Release Notes: This version adds an option to merge two spam databases together, and "make test" now runs much more quickly.
Release Notes: HTML parsing has been improved and the minimum token length increased, which gives better matching in general. A bug in the "X-Spam-Token" header code was also fixed.
Release Notes: Logarithmic scaling of the spam probabilities has been implemented. Benchmarking and informational options have been added. A database-independent backend storage module has been written to remove the dependency on GDBM or Berkeley DB.
Release Notes: A bug in the mailbox parser causing incorrect message counts was fixed, and the shifting weighting code was corrected to get rid of the "core dump on --train" bug. The maximum token length was increased, and binary attachments are now dealt with by using their checksum as a token.
Release Notes: The training algorithm has been altered to use shifting weightings, which reduces training time. The "make test" procedure has been altered to use autogenerated "fake" mailboxes. The GNU getopt_long() code has been removed.
Release Notes: A bug in the --restore option that prevented restoration was fixed.
Release Notes: Options for adding a numerical rating header and pruning the database were added. The tokenizer now strips HTML in a sensible manner and preserves hostnames. Debian build support was added, and new wrappers for Berkeley DB remove the dependency on GDBM. Tokens are now hashed for privacy, so words from emails no longer appear in the database. All users will need to rebuild their databases using the "--train" facility after upgrading.
Release Notes: Support for using both system-wide and per-user spam databases in tandem has been added, plus improved MIME handling, better training, and some minor bugfixes.
Release Notes: A bug in Procmail, causing mail filtered as spam to be corrupted, has been worked around.
Release Notes: This version has a known bug when used in a procmail recipe with a lockfile; be sure to read the man page.