re: [Doh] [Ext] A question on the mix of DNS and HTTP semantics

[Andrew Sullivan <[e-mail]>]

On Mon, Mar 19, 2018 at 03:39:58PM +0000, Stephane Bortzmeyer wrote:
> Truncated for HTTP reasons (error 500 : the server crashed), not for
> DNS reasons.

That's what I assumed, but my point is that the rule could still be
that the client could be required to process it, because the client
would fail.  The truncated message would have a section count that was
higher than the records actually contained there (because of the
truncated message), so the client would have a partial message and
would eventually do whatever it would do in that case (in ordinary
DNS, time out).  

I think erro 500 would probably return faster, however.

> If the DNS answer is truncated for DNS reasons (EDNS buffer size too
> small), I would expect the HTTP status code to be 200.

Also ok with me.  But since it's a wire message, you still need a TC
bit set, right?  (Hard for the as-yet-unwritten cases where there are
no such header bits, though, so 200 is probably better.)

A

-- 
Andrew Sullivan
[e-mail]