https://issues.apache.org/SpamAssassin/show_bug.cgi?id=7021
--- Comment #18 from Ivo Truxa <[email protected]> --- Another small change in the recently released version 1.0.11 of TxRep (available at https://github.com/truxoft/TxRep). The diff to the previous version is below. 1729a1730,1732 > if (!defined $pms->{relays_internal} && !defined $pms->{relays_external}) { > $pms->extract_message_metadata(); > } This is a change in the learn_message() hook sub. Since the permessagestatus is not available when TxRep is called through the learner, it is generated here, in the message_learn sub. The plain call of PerMsgStatus->new() does not parse the header metadata that are needed for some of the partial reputations (IP address, HELO name,...). In consequence, when the TxRep learning function is used (through sa-learn), only reputations for the plain email address and domain were stored in the reputation storage. EMAIL_IP and HELO reputations were not processed. To allow storing also the EMAIL_IP and HELO reputations, I now added the call to the extract_message_metadata() sub. Especially the EMAIL_IP is rather important, since (by default) it is the main marker. Additionally, the DKIM and SPF data are also not available in the permessagestatus in the learner, although they would be helpful for adjusting the right reputation records too. In this case, though, I am hesitating whether to add the calls to the respective DKIM/SPF functions or not. The DKIM and SPF calls may be relatively expensive (they may involve DNS calls), so they would negatively influence the learner performance. That may be especially noticeable when learning huge number of messages through sa-learn. So perhaps allowing the DKIM and SPF parsing in the learner only optionally might be wiser. Anyway, this is currently no implemented, so there will be (probably) another update later. -- You are receiving this mail because: You are the assignee for the bug.
