mmintel/instagram-bot
1
0
Fork 0
Code Issues Pull Requests 4 Actions Packages Projects Releases Wiki Activity

fix(sae): stabilize navigation engine, fix container filtering, and negative reinforcement logic

Browse Source
This commit is contained in:
Marc Mintel
2026-04-25 13:09:12 +02:00
parent ad1af4edfe
commit 77e8251aa7
43 changed files with 4107 additions and 4075 deletions
Download Patch File Download Diff File Expand all files Collapse all files

151
tests/e2e/conftest.py
View File

@@ -1,11 +1,13 @@
import sys
import os
import pytest
import sys
import time
from unittest.mock import MagicMock
import pytest
from GramAddict.core import utils
# Force Qdrant mocking globally across ALL E2E tests so we never
# Force Qdrant mocking globally across ALL E2E tests so we never
# block on connection refused trying to hit localhost:6344
mock_qdrant = MagicMock()
@@ -16,11 +18,12 @@ mock_qdrant.get_collection.return_value = mock_collection
sys.modules["qdrant_client"].QdrantClient = MagicMock(return_value=mock_qdrant)
@pytest.fixture
def e2e_device_dump_injector(request):
"""
Provides a factory to mock device.dump_hierarchy using real XML files.
Will gracefully fail with a comprehensive assertion if the file is missing
Will gracefully fail with a comprehensive assertion if the file is missing
(per 'ECHTE DUMPS fehlen' reporting requirement).
"""
if request.config.getoption("--live"):
@@ -29,30 +32,36 @@ def e2e_device_dump_injector(request):
def _inject_dump(device_mock, xml_filename):
fix_dir = os.path.join(os.path.dirname(os.path.dirname(__file__)), "fixtures")
xml_path = os.path.join(fix_dir, xml_filename)
if not os.path.exists(xml_path):
pytest.fail(f"MISSING REAL DUMP: required XML fixture '{xml_filename}' for full E2E workflow testing could not be found at {xml_path}. FAKE_NOTHING policy implies dropping this test execution until it is captured.", pytrace=False)
pytest.fail(
f"MISSING REAL DUMP: required XML fixture '{xml_filename}' for full E2E workflow testing could not be found at {xml_path}. FAKE_NOTHING policy implies dropping this test execution until it is captured.",
pytrace=False,
)
with open(xml_path, "r") as f:
real_xml = f.read()
device_mock.dump_hierarchy.return_value = real_xml
return real_xml
return _inject_dump
class VirtualClock:
def __init__(self):
self.time = 0.0
self.animation_target_time = 0.0
def sleep(self, seconds):
if hasattr(seconds, '__iter__'):
return # For edge case where something weird is passed
if hasattr(seconds, "__iter__"):
return # For edge case where something weird is passed
self.time += float(seconds)
clock = VirtualClock()
@pytest.fixture
def dynamic_e2e_dump_injector(monkeypatch, request):
"""
@@ -66,9 +75,9 @@ def dynamic_e2e_dump_injector(monkeypatch, request):
def _inject(device_mock, state_map, initial_xml):
from GramAddict.core.q_nav_graph import QNavGraph
fix_dir = os.path.join(os.path.dirname(os.path.dirname(__file__)), "fixtures")
def load_xml(filename):
path = os.path.join(fix_dir, filename)
if not os.path.exists(path):
@@ -79,47 +88,56 @@ def dynamic_e2e_dump_injector(monkeypatch, request):
# History stack to allow "back" navigation
device_mock._xml_history = [load_xml(initial_xml)]
device_mock._current_active_xml = device_mock._xml_history[-1]
import uuid
def _dump_hierarchy_hook():
if clock.time < clock.animation_target_time:
pytest.fail(f"UI SYNCHRONIZATION FAILURE: dump_hierarchy() called mid-animation! "
f"Virtual Clock is at {clock.time:.1f}s but UI needs until {clock.animation_target_time:.1f}s to settle. "
f"Add a time.sleep() guard before interacting with the UI after a click.", pytrace=False)
pytest.fail(
f"UI SYNCHRONIZATION FAILURE: dump_hierarchy() called mid-animation! "
f"Virtual Clock is at {clock.time:.1f}s but UI needs until {clock.animation_target_time:.1f}s to settle. "
f"Add a time.sleep() guard before interacting with the UI after a click.",
pytrace=False,
)
xml = device_mock._current_active_xml
if xml and "</hierarchy>" in xml:
xml = xml.replace("</hierarchy>", f"<node sid=\"{uuid.uuid4()}\" /></hierarchy>")
xml = xml.replace("</hierarchy>", f'<node sid="{uuid.uuid4()}" /></hierarchy>')
return xml
device_mock.dump_hierarchy.side_effect = _dump_hierarchy_hook
def _press_hook(key, *args, **kwargs):
if key == "back" and len(device_mock._xml_history) > 1:
device_mock._xml_history.pop()
device_mock._current_active_xml = device_mock._xml_history[-1]
clock.animation_target_time = clock.time + 1.5
device_mock.press.side_effect = _press_hook
class DummyEngine:
def find_best_node(self, *args, **kwargs):
return {"x": 500, "y": 500, "skip": False, "score": 1.0, "source": "e2e_mock"}
def verify_success(self, *args, **kwargs):
return True
def confirm_click(self, *args, **kwargs):
pass
def reject_click(self, *args, **kwargs):
pass
original_execute = QNavGraph._execute_transition
from GramAddict.core.goap import GoalExecutor
original_goap_execute = GoalExecutor._execute_action
def _mock_execute_transition(nav_self, action, zero_engine=None, max_retries=2):
if action == 'tap_post_username':
