genericize chain remover to use it for bridge and addressed chains

will be used in a future change to clean up markov chains

ircbot/migrations/0019_ircchannel_discord_bridge.py

@@ -0,0 +1,18 @@
+# Generated by Django 3.2.18 on 2023-02-16 22:38
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('ircbot', '0018_ircserver_replace_irc_control_with_markdown'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='ircchannel',
+            name='discord_bridge',
+            field=models.CharField(blank=True, default='', max_length=32),
+        ),
+    ]

ircbot/models.py

@@ -104,6 +104,8 @@ class IrcChannel(models.Model):
 
     markov_learn_from_channel = models.BooleanField(default=True)
 
+    discord_bridge = models.CharField(default='', max_length=32, blank=True)
+
     class Meta:
         """Settings for the model."""
 

markov/ircplugin.py

@@ -60,15 +60,10 @@ class Markov(Plugin):
     def handle_chatter(self, connection, event):
         """Learn from IRC chatter."""
         what = event.arguments[0]
-        if connection.server_config.additional_addressed_nicks:
+        who = irc.client.NickMask(event.source).nick
-            all_nicks = '|'.join(connection.server_config.additional_addressed_nicks.split('\n') +
-                                 [connection.get_nickname()])
-        else:
-            all_nicks = connection.get_nickname()
-        trimmed_what = re.sub(r'^(({nicks})[:,]|@({nicks}))\s+'.format(nicks=all_nicks), '', what)
-        nick = irc.client.NickMask(event.source).nick
         target = reply_destination_for_event(event)
 
+        log.debug("what: '%s', who: '%s', target: '%s'", what, who, target)
         # check to see whether or not we should learn from this channel
         channel = None
         if irc.client.is_channel(target):
@@ -78,11 +73,25 @@ class Markov(Plugin):
             log.debug("not learning from %s as i've been told to ignore it", channel)
         else:
             # learn the line
+            learning_what = what
+
+            # don't learn the speaker's nick if this came over a bridge
+            if channel and who == channel.discord_bridge:
+                learning_what = ' '.join(learning_what.split(' ')[1:])
+
+            # remove our own nick and aliases from what we learn
+            if connection.server_config.additional_addressed_nicks:
+                all_nicks = '|'.join(connection.server_config.additional_addressed_nicks.split('\n') +
+                                     [connection.get_nickname()])
+            else:
+                all_nicks = connection.get_nickname()
+            learning_what = re.sub(r'^(({nicks})[:,]|@({nicks}))\s+'.format(nicks=all_nicks), '', learning_what)
+
             recursing = getattr(event, 'recursing', False)
             if not recursing:
-                log.debug("learning %s", trimmed_what)
+                log.debug("learning %s", learning_what)
                 context = markovlib.get_or_create_target_context(target)
-                markovlib.learn_line(trimmed_what, context)
+                markovlib.learn_line(learning_what, context)
 
         log.debug("searching '%s' for '%s'", what, all_nicks)
         if re.search(all_nicks, what, re.IGNORECASE) is not None:
@@ -96,7 +105,7 @@ class Markov(Plugin):
                 topics = [x for x in match.group('addressed_msg').split(' ') if len(x) >= 3]
 
                 return self.bot.reply(event, "{0:s}: {1:s}"
-                                      "".format(nick, " ".join(markovlib.generate_line(context, topics=topics))))
+                                      "".format(who, " ".join(markovlib.generate_line(context, topics=topics))))
             else:
                 # i wasn't addressed directly, so just respond
                 topics = [x for x in what.split(' ') if len(x) >= 3]

markov/lib.py

@@ -1,3 +1,4 @@
+"""Provide methods for manipulating markov chain processing."""
 import logging
 import random
 
@@ -5,13 +6,11 @@ from django.db.models import Sum
 
 from markov.models import MarkovContext, MarkovState, MarkovTarget
 
-
+log = logging.getLogger(__name__)
-log = logging.getLogger('markov.lib')
 
 
 def generate_line(context, topics=None, min_words=15, max_words=30, sentence_bias=2, max_tries=5):
-    """String multiple sentences together into a coherent sentence."""
+    """Combine multiple sentences together into a coherent sentence."""
-
     tries = 0
     line = []
     min_words_per_sentence = min_words / sentence_bias
@@ -23,7 +22,7 @@ def generate_line(context, topics=None, min_words=15, max_words=30, sentence_bia
         else:
             if len(line) > 0:
                 if line[-1][-1] not in [',', '.', '!', '?', ':']:
-                    line[-1] += random.choice(['?', '.', '!'])
+                    line[-1] += random.SystemRandom().choice(['?', '.', '!'])
 
