[ietf-nntp] :bytes metadata

[Russ Allbery]

Clive D W Feather <clive at demon.net> writes:
> Russ Allbery said:

>> The only thing that it's used for in practice so far as I know is
>> killfiling and rough size estimates.  Do you have any reason to believe
>> that it's used for anything more important than that?

> You don't think it's used for space allocation, as I suggested?

No, I'm pretty sure it's not.  (That seems to have been backed up by the
one post you got in response to your message to news.software.nntp,
although it would have been good to get more feedback.)

> Perhaps. It feels wrong to me to have these under-specified features,
> given that we've made changes from historical practice (calling it
> ":bytes", for example). But it looks like I'm alone on this.

Oh, I absolutely agree that it's annoying to have under-specified
features, but one of the goals for OVER was to make it really easy for
existing implementations to also implement.  If we make it too hard for
clusters to implement, it will defeat a lot of the point.

It's definitely not ideal.  I agree that we should encourage servers to
make it an accurate byte count.  I just don't think that we can require it
or tell clients that they can rely on it.

> If you don't mind, I'll add a few words about what existing
> implementations do, then call it a day:

>    Note to client implementers: some existing servers return a value
>    different to that above. The commonest reasons for this are:
   
>    o  counting a CRLF pair as one octet;
   
>    o  including the "." character used for byte-stuffing in the number;
   
>    o  including the terminating "." CRLF in the number;
   
>    o  using one copy of an article for counting the octets but then
>       returning another one that differs in some (permitted) manner.
   
>    Implementations should be prepared for such variation (for example,
>    if allocating space to store a copy of the article).

This sounds like a good idea to me, although I do think it's fine to
encourage servers *not* to do this (I think that's still in there).

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>