Asset-Frameworker/gui/llm_interaction_handler.py

424 lines
23 KiB
Python

import os
import json # Added for direct config loading
import logging
from pathlib import Path
from PySide6.QtCore import QObject, Signal, QThread, Slot, QTimer
# --- Backend Imports ---
# Assuming these might be needed based on MainWindow's usage
try:
# Removed load_base_config import
# Removed Configuration import as we load manually now
from configuration import ConfigurationError # Keep error class
from .llm_prediction_handler import LLMPredictionHandler # Backend handler
from rule_structure import SourceRule # For signal emission type hint
except ImportError as e:
logging.getLogger(__name__).critical(f"Failed to import backend modules for LLMInteractionHandler: {e}")
LLMPredictionHandler = None
# load_base_config = None # Removed
ConfigurationError = Exception
SourceRule = None # Define as None if import fails
# Configuration = None # Removed
log = logging.getLogger(__name__)
# Define config file paths relative to this handler's location
CONFIG_DIR = Path(__file__).parent.parent / "config"
APP_SETTINGS_PATH = CONFIG_DIR / "app_settings.json"
LLM_SETTINGS_PATH = CONFIG_DIR / "llm_settings.json"
class LLMInteractionHandler(QObject):
"""
Handles the logic for interacting with the LLM prediction service,
including managing the queue, thread, and communication.
"""
# Signals to communicate results/status back to MainWindow or other components
llm_prediction_ready = Signal(str, list) # input_path, List[SourceRule]
llm_prediction_error = Signal(str, str) # input_path, error_message
llm_status_update = Signal(str) # status_message
llm_processing_state_changed = Signal(bool) # is_processing (True when busy, False when idle)
def __init__(self, main_window_ref, parent=None):
"""
Initializes the handler.
Args:
main_window_ref: A reference to the MainWindow instance for accessing
shared components like status bar or models if needed.
parent: The parent QObject.
"""
super().__init__(parent)
self.main_window = main_window_ref # Store reference if needed for status updates etc.
self.llm_processing_queue = [] # Unified queue for initial adds and re-interpretations
self.llm_prediction_thread = None
self.llm_prediction_handler = None
self._is_processing = False # Internal flag to track processing state
def _set_processing_state(self, processing: bool):
"""Updates the internal processing state and emits a signal."""
if self._is_processing != processing:
self._is_processing = processing
log.debug(f"LLM Handler processing state changed to: {processing}")
self.llm_processing_state_changed.emit(processing)
def force_reset_state(self):
"""Forces the processing state to False. Use with caution."""
log.warning("Forcing LLMInteractionHandler state reset.")
if self.llm_prediction_thread and self.llm_prediction_thread.isRunning():
log.warning("Force reset called while thread is running. Attempting to stop thread.")
# Attempt graceful shutdown first
self.llm_prediction_thread.quit()
if not self.llm_prediction_thread.wait(500): # Wait 0.5 sec
log.warning("LLM thread did not quit gracefully after force reset. Terminating.")
self.llm_prediction_thread.terminate()
self.llm_prediction_thread.wait() # Wait after terminate
self.llm_prediction_thread = None
self.llm_prediction_handler = None
self._set_processing_state(False)
# Do NOT clear the queue here, let the user decide via Clear Queue button
@Slot(str, list)
def queue_llm_request(self, input_path: str, file_list: list | None):
"""Adds a request to the LLM processing queue."""
log.debug(f"Queueing LLM request for '{input_path}'. Current queue size: {len(self.llm_processing_queue)}")
# Avoid duplicates? Check if already in queue
is_in_queue = any(item[0] == input_path for item in self.llm_processing_queue)
if not is_in_queue:
self.llm_processing_queue.append((input_path, file_list))
log.info(f"Added '{input_path}' to LLM queue. New size: {len(self.llm_processing_queue)}")
# If not currently processing, start the queue
if not self._is_processing:
# Use QTimer.singleShot to avoid immediate processing if called rapidly
QTimer.singleShot(0, self._process_next_llm_item)
else:
log.debug(f"Skipping duplicate add to LLM queue for: {input_path}")
@Slot(list)
def queue_llm_requests_batch(self, requests: list[tuple[str, list | None]]):
"""Adds multiple requests to the LLM processing queue."""