if action == "tap_post_username":
return True
original_click = nav_self.device.click
def _click_hook(obj=None, *args, **kwargs):
original_click(obj, *args, **kwargs)
if action in state_map:
@@ -127,22 +145,24 @@ def dynamic_e2e_dump_injector(monkeypatch, request):
device_mock._xml_history.append(new_xml)
device_mock._current_active_xml = new_xml
clock.animation_target_time = clock.time + 1.5
nav_self.device.click = _click_hook
try:
success = original_execute(nav_self, action, mock_semantic_engine=DummyEngine(), max_retries=max_retries)
success = original_execute(
nav_self, action, mock_semantic_engine=DummyEngine(), max_retries=max_retries
)
return success
finally:
nav_self.device.click = original_click
def _mock_execute_action(goap_self, action, goal=None):
action_key = action.replace(" ", "_")
if action_key == 'tap_post_username':
if action_key == "tap_post_username":
return True
original_click = goap_self.device.click
def _click_hook(obj=None, *args, **kwargs):
original_click(obj, *args, **kwargs)
if action_key in state_map:
@@ -155,20 +175,21 @@ def dynamic_e2e_dump_injector(monkeypatch, request):
device_mock._xml_history.append(new_xml)
device_mock._current_active_xml = new_xml
clock.animation_target_time = clock.time + 1.5
goap_self.device.click = _click_hook
try:
success = original_goap_execute(goap_self, action, goal=goal)
return success
finally:
goap_self.device.click = original_click
monkeypatch.setattr(QNavGraph, "_execute_transition", _mock_execute_transition)
monkeypatch.setattr(GoalExecutor, "_execute_action", _mock_execute_action)
return _inject
@pytest.fixture(autouse=True)
def mock_all_delays(monkeypatch, request):
"""
@@ -180,49 +201,72 @@ def mock_all_delays(monkeypatch, request):
return
global clock
clock.time = 0.0 # reset for test
clock.time = 0.0 # reset for test
clock.animation_target_time = 0.0
def simulate_sleep(seconds):
clock.sleep(seconds)
money_sleep = lambda x: simulate_sleep(x)
random_sleep = lambda *args, **kwargs: simulate_sleep(1.0) # Assume 1.0 minimum for randoms
random_sleep = lambda a=1.0, b=2.0, *args, **kwargs: simulate_sleep(max(1.5, float(a)))
monkeypatch.setattr(time, "sleep", money_sleep)
monkeypatch.setattr(utils, "random_sleep", random_sleep)
monkeypatch.setattr(utils, "sleep", money_sleep)
# Needs to capture specific module sleeps depending on how they imported it
try:
from GramAddict.core import bot_flow
monkeypatch.setattr(bot_flow, "sleep", money_sleep)
monkeypatch.setattr(bot_flow.random, "uniform", lambda a, b: float(a)) # deterministic lower bound
monkeypatch.setattr(bot_flow.random, "uniform", lambda a, b: float(a)) # deterministic lower bound
if hasattr(bot_flow, "random_sleep"):
monkeypatch.setattr(bot_flow, "random_sleep", random_sleep)
from GramAddict.core import q_nav_graph
monkeypatch.setattr(q_nav_graph.random, "uniform", lambda a, b: float(a))
if hasattr(q_nav_graph, "random_sleep"):
monkeypatch.setattr(q_nav_graph, "random_sleep", random_sleep)
from GramAddict.core import goap
if hasattr(goap, "random"):
monkeypatch.setattr(goap.random, "uniform", lambda a, b: float(a))
if hasattr(goap, "random_sleep"):
monkeypatch.setattr(goap, "random_sleep", random_sleep)
monkeypatch.setattr(utils.random, "uniform", lambda a, b: float(a))
from GramAddict.core import device_facade
monkeypatch.setattr(device_facade, "sleep", money_sleep)
monkeypatch.setattr(device_facade.random, "uniform", lambda a, b: float(a))
except Exception:
pass
if hasattr(device_facade, "random_sleep"):
monkeypatch.setattr(device_facade, "random_sleep", random_sleep)
except Exception as e:
print(f"Mocking delays exception: {e}")
# Standardize DarwinEngine across tests to prevent mockup math errors on session end
try:
from GramAddict.core.darwin_engine import DarwinEngine
monkeypatch.setattr(DarwinEngine, "evaluate_session_end", lambda *args, **kwargs: None)
except ImportError:
pass
@pytest.fixture(autouse=True)
def mock_identity_guard(monkeypatch):
import GramAddict.core.bot_flow
monkeypatch.setattr(GramAddict.core.bot_flow, "verify_and_switch_account", lambda *args, **kwargs: True)
@pytest.fixture
def e2e_configs():
import argparse
configs = MagicMock()
configs.username = "testuser"
configs.args = argparse.Namespace(
@@ -254,6 +298,7 @@ def e2e_configs():
)
return configs
@pytest.fixture(autouse=True)
def mock_sae_perceive(request, monkeypatch):
"""
@@ -263,9 +308,15 @@ def mock_sae_perceive(request, monkeypatch):
"""
if "test_e2e_sae.py" in str(request.node.fspath):
return
if "test_e2e_real_llm_learning.py" in str(request.node.fspath):
return
if request.config.getoption("--live"):
return
import GramAddict.core.situational_awareness
monkeypatch.setattr(GramAddict.core.situational_awareness.SituationalAwarenessEngine, "perceive", lambda self, xml: GramAddict.core.situational_awareness.SituationType.NORMAL)
import GramAddict.core.situational_awareness
monkeypatch.setattr(
GramAddict.core.situational_awareness.SituationalAwarenessEngine,
"perceive",
lambda self, xml: GramAddict.core.situational_awareness.SituationType.NORMAL,
)