# $Id: TODO 6705 2008-03-19 11:20:40Z m-a $

			bogofilter TODO list

**** If insufficient data is present and the default "undecided"
     bogosity is added in -p mode, add also a comment stating that
     bogofilter needs more training first

**** Add a "reservation lock" (fcntl style on separate file) so that a
     writer can prevent new readers from starting, so that busy scoring
     systems don't starve registration processes. (Figure out the
     details to avoid deadlock.)

**** Drop/fix MAXTOKENLEN: where it is an allocation, it must die.
     Where it is a character limit, count characters, not octets, to
     support UTF-8.

**** Implement Concurrent Data store, quite similar to Transactional.

**** Make sure that RFC-2047 decoder runs only once, not recursively.

**** Implement RFC-2046 section 5.2.2 (message/partial reassembly rules,
     Take most headers from enclosing message except Content-*, Subject,
     Message-ID, Encrypted, MIME-Version, which are taken from the
     enclosed message).

**** Reimplement seeking passthrough mode that got dropped on 2003-08-23
     with the switch to bogoreader.*
     http://article.gmane.org/gmane.mail.bogofilter.general/9035 and
     followups. (MID <20041222105734.GA30574@sela.f4n.org>, by "John"
     Subject "Size limit?" on 2004-12-22)
     The fseek() code to determine if the input is seekable got removed
     when the reader moved out of main.c between 1.66 and 1.67 (CVS) and
     has never been in bogoreader.c.

**** New Feature: Token aging. Support for struct data in the wordlists is
already present.

**** New feature: Token merging, based on delta tokens (Andras Salamon,
     andras@dns.net on bogofilter-dev, 2005-01-25)

**** Two deletes for kmail?  This wouldn't be a patch for bogofilter
itself, but a change to give kmail delete-as-spam and delete-as-
nonspam buttons.  Similarly for other MUAs.

**** New Feature: Make it a milter?

**** New Feature: Multiple list file support with weights and rules. Wordlist verfification.
Eric Seppanen:
> Allow use of a variable number of list files, each with their
> own weights and rules.
> Possible uses:
> - hand-maintained "whitelist" or "blacklist" files, with massive 
> weighting to override everything else.
> - allow users to use system-wide list files and their own files.
> 
Shared-database version based on the autodaemon code,
In the shared-database version (which doesn't yet exist) worldlist
verification to avoid attacks on posters (thanks, Barry!).
Emulate the Vipul's Razor reputation scheme for people reporting tokens?
http://razor.sourceforge.net/

**** What this software is probably heading towards is a scheme in which
there's a general notion of tagged categories (spam being one) with
cluster analysis being applied to categorize which categories a
message belongs to at above 0.9 confidence level.

**** New Feature: Web based tool for wordlist management. Allow message
registration and whitelist management. HTML Templatized for easy integration
with existing web mail systems.

**** New Feature: Add support for a user configurable list of headers that 
should be used to ignore (single or multi-line) headers that appear in the 
list. The list should be used to ignore headers both during the message 
registration and evaluation procedures.