Properly learned Spamhalter is very bipolar due used technology. Nearly all classifications have 0% or 100% result. Other results are only marginal or extreme cases.

Spamhalter is designed to decide if message is spam or not. It cannot produce result like 'it is maybe spam'. So, it have just one level - if classification is higher - message is spam. This configured level is essencial for corrections and training functions, becasue corrections and learning process working with this level value.

So, do not filter spam by your spam probability level. Allways use leve configured in Spamhalter. If classification is higher, then Spamhalter add "Spam detected" additional header what is designed for filtering purpose.

By this level Spamhalter knows itself, if message is spam or not, and this information is used for self-learning. If message is corrected, then this level is used for check, if correction is sucessfull or not.

If you are adding your individual levels by percentage instead of "spam detected" headers, then all of this cannot work properly. Please, use only configured spam level. Spamhalter is really sure if message is spam or not. There is not place for 'maybe'. And if human found error, then made correction immediately.
 

I know your post is old, but just wanted to add my voice that I think this would be a killer feature to be added to graywall.  Allowing any mail to the honeypot address address to get through.  That way those messages could be tagged as spam at least 15 minutes before they are sent to other users. 

The way it is now if a spam message is being sent to a bunch of users there is no guarantee that it will be sent to the honeypot address first.  It may in fact go to all the other users before ever being sent to the honeypot address.

Oh yes - I forgot about that small detail. BLAT is set to use our ISP's SMTP server for sending. I set the "outgoing" address as noreply@treverton.co.za and, of course, that address does not exist. I do see the SMTP server trying to deliver all of the failure messages to Mercury, but of course, they are never accepted.

The message that gets sent is:

Treverton has a very strict policy on unsolicited commercial email or SPAM.
Due to some wording or other part of your message to a user at treverton.co.za, your message was classified as SPAM and deleted.
You may resend your message with the code “jKx58Fq” appearing somewhere in the Subject line in order to ensure delivery.
You may also contact the IT Manager at postmaster@treverton.co.za, remembering to place the code “jKx58Fq" in the Subject line, and ask for your address to be whitelisted.

The other detail I omitted to mention: When my script parses a messages classified as SPAM but cannot find a local addressee in the headers, the message is saved in an "Orphans" folder. About 10 messages end up there every day, mostly "lottery winners" and "419" letters. Maybe 1 or 2 per day are send via a mailing list and I then have to go to the MercuryS logs and check the date and time and find out who the message was intended for. I then move the messages from the Orphans folder to the user's mail folder and add the sender to the whitelist.

> Yes I have but AJAX is not the only way to do things.

Right now it's the only way to avoid to reload an entire page every time you click somewhere. And it is done in JavaScript. And even if you don't use AJAX if you want a modern looking web page you will end up writing a lot of JavaScript. HTML itself is no longer enough.

> if there is no need for anything more than the stdin/stdout redirection then anything more is a moot point. 

Never said that CGI can't work - just that it's being phased out due to its shortcomings. You can code the application of tomorrow, or that of yesterday.

> Ok, I see you want to be picky.

No, I am just correct. That's science - you can be correct or wrong, not "picky". Therefore I have no misconception about what is needed and what is not.

> Now to address the subject of dll files. As of Windows 2000,  dll files once loaded stay in memory until the system is rebooted,

Again, not true.

> HKEY_LOCAL_MACHINE > Software > Microsoft > Windows > CurrentVersion > Explorer > AlwaysUnloadDLL

That's just avoid the "cache time" Windows Explorer uses when an application exits and its dll are unloaded - as long no other application uses them. That's to avoid to reload them from disk if the next app needs them. Ask yourself why this key is not the default if so useful...

>. You cannot tax 486 DX2 66 on a T1 with NT4 running, with todays computers....

Don't know where you live, but I can easily overwhelm a P4 with Windows 2003 with the bandwith and users available here.

> So you are implying that I cannot write a webmail system that is useful and should use an existing package. 

No, I am just implying that your foundations are probably wrong, or obsolete. Feel free to write your own, but I suggest you to compare yours with those already available, and ask yourself why someone should use yours instead of the others.

> Because I want to. 

Good luck. Bidirectional synchronization is always a nightmare, believe me. You are just reimplmenting the wheel, adding more complexity and the need of more resources, with no benefits. But you're the kind of guy who needs to hit the wall to acknowledge it's there.

Then feel free to insult me when you're unable to support you claims with facts.

Very useful, thanks. 

Perhaps using the existing add a block of text tool as a basis allow administrators to set up a logo and effectively company wide signature for all outgoing email.   In our case we would use the current disclaimer text along with company contact information and our logo.

The transaction filter on the HELO state works well if the announced name is the actual name.  Spammers very often insert something else that changes faster than I can change my socks.  This makes catching things like *adsl* not work all the time.  If Mercury did an RDNS and the transaction could home in on the RDNS result then this would defeat the spammers.  If when the RDNS finds that there is no name associated with the IP address, Mercury could insert, say "NULL" (as that is what gethostbyaddr returns), that could also be picked up by the transaction filter.

