Re: [Pan-users] Scoring based on arbitrary headers?

[Duncan]

Jim Henderson posted on Sat, 10 Jan 2015 21:08:27 +0000 as excerpted:

> That did it.  It was the previous absolute score that was causing the
> issue.  In fact, as soon as I removed it from the scoring window, the
> new score showed up.
> 
> Awesome help, thanks, Duncan. :)

=:^)

And as often the case, once I got testing I learned more about it myself 
too, so thanks for the learning stimulus! =:^)


Years ago, I was I think the first or one of the first to request the 
"arbitrary-header" scoring.  That was definitely before the C++ rewrite 
that became 0.90, and may well have been before pan even had actual 
scoring -- when it only had the trinary kill/normal/watch and filters 
that could only kill or watch.

I remember well Charles explaining the overviews issue back then.  My 
response was as I explained here, that while having to download first 
wasn't ideal, just the fact that I could then automate the killfiling of 
one deliberately nymshifting troll with one identifying header that 
unfortunately wasn't an overview header, instead of having to manually 
delete those posts, even if the posts had to be downloaded in ordered to 
be killfiled, would still be a bonus over the manual deletion I was 
having to do at that point.

But Charles marked the enhancement as target "bluesky", which means OK, 
it'd be nice to have, and if someone codes it up I'll likely take it, but 
don't expect any code from me (author/lead-dev, Charles in this case) 
implementing it any time soon.

And because I'm not a coder I couldn't make it happen no matter what I 
wished. =:^(

At that's exactly where it remained thru the rewrite until the end of the 
Charles era and thru the abandoned era.

Then Heinrich picked up development and that was one of several features 
that he implemented that Charles never got around to.  I guess binary 
posting was probably the big one for many, and score-based automatic 
actions was the big thing for others, but while my specific need that had 
generated the original request all those years ago was long gone, I'm 
still very much of the opinion that a news client that can't handle 
arbitrary-header scoring, even if it does require downloading first in 
ordered to do, is missing an important feature.  And I'm very glad pan 
isn't missing it any more, because there **ARE** nymshifting trolls out 
there, and IMO a news client that lacks the tools to deal with that is 
lacking an important features indeed!


Meanwhile, while I was prepared for the hit of having to download the 
messages, the big negative surprise of the feature for me was the effect 
it had on group-switch time due to having to rescan all those cached 
messages.  Of course that's primarily because I'm using pan as a news 
archiver and have over a gig of text messages distributed between my 
active groups, for pan to scan.  Were I to use it normally and expire 
headers in some reasonable time while actually caching far fewer than 
that, that problem wouldn't be as bad.

I've actually been thinking about splitting my pan text instance in half, 
making it two separate instances, a text-archive instance I can open if I 
want to look something up, pointed at a $PAN_HOME that can actually be on 
a partition I don't keep mounted all the time, and a new text-current 
instance with a much smaller cache, that expires messages in say 60-180 
days (maybe set the expiry to 180 days for headers/overviews, and set a 
cache size that gives me about 60 days worth of messages actually 
cached).  That would be small enough that I should be able to run 
arbitrary-header scores if I wanted to without taking unduly long to 
switch groups.

It'd also let pan startup far faster than it does now -- I actually have 
pan set to start with kde and keep it running basically aways, because 
with that many messages cached it takes too long to start, especially the 
first time after boot when the storage cache (even SSD!) is cold.

But the idea isn't fully formed in my head yet.  Ideally I'd have the 
cached messages from text-current copy automatically to text-archive 
before the cache rotated them out, so the archive instance could stay 
updated without actually having to download all those messages over 
again, but I have to figure out how I'm going to implement that and then 
actually do it...

So it might be awhile or I might never do it.  But I'm thinking about it, 
and if I did do it, that'd make arbitrary-header scoring much more 
workable for me, as I could actually change groups in reasonable time 
again, that being the problem that had me turning off arbitrary-header 
scoring so fast after the test this time.


Another thing that could make a difference but that I didn't test, would 
be strictly limiting arbitrary-header scoring to specific groups.  I'm 
guessing that would limit the group-switch time penalty to only those 
groups where I had arbitrary-header scores active.  This is something 
most folks could do pretty easily, as most folks won't need to arbitrary-
header score on /all/ their groups, only specific groups with specific 
issues, like that nymshifting troll I wanted to killfile that triggered 
the initial feature request all those years ago, or the specific 
forwarded-for watch it appears you are doing.


Hmm... That was /supposed/ to be a quick ack-N-thanks reply, basically 
the smiley and first sentence/paragraph, but it didn't turn out that way! 
=:^0

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman