Changes from rspamd 0.3 to rspamd 0.4
- Statistic of rspamd 0.4 uses new tokenization logic so old tokens are incompatible with rspamd 0.4, and there is no way to convert 0.3 statfiles to 0.4 ones. Sorry for this, but hashing used in rspamd is not revertible.
- Fuzzy hashes now skips unnecessary data (like urls) and uses double hashing to avoid border problems, so old hashes are incompatible with new rspamd.
- Rspamd now works in utf8 mode by default. Statistics, fuzzy hashes and other things are now working with utf8 unichars making statistics more accurate.
- Rspamd now can call pre-classifier function to select a set of statfiles for this message.
- SURBL can now has option noip to skip ip requests to this uribl (useful for dbl.spamhaus.org).
- Settings works now much faster by applying only once per message.
- Parts language detector (based on unicode script) was added.
- Difference between parts is now calculated more accurately using diff algorithm.
- "learn_spam" and "learn_ham" commands were added to automatic learning of classifiers
- Fixed bug with synchronization of statfiles: now master/slave synchronization is working
- Fixed bug with IO dispatcher that may cause sessions freezes.
- Fixed race condition in settings by adding reference counter.
- Many of minor bugs were fixed.
- Code style was improved by adding some more comments.
- Sample configuration file was adopted to the new version.