I certainly understood your desires....I was
extending it to what was originally proposed a year ago for combo testing within
Declude. I think most people using Pro have gone to filters to do this
instead, since it's easier that way. But those on Standard could use
bitmasking to achieve the same combo test results.
Darin.
----- Original Message -----
Sent: Friday, November 05, 2004 9:04 AM
Subject: Re: [Declude.JunkMail] ANN: SPAMC32 (SpamAssassin SPAMC for
Declude) 0.5.57 released
Darin,
Correct me if I'm wrong, but I was expecting that
this would only be internal to one external test at one time and have no effect
on anything else, i.e. DNSBL's. So the only limitation would be 32 result
codes for each external test which is workable. I would also imagine that
a different variable type could be used for a 'bitmask' type rather than a
'nonzero', 'weight' or 'external' type.
Matt
Darin
Cox wrote:
Certainly...I was thinking of it in the broader
sense, though. For example, we run more than 32 tests within Declude, so
it would only work for us if we culled the list down a bit, which we could
probably do quite easily with a lot of the DNSBLs that rarely get hit and are
almost always covered by others.
Also, I don't know for sure whether Scott or Pete
use unsigned 4-byte ints for the weights. Scott actually probably uses
signed ints, so you lose half of the bits...and if the weight is a 2-byte
signed int then the number of available bits drops to 15.
Darin.
-----
Original Message -----
Sent: Friday, November 05, 2004 8:41 AM
Subject: Re: [Declude.JunkMail] ANN: SPAMC32 (SpamAssassin SPAMC
for Declude) 0.5.57 released
I could deal with 32 result codes for a single test
:)
I'm hoping that Pete will weigh in on this. We had a
discussion once about how to weight multiple hits, and he was leaning towards
an internal probability based method, but this would give us far more
flexibility as administrators IMO.
Yesterday on my system Sniffer
returned 118,909 results (clean and failed), and of the 104,942 failed result
codes, there were a total of 316,206 result codes meaning an average of just
about 3 result codes for each time a message failed Sniffer. I was
careful not to double count the final result with each result
code.
Being able to get an average of 3 Sniffer hits per message would
allow me to reduce the weights slightly to protect from false positives, and
end up scoring spam with much higher weights as a result. This would
help my system immensely.
I could also use this for my own programming,
but enhancing Sniffer in this way would have broad implications across
Declude's customer base.
Matt
Darin Cox wrote:
This is the same idea I mentioned a year
ago when we were all talking about combo tests in Declude....only
problem being if you use more unique tests than the numeric type
supported. Assuming the weight/bitmask number is a 4-byte unsigned
int, then we have a maximum of 32 tests.
Darin.
-----
Original Message -----
Sent: Friday, November 05, 2004 7:35 AM
Subject: Re: [Declude.JunkMail] ANN: SPAMC32 (SpamAssassin SPAMC
for Declude) 0.5.57 released
If you don't mind me expanding on the bitmask idea....Sniffer
users would benefit from this greatly as many spams fail multiple Sniffer
tests. This would allow us to score each result code that it returned,
i.e.
SNIFFER-GENERAL
bitmask 1
"C:\IMail\Declude\Sniffer\execode.exe mycode"
6 0
SNIFFER-EXPERIMENTAL bitmask 2
"C:\IMail\Declude\Sniffer\execode.exe mycode"
6 0 SNIFFER-OBFUSCATION
bitmask 4
"C:\IMail\Declude\Sniffer\execode.exe mycode"
6 0 SNIFFER-IP
bitmask 8
"C:\IMail\Declude\Sniffer\execode.exe mycode"
4 0 SNIFFER-CASINO
bitmask 16 "C:\IMail\Declude\Snifferexecode.exe
mycode" 8 0
...
So if a test such as Sniffer returned a result code of 26, that
would mean it hit SNIFFER-CASINO, SNIFFER-IP and
SNIFFER-EXPERIMENTAL.
That would be huge
:)
Matt
Matt wrote:
Yes, I would
be interested in this very much since it would greatly ease the
management, testing and reporting of such tests, and I have been working
on something myself that would be capable of returning both positive and
negative weights and I didn't want to be running it twice to get the
separation in log lines.
Something else that is a bit OT regarding
external tests...I would be very interested in finding a way to run an
external test once and return multiple result codes, that way if you for
instance were testing different things that both required substantial code
and extra I/O, you could make things much more efficient and also greatly
simplify the management of your code. I understand of course that
you could create a set of 4 result codes to represent the combination of
two hits, but it quickly becomes unwieldy as it grows exponentially.
Is there a way that you could return multiple result codes and have
Declude fail multiple tests without running the test multiple times?
I'm thinking that something like a bitmask returned and then interpreted
by Declude to match zero to many tests.
http://www.joestump.net/170933118/a-quick-bitmask-howto-for-programmers
Note
that if this was available, I would probably prefer this over weight+ and
weight- for my own needs since I don't perceive being able to do both
:)
Thanks,
Matt
Markus Gufler wrote:
Yet another update to SPAMC32 that's useful when deployed as
a Declude 'weight' test type. See the release notes below
and download from the traditional /release folder.
As SpamChk is not anymore alone as external 'weight' test maybe also SPAMC32
users are interested in having 'weight+' and 'weight-'
So it would be possible to confgure two config lines one for a positive the
other for negative results.
For example
SPAMASSASSIN+ weight+ c:\imail\...
SPAMASSASSIN- weight- c:\imail\...
The benefits?
1.) It would become possible to use the results of weight tests for
combination filters.
Up to now it was not possible to assign extra points, for example if an
IP4R-test and SPAMCHK has failed.
As both tests are tecnicaly completely different the combination would be
highly accurate.
You can see this for example on http://www2.spamchk.com/public.html on the
already existing COMBO-... tests.
2.) Creating reports would be much easier and more clear if weight tests can
be separated like showed above.
I've suggested this some months ago to Scott. Maybe now with some additional
interested parties...
Markus
---
[This E-mail was scanned for viruses by Declude Virus (http://www.declude.com)]
---
This E-mail came from the Declude.JunkMail mailing list. To
unsubscribe, just send an E-mail to [EMAIL PROTECTED], and
type "unsubscribe Declude.JunkMail". The archives can be found
at http://www.mail-archive.com.
--
=====================================================
MailPure custom filters for Declude JunkMail Pro.
http://www.mailpure.com/software/
=====================================================
--
=====================================================
MailPure custom filters for Declude JunkMail Pro.
http://www.mailpure.com/software/
=====================================================
--
=====================================================
MailPure custom filters for Declude JunkMail Pro.
http://www.mailpure.com/software/
=====================================================
--
=====================================================
MailPure custom filters for Declude JunkMail Pro.
http://www.mailpure.com/software/
=====================================================
|