added_count = 0
log.debug(f"Queueing batch. Current queue content: {self.llm_processing_queue}") # ADDED DEBUG LOG
for input_path, file_list in requests:
is_in_queue = any(item[0] == input_path for item in self.llm_processing_queue)
if not is_in_queue:
self.llm_processing_queue.append((input_path, file_list))
added_count += 1
else:
log.debug(f"Skipping duplicate add to LLM queue for: {input_path}")
if added_count > 0:
log.info(f"Added {added_count} requests to LLM queue. New size: {len(self.llm_processing_queue)}")
# If not currently processing, start the queue
if not self._is_processing:
QTimer.singleShot(0, self._process_next_llm_item)
# --- Methods to be moved from MainWindow ---
@Slot()
def _reset_llm_thread_references(self):
"""Resets LLM thread and handler references after the thread finishes."""
log.debug("--> Entered LLMInteractionHandler._reset_llm_thread_references")
log.debug("Resetting LLM prediction thread and handler references.")
self.llm_prediction_thread = None
self.llm_prediction_handler = None
# --- Process next item now that the previous thread is fully finished ---
log.debug("Previous LLM thread finished. Setting processing state to False.")
self._set_processing_state(False) # Mark processing as finished
# The next item will be processed when _handle_llm_result or _handle_llm_error
# calls _process_next_llm_item after popping the completed item.
log.debug("<-- Exiting LLMInteractionHandler._reset_llm_thread_references")
def _start_llm_prediction(self, input_path_str: str, file_list: list = None):
"""
Sets up and starts the LLMPredictionHandler in a separate thread.
Emits signals for results, errors, or status updates.
If file_list is not provided, it will be extracted.
"""
log.debug(f"Attempting to start LLM prediction for: {input_path_str}")
# Extract file list if not provided (needed for re-interpretation calls)
if file_list is None:
log.debug(f"File list not provided for {input_path_str}, extracting...")
if hasattr(self.main_window, '_extract_file_list'):
file_list = self.main_window._extract_file_list(input_path_str)
if file_list is None:
error_msg = f"Failed to extract file list for {input_path_str} in _start_llm_prediction."
log.error(error_msg)
self.llm_status_update.emit(f"Error extracting files for {os.path.basename(input_path_str)}")
self.llm_prediction_error.emit(input_path_str, error_msg) # Signal error
return # Stop if extraction failed
else:
error_msg = f"MainWindow reference does not have _extract_file_list method."
log.error(error_msg)
self.llm_status_update.emit(f"Internal Error: Cannot extract files for {os.path.basename(input_path_str)}")
self.llm_prediction_error.emit(input_path_str, error_msg)
return # Stop
input_path_obj = Path(input_path_str) # Still needed for basename
if not file_list:
error_msg = f"LLM Error: No files found/extracted for {input_path_str}"
log.error(error_msg)
self.llm_status_update.emit(f"LLM Error: No files found for {input_path_obj.name}")
self.llm_prediction_error.emit(input_path_str, error_msg)
return
# --- Load Required Settings Directly ---
llm_settings = {}
try:
log.debug(f"Loading LLM settings from: {LLM_SETTINGS_PATH}")
with open(LLM_SETTINGS_PATH, 'r') as f:
llm_data = json.load(f)
# Extract required fields with defaults
llm_settings['endpoint_url'] = llm_data.get('llm_endpoint_url')
llm_settings['api_key'] = llm_data.get('llm_api_key') # Can be None
llm_settings['model_name'] = llm_data.get('llm_model_name', 'local-model')
llm_settings['temperature'] = llm_data.get('llm_temperature', 0.5)
llm_settings['request_timeout'] = llm_data.get('llm_request_timeout', 120)
llm_settings['predictor_prompt'] = llm_data.get('llm_predictor_prompt', '')
llm_settings['examples'] = llm_data.get('llm_examples', [])
log.debug(f"Loading App settings from: {APP_SETTINGS_PATH}")
with open(APP_SETTINGS_PATH, 'r') as f:
app_data = json.load(f)
# Extract required fields
llm_settings['asset_type_definitions'] = app_data.get('ASSET_TYPE_DEFINITIONS', {})
llm_settings['file_type_definitions'] = app_data.get('FILE_TYPE_DEFINITIONS', {})
# Validate essential settings
if not llm_settings['endpoint_url']:
raise ValueError("LLM endpoint URL is missing in llm_settings.json")
if not llm_settings['predictor_prompt']:
raise ValueError("LLM predictor prompt is missing in llm_settings.json")
log.debug("LLM and App settings loaded successfully for LLMInteractionHandler.")
except FileNotFoundError as e:
error_msg = f"LLM Error: Configuration file not found: {e.filename}"
log.critical(error_msg)
self.llm_status_update.emit("LLM Error: Cannot load configuration file.")
self.llm_prediction_error.emit(input_path_str, error_msg)
return
except json.JSONDecodeError as e:
error_msg = f"LLM Error: Failed to parse configuration file: {e}"
log.critical(error_msg)
self.llm_status_update.emit("LLM Error: Cannot parse configuration file.")
self.llm_prediction_error.emit(input_path_str, error_msg)
return
except ValueError as e: # Catch validation errors
error_msg = f"LLM Error: Invalid configuration - {e}"
log.critical(error_msg)
self.llm_status_update.emit("LLM Error: Invalid configuration.")
self.llm_prediction_error.emit(input_path_str, error_msg)
return
except Exception as e: # Catch other potential errors
error_msg = f"LLM Error: Unexpected error loading configuration: {e}"
log.critical(error_msg, exc_info=True)
self.llm_status_update.emit("LLM Error: Cannot load application configuration.")
self.llm_prediction_error.emit(input_path_str, error_msg)
return
# --- Wrap thread/handler setup and start in try...except ---
try:
# --- Check if Handler Class is Available ---
if LLMPredictionHandler is None:
# Raise ValueError to be caught below
raise ValueError("LLMPredictionHandler class not available.")
# --- Clean up previous thread/handler if necessary ---
# (Keep this cleanup logic as it handles potential stale threads)
if self.llm_prediction_thread and self.llm_prediction_thread.isRunning():
log.warning("Warning: Previous LLM prediction thread still running when trying to start new one. Attempting cleanup.")
if self.llm_prediction_handler:
if hasattr(self.llm_prediction_handler, 'cancel'):
self.llm_prediction_handler.cancel()
self.llm_prediction_thread.quit()
if not self.llm_prediction_thread.wait(1000): # Wait 1 sec
log.warning("LLM thread did not quit gracefully. Forcing termination.")
self.llm_prediction_thread.terminate()
self.llm_prediction_thread.wait() # Wait after terminate
self.llm_prediction_thread = None
self.llm_prediction_handler = None
log.info(f"Starting LLM prediction thread for source: {input_path_str} with {len(file_list)} files.")
self.llm_status_update.emit(f"Starting LLM interpretation for {input_path_obj.name}...")
