def _get_message_keys(self, messageid):
keys = self.key_cache.get(messageid, [])
if not keys:
email = Email(self._idx(), messageid)
# First we check the Autocrypt headers
msg = email.get_msg(pgpmime='all')
for ach in ([extract_autocrypt_header(msg)] +
extract_autocrypt_gossip_headers(msg)):
if 'keydata' in ach:
for keydata in get_keydata(ach['keydata'],
autocrypt_header=ach,
include_subkeys=False):
keys.append((keydata, ach['keydata']))
# Then go looking at the attachments
attachments = email.get_message_tree(want=["attachments"]
)["attachments"]
for part in attachments:
if len(keys) > 100: # Just to set some limit...
break
if _might_be_pgp_key(part["filename"], part["mimetype"]):
key = part["part"].get_payload(None, True)
for keydata in get_keydata(key, include_subkeys=False):
keys.append((keydata, key))
self.key_cache[messageid] = keys
return keys
def _get_message_keys(self,
messageid,
autocrypt=True,
autocrypt_gossip=True,
attachments=True):
keys = self.key_cache.get(messageid, [])
if not keys:
email = Email(self._idx(), messageid)
# First we check the Autocrypt headers
loop_count = 0
msg = email.get_msg(pgpmime='all')
ac_headers = []
if autocrypt:
ac_headers.append(extract_autocrypt_header(msg))
if autocrypt_gossip:
ac_headers.extend(extract_autocrypt_gossip_headers(msg))
for ach in ac_headers:
loop_count += 1
if 'keydata' in ach:
for keyinfo in get_keyinfo(ach['keydata'],
autocrypt_header=ach,
key_info_class=MailpileKeyInfo):
keyinfo.is_autocrypt = True
keyinfo.is_gossip = (loop_count > 1)
keys.append((keyinfo, ach['keydata']))
# Then go looking at the attachments
atts = []
if attachments:
atts.extend(
email.get_message_tree(
want=["attachments"])["attachments"])
for part in atts:
if len(keys) > 100: # Just to set some limit...
break
if _might_be_pgp_key(part["filename"], part["mimetype"]):
key = part["part"].get_payload(None, True)
for keyinfo in get_keyinfo(key,
key_info_class=MailpileKeyInfo):
keys.append((keyinfo, key))
self.key_cache[messageid] = keys
return keys
def _retrain(self, tags=None):
"Retrain autotaggers"
session, config, idx = self.session, self.session.config, self._idx()
tags = tags or [asb.match_tag for asb in autotag_configs(config)]
tids = [config.get_tag(t)._key for t in tags if t]
session.ui.mark(_('Retraining SpamBayes autotaggers'))
if not config.real_hasattr('autotag'):
config.real_setattr('autotag', {})
# Find all the interesting messages! We don't look in the trash,
# but we do look at interesting spam.
#
# Note: By specifically stating that we DON'T want trash, we
# disable the search engine's default result suppression
# and guarantee these results don't corrupt the somewhat
# lame/broken result cache.
#
no_trash = ['-in:%s' % t._key for t in config.get_tags(type='trash')]
interest = {}
for ttype in ('replied', 'read', 'tagged'):
interest[ttype] = set()
for tag in config.get_tags(type=ttype):
interest[ttype] |= idx.search(session, ['in:%s' % tag.slug] +
no_trash).as_set()
session.ui.notify(
_('Have %d interesting %s messages') %
(len(interest[ttype]), ttype))
retrained, unreadable = [], []
count_all = 0
for at_config in autotag_configs(config):
at_tag = config.get_tag(at_config.match_tag)
if at_tag and at_tag._key in tids:
session.ui.mark('Retraining: %s' % at_tag.name)
yn = [(set(), set(), 'in:%s' % at_tag.slug, True),
(set(), set(), '-in:%s' % at_tag.slug, False)]
# Get the current message sets: tagged and untagged messages
# excluding trash.
for tset, mset, srch, which in yn:
mset |= idx.search(session, [srch] + no_trash).as_set()
# If we have any exclude_tags, they are particularly
# interesting, so we'll look at them first.
interesting = []
for etagid in at_config.exclude_tags:
etag = config.get_tag(etagid)
if etag._key not in interest:
srch = ['in:%s' % etag._key] + no_trash
interest[etag._key] = idx.search(session,
srch).as_set()
interesting.append(etag._key)
interesting.extend(['replied', 'read', 'tagged', None])
# Go through the interest types in order of preference and
# while we still lack training data, add to the training set.
for ttype in interesting:
for tset, mset, srch, which in yn:
# False positives are really annoying, and generally
# speaking any autotagged subset should be a small
# part of the Universe. So we divide the corpus
# budget 33% True, 67% False.
full_size = int(at_config.corpus_size *
(0.33 if which else 0.67))
want = min(full_size // len(interesting),
max(0, full_size - len(tset)))
# Make sure we always fully utilize our budget
if full_size > len(tset) and not ttype:
want = full_size - len(tset)
if want:
if ttype:
adding = sorted(list(mset & interest[ttype]))
else:
adding = sorted(list(mset))
adding = set(list(reversed(adding))[:want])
tset |= adding
mset -= adding
# Load classifier, reset
atagger = config.load_auto_tagger(at_config)
atagger.reset(at_config)
for tset, mset, srch, which in yn:
