Re: Broken References (was: Cups)

[Hrvoje Niksic <hniksic@iskon.hr>]

Toby Speight <Toby.Speight@streapadair.freeserve.co.uk> writes:

> >> References: <m34sef90y1.fsf@quimbies.gnus.org> <sdg0xyh1vv.fsf@wanderer.dcn.davis.ca.us> <m3ln7nn0mq.fsf@quimbies.gnus.org> <86903n762l.fsf@megalith.bp.aventail.com> <vafu2mblylj.fsf@lucy.cs.uni-dortmund.de> <k8zd7sydabl.fsf@alcatel.de> <86wvr6v9sb.fsf@me
> 
> Gunnar> is that a known bug?
> 
> It's probably an MTA bug somewhere (or feature - I can't remember
> what the line-length limit is on headers).  Perhaps we should be
> able to tell Gnus that we're passing the message to a network that
> does this.

There are line-length limit suggestions, but they are much larger than
the above.  USEFOR and MESSFOR suggest 998 bytes if memory serves me.

Anyway, Gnus does limit References to around 1000 octets.
Furthermore, it does so intelligently, being careful not to cut the
line in the middle as in the above example.  Look at
message-shorten-references to see how it's done.

> In times gone past, Gnus used to wrap long References lines, but no
> longer does (in order to coddle broken readers).

Correction: it's the servers that are/were broken.  More precisely,
the servers didn't care, but the engine that creates NOV databases
wouldn't handle folded headers correctly.  This resulted in articles
produced by Gnus being threaded incorrectly despite their correct
contents and despite newsreaders being perfectly able to process them.

In the spirit of "be conservative in what you send," Gnus was modified
not to fold References by default, although folding them was clearly
the right thing.

Interoperability is sometimes _so_ sucky.