# --- Create Thread and Handler ---
self.llm_prediction_thread = QThread(self) # Parent thread to self
# Pass the loaded settings dictionary
self.llm_prediction_handler = LLMPredictionHandler(input_path_str, file_list, llm_settings)
self.llm_prediction_handler.moveToThread(self.llm_prediction_thread)
# Connect signals from handler to *internal* slots or directly emit signals
self.llm_prediction_handler.prediction_ready.connect(self._handle_llm_result)
self.llm_prediction_handler.prediction_error.connect(self._handle_llm_error)
self.llm_prediction_handler.status_update.connect(self.llm_status_update) # Pass status through
# Connect thread signals
self.llm_prediction_thread.started.connect(self.llm_prediction_handler.run)
# Clean up thread and handler when finished
self.llm_prediction_thread.finished.connect(self._reset_llm_thread_references)
self.llm_prediction_thread.finished.connect(self.llm_prediction_handler.deleteLater)
self.llm_prediction_thread.finished.connect(self.llm_prediction_thread.deleteLater)
# Also ensure thread quits when handler signals completion/error
self.llm_prediction_handler.prediction_ready.connect(self.llm_prediction_thread.quit)
self.llm_prediction_handler.prediction_error.connect(self.llm_prediction_thread.quit)
# TODO: Add a logging.debug statement at the very beginning of LLMPredictionHandler.run()
# to confirm if the method is being reached. Example:
# log.debug(f"--> Entered LLMPredictionHandler.run() for {self.input_path}")
self.llm_prediction_thread.start()
log.debug(f"LLM prediction thread start() called for {input_path_str}. Is running: {self.llm_prediction_thread.isRunning()}") # ADDED DEBUG LOG
# Log success *after* start() is called successfully
log.debug(f"Successfully initiated LLM prediction thread for {input_path_str}.") # MOVED/REWORDED LOG
except Exception as e:
# --- Handle errors during setup/start ---
log.exception(f"Critical error during LLM thread setup/start for {input_path_str}: {e}")
error_msg = f"Error initializing LLM task for {input_path_obj.name}: {e}"
self.llm_status_update.emit(error_msg)
self.llm_prediction_error.emit(input_path_str, error_msg) # Signal the error
# --- Crucially, reset processing state if setup fails ---
log.warning("Resetting processing state due to thread setup/start error.")
self._set_processing_state(False)
# Clean up potentially partially created objects
if self.llm_prediction_handler:
self.llm_prediction_handler.deleteLater()
self.llm_prediction_handler = None
if self.llm_prediction_thread:
if self.llm_prediction_thread.isRunning():
self.llm_prediction_thread.quit()
self.llm_prediction_thread.wait(500)
self.llm_prediction_thread.terminate() # Force if needed
self.llm_prediction_thread.wait()
self.llm_prediction_thread.deleteLater()
self.llm_prediction_thread = None
# Do NOT automatically try the next item here, as the error might be persistent.
# Let the error signal handle popping the item and trying the next one.
# The error signal (_handle_llm_error) will pop the item and call _process_next_llm_item.
def is_processing(self) -> bool:
"""Safely checks if the LLM prediction thread is currently running."""
# Use the internal flag, which is more reliable than checking thread directly
# due to potential race conditions during cleanup.
# The thread check can be a fallback.
is_running_flag = self._is_processing
# Also check thread as a safeguard, though the flag should be primary
try:
is_thread_alive = self.llm_prediction_thread is not None and self.llm_prediction_thread.isRunning()
if is_running_flag != is_thread_alive:
# This might indicate the flag wasn't updated correctly, log it.
log.warning(f"LLM Handler processing flag ({is_running_flag}) mismatch with thread state ({is_thread_alive}). Flag is primary.")
return is_running_flag
except RuntimeError:
log.debug("is_processing: Caught RuntimeError checking isRunning (thread likely deleted).")
# If thread died unexpectedly, the flag might be stale. Reset it.
if self._is_processing:
self._set_processing_state(False)
return False
def _process_next_llm_item(self):
"""Processes the next directory in the unified LLM processing queue."""
log.debug(f"--> Entered _process_next_llm_item. Queue size: {len(self.llm_processing_queue)}")
if self.is_processing():
log.info("LLM processing already running. Waiting for current item to finish.")