         tries += 1
 
@@ -33,7 +32,6 @@ def generate_line(context, topics=None, min_words=15, max_words=30, sentence_bia
 
 def generate_longish_sentence(context, topics=None, min_words=15, max_words=30, max_tries=100):
     """Generate a Markov chain, but throw away the short ones unless we get desperate."""
-
     sent = ""
     tries = 0
     while tries < max_tries:
@@ -52,20 +50,19 @@ def generate_longish_sentence(context, topics=None, min_words=15, max_words=30,
 
 def generate_sentence(context, topics=None, min_words=15, max_words=30):
     """Generate a Markov chain."""
-
     words = []
     # if we have topics, try to work from it and work backwards
     if topics:
-        topic_word = random.choice(topics)
+        topic_word = random.SystemRandom().choice(topics)
         topics.remove(topic_word)
-        log.debug("looking for topic '{0:s}'".format(topic_word))
+        log.debug("looking for topic '%s'", topic_word)
         new_states = MarkovState.objects.filter(context=context, v=topic_word)
 
         if len(new_states) > 0:
-            log.debug("found '{0:s}', starting backwards".format(topic_word))
+            log.debug("found '%s', starting backwards", topic_word)
             words.insert(0, topic_word)
             while len(words) <= max_words and words[0] != MarkovState._start2:
-                log.debug("looking backwards for '{0:s}'".format(words[0]))
+                log.debug("looking backwards for '%s'", words[0])
                 new_states = MarkovState.objects.filter(context=context, v=words[0])
                 # if we find a start, use it
                 if MarkovState._start2 in new_states:
@@ -87,7 +84,7 @@ def generate_sentence(context, topics=None, min_words=15, max_words=30):
 
     i = len(words)
     while words[-1] != MarkovState._stop:
-        log.debug("looking for '{0:s}','{1:s}'".format(words[i-2], words[i-1]))
+        log.debug("looking for '%s','%s'", words[i-2], words[i-1])
         new_states = MarkovState.objects.filter(context=context, k1=words[i-2], k2=words[i-1])
         log.debug("states retrieved")
 
@@ -103,7 +100,7 @@ def generate_sentence(context, topics=None, min_words=15, max_words=30):
             words.append(MarkovState._stop)
         elif len(target_hits) > 0:
             # if there's a target word in the states, pick it
-            target_hit = random.choice(target_hits)
+            target_hit = random.SystemRandom().choice(target_hits)
             log.debug("found a topic hit %s, using it", target_hit)
             topics.remove(target_hit)
             words.append(target_hit)
@@ -129,7 +126,6 @@ def generate_sentence(context, topics=None, min_words=15, max_words=30):
 
 def get_or_create_target_context(target_name):
     """Return the context for a provided nick/channel, creating missing ones."""
-
     target_name = target_name.lower()
 
     # find the stuff, or create it
@@ -156,7 +152,6 @@ def get_or_create_target_context(target_name):
 
 def get_word_out_of_states(states, backwards=False):
     """Pick one random word out of the given states."""
-
     # work around possible broken data, where a k1,k2 should have a value but doesn't
     if len(states) == 0:
         states = MarkovState.objects.filter(v=MarkovState._stop)
@@ -168,9 +163,9 @@ def get_word_out_of_states(states, backwards=False):
         # this being None probably means there's no data for this context
         raise ValueError("no markov states to generate from")
 
-    hit = random.randint(0, count_sum)
+    hit = sysrand.randint(0, count_sum)
 
-    log.debug("sum: {0:d} hit: {1:d}".format(count_sum, hit))
+    log.debug("sum: %s hit: %s", count_sum, hit)
 
     states_itr = states.iterator()
     for state in states_itr:
@@ -183,13 +178,12 @@ def get_word_out_of_states(states, backwards=False):
 
             break
 
-    log.debug("found '{0:s}'".format(new_word))
+    log.debug("found '%s'", new_word)
     return new_word
 
 
 def learn_line(line, context):
     """Create a bunch of MarkovStates for a given line of text."""
-
     log.debug("learning %s...", line[:40])
 
     words = line.split()
@@ -200,7 +194,7 @@ def learn_line(line, context):
             return
 
     for i, word in enumerate(words):
-        log.debug("'{0:s}','{1:s}' -> '{2:s}'".format(words[i], words[i+1], words[i+2]))
+        log.debug("'%s','%s' -> '%s'", words[i], words[i+1], words[i+2])
         state, created = MarkovState.objects.get_or_create(context=context,
                                                            k1=words[i],
                                                            k2=words[i+1],

markov/management/__init__.py

@@ -0,0 +1 @@
+"""Management operations for the markov plugin and models."""

markov/management/commands/__init__.py

@@ -0,0 +1 @@
+"""Management commands for the markov plugin and models."""

