Re: Broken References (was: Cups)

[Hrvoje Niksic <hniksic@iskon.hr>]

Russ Allbery <rra@stanford.edu> writes:

> Hrvoje Niksic <hniksic@iskon.hr> writes:
> 
> > Except that wrapping doesn't preclude shortening.  When speaking of
> > header size limits, folded headers count as one line.  Or doesn't it?
> 
> Not according to the standards, and not in INN I'm fairly sure.

<check check>  You are right.  An oldish USEFOR draft says:

          NOTE: There is NO restriction on the number of lines into
          which a header may be split, and hence there is NO
          restriction on the total length of a header (in particular
          it may, by suitable folding, be made to exceed the 998
          octets restriction pertaining to a single header line).

However, References still needs to be trimmed, based on the number of
message-id's it contains:

          Followup Agents SHOULD NOT trim message ids out of the
          References content unless the number of message ids exceeds 31
          in which case message ids SHOULD be trimmed until there are
          only 31.

This means that once Gnus starts ignoring ancient INN's,
we'll be able to rewrite message-shorten-references to:

a) Always fold references nicely, as old Gnus did.

b) When such folded references contains more than 31 message-id, trim
   them until there is only 31.

In fact, we could code it that way even now, and provide an option
`message-cater-to-broken-inn' defaulting to t.  That way when the time
is ripe, we'll be able to change the default and, with the flip of a
lever, make the world a better place.

If Lars agrees, I can provide a patch.  Lars, what thinks you?