# Copyright (C) 2001-2003 by the Free Software Foundation, Inc.
#
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License
# as published by the Free Software Foundation; either version 2
# of the License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
import re
from Mailman import mm_cfg
from Mailman import Utils
from Mailman.i18n import _
from Mailman.Logging.Syslog import syslog
from Mailman.Gui.GUIBase import GUIBase
try:
True, False
except NameError:
True = 1
False = 0
class Topics(GUIBase):
def GetConfigCategory(self):
return 'topics', _('Topics')
def GetConfigInfo(self, mlist, category, subcat=None):
if category <> 'topics':
return None
WIDTH = mm_cfg.TEXTFIELDWIDTH
return [
_('List topic keywords'),
('topics_enabled', mm_cfg.Radio, (_('Disabled'), _('Enabled')), 0,
_('''Should the topic filter be enabled or disabled?'''),
_("""The topic filter categorizes each incoming email message
according to <a
href="http://www.python.org/doc/current/lib/module-re.html">regular
expression filters</a> you specify below. If the message's
<code>Subject:</code> or <code>Keywords:</code> header contains a
match against a topic filter, the message is logically placed
into a topic <em>bucket</em>. Each user can then choose to only
receive messages from the mailing list for a particular topic
bucket (or buckets). Any message not categorized in a topic
bucket registered with the user is not delivered to the list.
<p>Note that this feature only works with regular delivery, not
digest delivery.
<p>The body of the message can also be optionally scanned for
<code>Subject:</code> and <code>Keywords:</code> headers, as
specified by the <a
href="?VARHELP=topics/topics_bodylines_limit">topics_bodylines_limit</a>
configuration variable.""")),
('topics_bodylines_limit', mm_cfg.Number, 5, 0,
_('How many body lines should the topic matcher scan?'),
_("""The topic matcher will scan this many lines of the message
body looking for topic keyword matches. Body scanning stops when
either this many lines have been looked at, or a non-header-like
body line is encountered. By setting this value to zero, no body
lines will be scanned (i.e. only the <code>Keywords:</code> and
<code>Subject:</code> headers will be scanned). By setting this
value to a negative number, then all body lines will be scanned
until a non-header-like line is encountered.
""")),
('topics', mm_cfg.Topics, 0, 0,
_('Topic keywords, one per line, to match against each message.'),
_("""Each topic keyword is actually a regular expression, which is
matched against certain parts of a mail message, specifically the
<code>Keywords:</code> and <code>Subject:</code> message headers.
Note that the first few lines of the body of the message can also
contain a <code>Keywords:</code> and <code>Subject:</code>
"header" on which matching is also performed.""")),
]
def handleForm(self, mlist, category, subcat, cgidata, doc):
topics = []
# We start i at 1 and keep going until we no longer find items keyed
# with the marked tags.
i = 1
while True:
deltag = 'topic_delete_%02d' % i
boxtag = 'topic_box_%02d' % i
reboxtag = 'topic_rebox_%02d' % i
desctag = 'topic_desc_%02d' % i
wheretag = 'topic_where_%02d' % i
addtag = 'topic_add_%02d' % i
newtag = 'topic_new_%02d' % i
i += 1
# Was this a delete? If so, we can just ignore this entry
if cgidata.has_key(deltag):
continue
# Get the data for the current box
name = cgidata.getvalue(boxtag)
pattern = cgidata.getvalue(reboxtag)
desc = cgidata.getvalue(desctag)
if name is None:
# We came to the end of the boxes
break
if cgidata.has_key(newtag) and (not name or not pattern):
# This new entry is incomplete.
doc.addError(_("""Topic specifications require both a name and
a pattern. Incomplete topics will be ignored."""))
continue
# Make sure the pattern was a legal regular expression
name = Utils.websafe(name)
try:
re.compile(pattern)
except (re.error, TypeError):
safepattern = Utils.websafe(pattern)
doc.addError(_("""The topic pattern '%(safepattern)s' is not a
legal regular expression. It will be discarded."""))
continue
# Was this an add item?
if cgidata.has_key(addtag):
# Where should the new one be added?
where = cgidata.getvalue(wheretag)
if where == 'before':
# Add a new empty topics box before the current one
topics.append(('', '', '', True))
topics.append((name, pattern, desc, False))
# Default is to add it after...
else:
topics.append((name, pattern, desc, False))
topics.append(('', '', '', True))
# Otherwise, just retain this one in the list
else:
topics.append((name, pattern, desc, False))
# Add these topics to the mailing list object, and deal with other
# options.
mlist.topics = topics
try:
mlist.topics_enabled = int(cgidata.getvalue(
'topics_enabled',
mlist.topics_enabled))
except ValueError:
# BAW: should really print a warning
pass
try:
mlist.topics_bodylines_limit = int(cgidata.getvalue(
'topics_bodylines_limit',
mlist.topics_bodylines_limit))
except ValueError:
# BAW: should really print a warning
pass