# Do not pop from queue if already running, wait for _reset_llm_thread_references to call this again
return
if not self.llm_processing_queue:
log.info("LLM processing queue is empty. Finishing.")
self.llm_status_update.emit("LLM processing complete.")
self._set_processing_state(False) # Ensure state is set to idle
log.debug("<-- Exiting _process_next_llm_item (queue empty)")
return
# Set state to busy *before* starting
self._set_processing_state(True)
# Get next item *without* removing it yet
next_item = self.llm_processing_queue[0] # Peek at the first item
next_dir, file_list = next_item # Unpack the tuple
# --- Update Status/Progress ---
total_in_queue_now = len(self.llm_processing_queue)
status_msg = f"LLM Processing {os.path.basename(next_dir)} ({total_in_queue_now} remaining)..."
self.llm_status_update.emit(status_msg)
log.info(status_msg)
# --- Start Prediction (which might fail) ---
try:
# Pass the potentially None file_list. _start_llm_prediction handles extraction if needed.
self._start_llm_prediction(next_dir, file_list=file_list)
# --- DO NOT pop item here. Item is popped in _handle_llm_result or _handle_llm_error ---
# Log message moved into the try block of _start_llm_prediction
# log.debug(f"Successfully started LLM prediction thread for {next_dir}. Item remains in queue until finished.")
except Exception as e:
# This block now catches errors from _start_llm_prediction itself
log.exception(f"Error occurred *during* _start_llm_prediction call for {next_dir}: {e}")
error_msg = f"Error starting LLM for {os.path.basename(next_dir)}: {e}"
self.llm_status_update.emit(error_msg)
self.llm_prediction_error.emit(next_dir, error_msg) # Signal the error
# --- Remove the failed item from the queue ---
try:
failed_item = self.llm_processing_queue.pop(0)
log.warning(f"Removed failed item {failed_item} from LLM queue due to start error.")
except IndexError:
log.error("Attempted to pop failed item from already empty LLM queue after start error.")
# --- Attempt to process the *next* item ---
# Reset processing state since this one failed *before* the thread finished signal could
self._set_processing_state(False)
# Use QTimer.singleShot to avoid deep recursion
QTimer.singleShot(100, self._process_next_llm_item) # Try next item after a short delay
# --- Internal Slots to Handle Results/Errors from LLMPredictionHandler ---
@Slot(str, list)
def _handle_llm_result(self, input_path: str, source_rules: list):
"""Internal slot to receive results, pop item, and emit the public signal."""
log.debug(f"LLM Handler received result for {input_path}. Removing from queue and emitting llm_prediction_ready.")
# Remove the completed item from the queue
try:
# Find and remove the item by input_path
self.llm_processing_queue = [item for item in self.llm_processing_queue if item[0] != input_path]
log.debug(f"Removed '{input_path}' from LLM queue after successful prediction. New size: {len(self.llm_processing_queue)}")
except Exception as e:
log.error(f"Error removing '{input_path}' from LLM queue after success: {e}")
self.llm_prediction_ready.emit(input_path, source_rules)
# Process the next item in the queue
QTimer.singleShot(0, self._process_next_llm_item)
@Slot(str, str)
def _handle_llm_error(self, input_path: str, error_message: str):
"""Internal slot to receive errors, pop item, and emit the public signal."""
log.debug(f"LLM Handler received error for {input_path}: {error_message}. Removing from queue and emitting llm_prediction_error.")
# Remove the failed item from the queue
try:
# Find and remove the item by input_path
self.llm_processing_queue = [item for item in self.llm_processing_queue if item[0] != input_path]
log.debug(f"Removed '{input_path}' from LLM queue after error. New size: {len(self.llm_processing_queue)}")
except Exception as e:
log.error(f"Error removing '{input_path}' from LLM queue after error: {e}")
self.llm_prediction_error.emit(input_path, error_message)
# Process the next item in the queue
QTimer.singleShot(0, self._process_next_llm_item)
def clear_queue(self):
"""Clears the LLM processing queue."""
log.info(f"Clearing LLM processing queue ({len(self.llm_processing_queue)} items).")
self.llm_processing_queue.clear()
# TODO: Should we also attempt to cancel any *currently* running LLM task?
# This might be complex. For now, just clears the queue of pending items.
if self.is_processing():
log.warning("LLM queue cleared, but a task is currently running. It will complete.")
else:
self.llm_status_update.emit("LLM queue cleared.")