You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
gpt-pilot/pilot/helpers/Debugger.py

137 lines
6.0 KiB
Python

import platform
import uuid
import re
from const.code_execution import MAX_COMMAND_DEBUG_TRIES, MAX_RECUSION_LAYER
from const.function_calls import DEBUG_STEPS_BREAKDOWN
from const.messages import AFFIRMATIVE_ANSWERS, NEGATIVE_ANSWERS
from helpers.exceptions.TokenLimitError import TokenLimitError
from helpers.exceptions.TooDeepRecursionError import TooDeepRecursionError
from logger.logger import logger
from prompts.prompts import ask_user
class Debugger:
def __init__(self, agent):
self.agent = agent
self.recursion_layer = 0
def debug(self, convo, command=None, user_input=None, issue_description=None, is_root_task=False, ask_before_debug=False):
"""
Debug a conversation.
Args:
convo (AgentConvo): The conversation object.
command (dict, optional): The command to debug. Default is None.
user_input (str, optional): User input for debugging. Default is None.
Should provide `command` or `user_input`.
issue_description (str, optional): Description of the issue to debug. Default is None.
ask_before_debug (bool, optional): True if we have to ask user for permission to start debugging.
Returns:
bool: True if debugging was successful, False otherwise.
"""
logger.info('Debugging %s', command)
self.recursion_layer += 1
if self.recursion_layer > MAX_RECUSION_LAYER:
self.recursion_layer = 0
raise TooDeepRecursionError()
function_uuid = str(uuid.uuid4())
convo.save_branch(function_uuid)
success = False
for i in range(MAX_COMMAND_DEBUG_TRIES):
if success:
break
if ask_before_debug or i > 0:
print('yes/no', type='button')
answer = ask_user(self.agent.project, 'Can I start debugging this issue?', require_some_input=False)
if answer.lower() in NEGATIVE_ANSWERS:
return True
if answer and answer.lower() not in AFFIRMATIVE_ANSWERS:
user_input = answer
llm_response = convo.send_message('dev_ops/debug.prompt',
{
'command': command['command'] if command is not None else None,
'user_input': user_input,
'issue_description': issue_description,
'os': platform.system(),
'context': convo.to_context_prompt()
},
DEBUG_STEPS_BREAKDOWN)
completed_steps = []
try:
while True:
logger.info('Thoughts: ' + llm_response['thoughts'])
logger.info('Reasoning: ' + llm_response['reasoning'])
steps = completed_steps + llm_response['steps']
# TODO refactor to nicely get the developer agent
result = self.agent.project.developer.execute_task(
convo,
f"Thoughts: {llm_response['thoughts']}\n\nReasoning: {llm_response['reasoning']}",
steps,
test_command=command,
test_after_code_changes=True,
continue_development=False,
is_root_task=is_root_task,
continue_from_step=len(completed_steps)
)
# in case one step failed or llm wants to see the output to determine the next steps
if 'step_index' in result:
result['os'] = platform.system()
step_index = result['step_index']
completed_steps = steps[:step_index+1]
result['completed_steps'] = completed_steps
result['current_step'] = steps[step_index]
result['next_steps'] = steps[step_index + 1:]
result['current_step_index'] = step_index
# Remove the previous debug plan and build a new one
convo.remove_last_x_messages(2)
# todo before updating task first check if update is needed
llm_response = convo.send_message('development/task/update_task.prompt', result,
DEBUG_STEPS_BREAKDOWN)
else:
success = result['success']
if not success:
convo.load_branch(function_uuid)
if 'cli_response' in result:
user_input = result['cli_response']
convo.messages[-2]['content'] = re.sub(
r'(?<=The output was:\n\n).*?(?=\n\nThink about this output)',
result['cli_response'],
convo.messages[-2]['content'],
flags=re.DOTALL
)
break
except TokenLimitError as e:
if self.recursion_layer > 0:
self.recursion_layer -= 1
raise e
else:
if not success:
convo.load_branch(function_uuid)
continue
# if not success:
# # TODO explain better how should the user approach debugging
# # we can copy the entire convo to clipboard so they can paste it in the playground
# user_input = convo.agent.project.ask_for_human_intervention(
# 'It seems like I cannot debug this problem by myself. Can you please help me and try debugging it yourself?' if user_input is None else f'Can you check this again:\n{issue_description}?',
# response['data']
# )
# if user_input == 'continue':
# success = True
self.recursion_layer -= 1
return success