Re: [sup-devel] [PATCH] XapianIndex.each_message_in_thread_for yields messages in cronological order

[Tero Tilus <tero@tilus.net>]

William Morgan, 2009-12-31 21:41:
> Out of curiousity, Tero, could the problem also be solved by giving
> the in-reply-to header precedence over the references header?

Well, yes and no.  ;)

I think what it needs is to do is 

 a) consider only the first message in In-reply-to: (like it already
    does),
 b) prioritize In-reply-to: ahead of References: (like it already
    does!) and
 c) if In-reply-to: would create a loop or diamond, resolve by
    dropping another link ("topmost" conflicting?) and keep the one
    from In-reply-to: (currently it drops the link suggested by
    In-reply-to: over another potentially coming from (messed up)
    References:).

Lemme speculate on this a bit.

Current threading implementation tries to give In-reply-to: precedence
over References: but it still could leave the (in my previous mail
described way) malformed References: affecting the real root of the
thread.  By the time we encounter the In-reply-to: headers which would
need to take precedence over the References:, there could already be
bogus parent to the root.

Say we have

First (no In-reply-to: or References:)
 +- Second (In-reply-to: First; References: First)
     +- Third (In-reply-to: Second, First; no References:)
         +- Fourth (In-reply-to: Third; References: Second, First, Third)

If Third is a reply to both Second and First (in that order). Then
Fourth might have References: Second, First, Third.  If, when
threading, Fourth is the first processed message then First is seen as
a reply to Second.  Now when itself Second is processed, the
In-reply-to: in it would create a loop and is discarded, (see
ThreadSet#link).  Resulting in

Second
 +- First
     +- Third
         +- Fourth

which is exactly what the example headers I posted seem to produce
(real root jumps in the middle of one of the branches).

-- 
Tero Tilus ## 050 3635 235 ## http://tero.tilus.net/
_______________________________________________
Sup-devel mailing list
Sup-devel@rubyforge.org
http://rubyforge.org/mailman/listinfo/sup-devel