Re: `gnus-agent-expire' - specifics?

[Mike McEwan <mike@lotusland.demon.co.uk>]

Lars Magne Ingebrigtsen <larsi@gnus.org> writes:

> Oh, uhm, never.  I forgot about that.  *sigh*
> 
> Hm...  I'll have to think about how to approach this problem.

  Well I had a think about it, hacked a solution, and then rejected it
as null and void :-(. I was nuking the NOV lines of articles in the
overview file with dates older then `gnus-agent-expire-days' that did
*not* have corresponding article files in my news directories. This is
fine when the NOV line date is an accurate reflection of when the
article header was downloaded, but I guess there are too many
instances when such a date bears little relation to when the article
was actually posted, arrived on the news server, and was
downloaded. Duh, I've realised the necessity for the agent history
files.

  With this in view, I was wondering if perhaps the best way to get
rid of redundant NOV lines might be if an extra line of some sort was
written to the history file for each group, detailing the highest
article *header* downloaded in a given session. Then, at the tail end
of current gnus-agent expire processing, the history file could be
scanned for the last `group header details' line with a date that was
less than or equal to current time minus `gnus-expire-days' (No NOV
lines would be deleted if the first `group header details' was greater
than the match value etc. etc.). Any NOV lines with article numbers
less than or equal to the `highest retrieved header' number would then
be nuked.

  An alternative would be to have a separate `history' file for
downloaded headers, but then there's probably too many files already
eh. 

  Just an idea, I'm in no way proffering this as the way to go, it's
just that I've wasted some time on it :-). What's the deal with those
.agentview files? I been pruning my .overviews, but these suckers seem 
to keep on growing (I think...).

-- 
Mike.