markov/management/commands/fix_chains_with_nicks_via_bridge.py

@@ -0,0 +1,77 @@
+"""Clean up learned chains with speaker nicks (from the bridge) or self (because the bridge broke the regex)."""
+from django.core.management import BaseCommand
+
+from ircbot.models import IrcChannel
+from markov.models import MarkovContext, MarkovState
+
+
+class Command(BaseCommand):
+    """Find markov chains that erroneously have speaker/self nicks and remove them."""
+
+    def handle(self, *args, **kwargs):
+        """Scan the DB, looking for bad chains, and repair them."""
+        candidate_channels = IrcChannel.objects.exclude(discord_bridge='')
+        markov_contexts = MarkovContext.objects.filter(markovtarget__name__in=list(candidate_channels))
+        for context in markov_contexts:
+            self.stdout.write(self.style.NOTICE(f"scanning context {context}..."))
+            # get starting states that look like they came over the bridge
+            bridge_states = context.states.filter(k1=MarkovState._start1, k2=MarkovState._start2,
+                                                  v__regex=r'<.*>')
+            self._chain_remover(context, bridge_states)
+
+    def _chain_remover(self, context, start_states):
+        """Remove a given k from markov states, deleting the found states after rebuilding subsequent states.
+
+        As in, if trying to remove A,B -> X, then B,X -> C and X,C -> D must be rebuilt (A,B -> C / B,C -> D)
+        then the three states with X deleted.
+        """
+        for start_state in start_states:
+            self.stdout.write(self.style.NOTICE(f"  diving into {start_state}..."))
+            # find the states that build off of the start
+            second_states = context.states.filter(k1=start_state.k2, k2=start_state.v)
+
+            for second_state in second_states:
+                self.stdout.write(self.style.NOTICE(f"    diving into {second_state}..."))
+                # find the third states
+                leaf_states = context.states.filter(k1=second_state.k2, k2=second_state.v)
+
+                for leaf_state in leaf_states:
+                    self.stdout.write(self.style.NOTICE(f"      upserting state based on {leaf_state}"))
+                    # get/update state without the nick from the bridge
+                    try:
+                        updated_leaf = MarkovState.objects.get(k1=second_state.k1, k2=leaf_state.k2, v=leaf_state.v)
+                        updated_leaf.count += leaf_state.count
+                        updated_leaf.save()
+                        self.stdout.write(self.style.SUCCESS(f"        updated count for {updated_leaf}"))
+                    except MarkovState.DoesNotExist:
+                        new_leaf = MarkovState.objects.create(k1=second_state.k1, k2=leaf_state.k2, v=leaf_state.v,
+                                                              context=context)
+                        new_leaf.count = leaf_state.count
+                        new_leaf.save()
+                        self.stdout.write(self.style.SUCCESS(f"        created {new_leaf}"))
+
+                    # remove the migrated leaf state
+                    self.stdout.write(self.style.SUCCESS(f"        deleting {leaf_state}"))
+                    leaf_state.delete()
+
+                # take care of the new middle state
+                self.stdout.write(self.style.NOTICE(f"    upserting state based on {second_state}"))
+                try:
+                    updated_second = MarkovState.objects.get(k1=start_state.k1, k2=start_state.k2, v=second_state.v)
+                    updated_second.count += second_state.count
+                    updated_second.save()
+                    self.stdout.write(self.style.SUCCESS(f"      updated count for {updated_second}"))
+                except MarkovState.DoesNotExist:
+                    new_second = MarkovState.objects.create(k1=start_state.k1, k2=start_state.k2, v=second_state.v,
+                                                            context=context)
+                    new_second.count = second_state.count
+                    new_second.save()
+                    self.stdout.write(self.style.SUCCESS(f"      created {new_second}"))
+
+                # remove the migrated second state
+                self.stdout.write(self.style.SUCCESS(f"      deleting {second_state}"))
+                second_state.delete()
+
+            # remove the dead end original start
+            self.stdout.write(self.style.SUCCESS(f"    deleting {start_state}"))
+            start_state.delete()

markov/migrations/0004_alter_markovstate_context.py

@@ -0,0 +1,19 @@
+# Generated by Django 3.2.18 on 2023-02-19 19:00
+
+from django.db import migrations, models
+import django.db.models.deletion
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('markov', '0003_auto_20161112_2348'),
+    ]
+
+    operations = [
+        migrations.AlterField(
+            model_name='markovstate',
+            name='context',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='states', to='markov.markovcontext'),
+        ),
+    ]

markov/models.py

@@ -41,7 +41,7 @@ class MarkovState(models.Model):
     v = models.CharField(max_length=128)
 
     count = models.IntegerField(default=0)
-    context = models.ForeignKey(MarkovContext, on_delete=models.CASCADE)
+    context = models.ForeignKey(MarkovContext, on_delete=models.CASCADE, related_name='states')
 
     class Meta:
         """Options for the model itself."""