count = 0
# We go through the list of message in order, to avoid
# thrashing caches too badly.
for msg_idx in sorted(list(tset)):
try:
e = Email(idx, msg_idx)
count += 1
count_all += 1
session.ui.mark(
_('Reading %s (%d/%d, %s=%s)') %
(e.msg_mid(), count, len(tset), at_tag.name,
which))
atagger.learn(at_config, e.get_msg(),
self._get_keywords(e), which)
play_nice_with_threads()
if mailpile.util.QUITTING:
return self._error('Aborted')
except (IndexError, TypeError, ValueError, OSError,
IOError):
if 'autotag' in session.config.sys.debug:
import traceback
traceback.print_exc()
unreadable.append(msg_idx)
session.ui.warning(
_('Failed to process message at =%s') %
(b36(msg_idx)))
# We got this far without crashing, so save the result.
config.save_auto_tagger(at_config)
retrained.append(at_tag.name)
message = _('Retrained SpamBayes auto-tagging for %s') % ', '.join(
retrained)
session.ui.mark(message)
return self._success(message,
result={
'retrained': retrained,
'unreadable': unreadable,
'read_messages': count_all
})
def _retrain(self, tags=None):
"Retrain autotaggers"
session, config, idx = self.session, self.session.config, self._idx()
tags = tags or [asb.match_tag for asb in autotag_configs(config)]
tids = [config.get_tag(t)._key for t in tags if t]
session.ui.mark(_('Retraining SpamBayes autotaggers'))
if not config.real_hasattr('autotag'):
config.real_setattr('autotag', {})
# Find all the interesting messages! We don't look in the trash,
# but we do look at interesting spam.
#
# Note: By specifically stating that we DON'T want trash, we
# disable the search engine's default result suppression
# and guarantee these results don't corrupt the somewhat
# lame/broken result cache.
#
no_trash = ['-in:%s' % t._key for t in config.get_tags(type='trash')]
interest = {}
for ttype in ('replied', 'read', 'tagged'):
interest[ttype] = set()
for tag in config.get_tags(type=ttype):
interest[ttype] |= idx.search(session,
['in:%s' % tag.slug] + no_trash
).as_set()
session.ui.notify(_('Have %d interesting %s messages'
) % (len(interest[ttype]), ttype))
retrained, unreadable = [], []
count_all = 0
for at_config in autotag_configs(config):
at_tag = config.get_tag(at_config.match_tag)
if at_tag and at_tag._key in tids:
session.ui.mark('Retraining: %s' % at_tag.name)
yn = [(set(), set(), 'in:%s' % at_tag.slug, True),
(set(), set(), '-in:%s' % at_tag.slug, False)]
# Get the current message sets: tagged and untagged messages
# excluding trash.
for tset, mset, srch, which in yn:
mset |= idx.search(session, [srch] + no_trash).as_set()
# If we have any exclude_tags, they are particularly
# interesting, so we'll look at them first.
interesting = []
for etagid in at_config.exclude_tags:
etag = config.get_tag(etagid)
if etag._key not in interest:
srch = ['in:%s' % etag._key] + no_trash
interest[etag._key] = idx.search(session, srch
).as_set()
interesting.append(etag._key)
interesting.extend(['replied', 'read', 'tagged', None])
# Go through the interest types in order of preference and
# while we still lack training data, add to the training set.
for ttype in interesting:
for tset, mset, srch, which in yn:
# False positives are really annoying, and generally
# speaking any autotagged subset should be a small
# part of the Universe. So we divide the corpus
# budget 33% True, 67% False.
full_size = int(at_config.corpus_size *
(0.33 if which else 0.67))
want = min(full_size // len(interesting),
max(0, full_size - len(tset)))
# Make sure we always fully utilize our budget
if full_size > len(tset) and not ttype:
want = full_size - len(tset)
if want:
if ttype:
adding = sorted(list(mset & interest[ttype]))
else:
adding = sorted(list(mset))
adding = set(list(reversed(adding))[:want])
tset |= adding
mset -= adding
# Load classifier, reset
atagger = config.load_auto_tagger(at_config)
atagger.reset(at_config)
for tset, mset, srch, which in yn:
count = 0
# We go through the list of message in order, to avoid
# thrashing caches too badly.
for msg_idx in sorted(list(tset)):
try:
e = Email(idx, msg_idx)
count += 1
count_all += 1
session.ui.mark(
_('Reading %s (%d/%d, %s=%s)'
) % (e.msg_mid(), count, len(tset),
at_tag.name, which))
atagger.learn(at_config,
e.get_msg(),
self._get_keywords(e),
which)
play_nice_with_threads()
if mailpile.util.QUITTING:
return self._error('Aborted')
except (IndexError, TypeError, ValueError,
OSError, IOError):
if 'autotag' in session.config.sys.debug:
import traceback
traceback.print_exc()
unreadable.append(msg_idx)
session.ui.warning(
_('Failed to process message at =%s'
) % (b36(msg_idx)))
# We got this far without crashing, so save the result.
config.save_auto_tagger(at_config)
retrained.append(at_tag.name)
message = _('Retrained SpamBayes auto-tagging for %s'
) % ', '.join(retrained)
session.ui.mark(message)
return self._success(message, result={
'retrained': retrained,
'unreadable': unreadable,
'read_messages': count_all
})