RDNS should be an option that can be turned on/off just in case there is too much of a performance hit for someone or is just not needed.

Mercury isn't free though. It is licensed software.

 http://community.pmail.com/pmail/MercuryPricing.aspx 

 As Thomas already stated, There is Squirrelmail for webmail, and with use of daemons  (even with out them) Mercury is already powerfull. Spamhalter, Greywall and Content Control and even the standard filtering Rules... and more...

I run Mercury as a service (win32) that starts up automatically when I start up the computer.
When starting, the startup screen always appears for couple of seconds.

It would be nice to have a parameter (for example mercury.exe -nosplash) that starts up Mercury without splash screen so that it feels as an "invisible" service when starting up the computer.

 
Comments:
Strangly if srvany.exe is used (part of Windows Resource Kit Tools) to create the effect of Mercury as nt service, there is no start-up screen when the parameter mercury.exe -m is used, but there is a minimized window of mercury in the taskbar that disappears when you click on it.

When the program FireDaemon () is used to let seem Mercury as a nt service, there is no minimized window but still a startup screen.

Sorry. I meant to post this in the Pegasus Wishlist!

I actually set mine to 1.  It prevents the hammering.  As for legit emails, a legit email will retry again some time after the blacklist clears.

Well, you can run the NLM version of Mercury handling mail delivery on a netware server and have a Windows server running Mercury/32 as an IMAP server in netware mode to read mail.

 You have to run Mercury/32 on windows though for this, even if it is just purely for IMAP, there is no way round that.

 David. 

For internal reasons (to do with the way Mercury manages the list files) it's not a trivial change to sort the subscriber list. It's a reasonably common request, but implementing it will take quite a bit of effort.

Cheers!

-- David --

Many thanks!

I'm trying to do this as well - I use popfile to classify mails into 5 separate categories, so I really want to use 5 folders inside a single user account.

Do you think a scheme like this work?

1) create a 'policy' that will run a program when mail arrives, like an antivirus scanner. This program must launch another process, then return control to mercury so it can finish delivering the mail

2) the process scans the mail headers to work out which user received mail. (optional, with only a few mailboxes you could just do them all)

3) the process waits a minute or two until mercury has finished delivering mail to the user's inbox

4) the process launches one of the perl scripts mentioned earlier, which then use IMAP to connect to the user's inbox and move any new mails it finds to the right folders.

Is there any way to make it more deterministic, i.e. remove the delay? For example, could a mercury daemon launch the perl script after the mail is delivered?

Thx, I'll try them

It's good to know someone is working on AD, but I'm still curious about a more generic LDAP version?

LDAP/OpenLDAP is arguably used in more places than AD.  Even in a Netware environment, LDAP is now the prefered method for connecting to an eDirectory tree from 3rd party apps...

We create the smtpauth.mer file with the same pairs as local mailbox name (= username) and the pop3 password. If your local mailboxes follow the syntax you should be able to have that combination  within smtpauth.mer.

However, we've found it to be limiting, narrowminded and confusing for the users to have local mailboxes equal the mail-address. It gets very hard to explain that a mailbox has mail addresses attached to it, and that you can have as many as you like connected with a mailbox. Also for our users when using the webmail interface our users can log in using either their email address or their mailbox name, and a conversion is made if needed into the pair of mailbox/password for proper pop and imap fetch.

At this point, you can't really write a Daemon that could have a significant impact on the way MercuryP operates - I haven't as yet opened up its innards in the way I have for MercuryS... GreyWall is a good example of the type of minute control that the new protocol-level event mechanism I have developed for Daemons allows, but I'm retrofitting the event mechanism on a module by module basis, and MercuryP hasn't been done yet (even for v4.5). It's on the list though.

That said, the sorting idea isn't a difficult option to add, so that seems like a reasonable compromise in the short term.

Cheers!

-- David --

Commentary on Point 1.
  The 30 char limit on email address in the Syslog (I and O records) will truncate this : Symantec_Mail_Security_for_SMTP@workorder.se
  Changing to variable-length records will also allow the addition of new fields more easily.

Commentary on Point 2 relating to mail stats.
   In trying to determine bandwidth usage per user, I have been examining the "I" and "O" records from the Syslog.
   2.1  Mail between Local users only generates an "I" record.
   2.2  A Local / Non-Local identifier on each address would be most useful for reporting (as indicated in the Core display)
          This to be able to split reporting between Local and Non-Local traffic.
   2.3  Inbound mail indicates the TO address as the "mailbox" name and not the "alias" (email address)
         Outbound mail shows the FROM address as the email address and not eh "mailbox" name.
         Correlating the two is not possible by looking at the data only - a matching table is required.