tests/test_markov_ircplugin.py

@@ -0,0 +1,102 @@
+"""Test IRC behavior of the markov plugin."""
+from unittest import mock
+
+from django.test import TestCase
+
+from ircbot.models import IrcServer
+from markov.ircplugin import Markov
+
+
+class MarkovTestCase(TestCase):
+    """Test the markov plugin."""
+
+    fixtures = ['tests/fixtures/irc_server_fixture.json']
+
+    def setUp(self):
+        """Create common objects."""
+        self.mock_bot = mock.MagicMock()
+        self.mock_connection = mock.MagicMock()
+
+        self.mock_connection.get_nickname.return_value = 'test_bot'
+        self.mock_connection.server_config = IrcServer.objects.get(pk=1)
+
+        self.plugin = Markov(self.mock_bot, self.mock_connection, mock.MagicMock())
+
+    def test_learn(self):
+        """Test that an IRC event triggers learning as expected."""
+        mock_event = mock.MagicMock()
+        mock_event.arguments = ['hello this is a test message']
+        mock_event.target = '#test'
+        mock_event.recursing = False
+
+        with mock.patch('markov.lib.learn_line') as mock_learn_line:
+            self.plugin.handle_chatter(self.mock_connection, mock_event)
+
+        self.assertEqual(mock_learn_line.call_args.args[0], 'hello this is a test message')
+
+    def test_learn_self_edit(self):
+        """Test that we don't learn our own name when learning something addressed to us."""
+        mock_event = mock.MagicMock()
+        mock_event.arguments = ['test_bot: hello this is a test message']
+        mock_event.target = '#test'
+        mock_event.recursing = False
+
+        with mock.patch('markov.lib.learn_line') as mock_learn_line:
+            with mock.patch('markov.lib.generate_line'):
+                self.plugin.handle_chatter(self.mock_connection, mock_event)
+
+        self.assertEqual(mock_learn_line.call_args.args[0], 'hello this is a test message')
+
+    def test_learn_variant_self_edit(self):
+        """Test that we don't learn our own name when learning something addressed to us, discord style."""
+        mock_event = mock.MagicMock()
+        mock_event.arguments = ['@test_bot hello this is a test message']
+        mock_event.target = '#test'
+        mock_event.recursing = False
+
+        with mock.patch('markov.lib.learn_line') as mock_learn_line:
+            with mock.patch('markov.lib.generate_line'):
+                self.plugin.handle_chatter(self.mock_connection, mock_event)
+
+        self.assertEqual(mock_learn_line.call_args.args[0], 'hello this is a test message')
+
+    def test_learn_bridge_edit(self):
+        """Test that we don't learn the speaker's nick when learning a message from the bridge."""
+        mock_event = mock.MagicMock()
+        mock_event.arguments = ['<tester> hello this is a test message']
+        mock_event.target = '#test'
+        mock_event.recursing = False
+        mock_event.source = 'bridge!bridge@localhost'
+
+        with mock.patch('markov.lib.learn_line') as mock_learn_line:
+            self.plugin.handle_chatter(self.mock_connection, mock_event)
+
+        self.assertEqual(mock_learn_line.call_args.args[0], 'hello this is a test message')
+
+    def test_learn_bridge_and_self_edit(self):
+        """Test that we don't learn our own name when learning something addressed to us, discord style."""
+        mock_event = mock.MagicMock()
+        mock_event.arguments = ['<tester> test_bot: hello this is a test message']
+        mock_event.target = '#test'
+        mock_event.recursing = False
+        mock_event.source = 'bridge!bridge@localhost'
+
+        with mock.patch('markov.lib.learn_line') as mock_learn_line:
+            with mock.patch('markov.lib.generate_line'):
+                self.plugin.handle_chatter(self.mock_connection, mock_event)
+
+        self.assertEqual(mock_learn_line.call_args.args[0], 'hello this is a test message')
+
+    def test_learn_bridge_and_variant_self_edit(self):
+        """Test that we don't learn our own name when learning something addressed to us, discord style."""
+        mock_event = mock.MagicMock()
+        mock_event.arguments = ['<tester> @test_bot hello this is a test message']
+        mock_event.target = '#test'
+        mock_event.recursing = False
+        mock_event.source = 'bridge!bridge@localhost'
+
+        with mock.patch('markov.lib.learn_line') as mock_learn_line:
+            with mock.patch('markov.lib.generate_line'):
+                self.plugin.handle_chatter(self.mock_connection, mock_event)
+
+        self.assertEqual(mock_learn_line.call_args.args[0], 'hello this is a test message')

tox.ini

@@ -194,5 +194,6 @@ python_files =
     *_tests.py
     tests.py
     test_*.py
+log_level=DEBUG
 DJANGO_SETTINGS_MODULE = dr_botzo.settings
 django_find_project = false