Merge pull request #695 from Pythagora-io/development

Development

Author: LeonOstrez

pilot/.env.example

@@ -1,7 +1,8 @@
 # OPENAI or AZURE or OPENROUTER
 ENDPOINT=OPENAI
 
-OPENAI_ENDPOINT=https://api.openai.com/v1/chat/completions
+# OPENAI_ENDPOINT=https://api.openai.com/v1/chat/completions
+OPENAI_ENDPOINT=
 OPENAI_API_KEY=
 
 AZURE_API_KEY=

pilot/const/common.py

@@ -29,6 +29,7 @@
     '.next',
     '.DS_Store',
     '__pycache__',
+    "site-packages",
     'node_modules',
     'package-lock.json',
     'venv',

pilot/const/ipc.py

@@ -16,6 +16,7 @@
     'loopTrigger': 'loopTrigger',          # Trigger loop feedback popup in extension
     'progress': 'progress',                # Progress bar for extension only
     'projectStats': 'projectStats',        # Project stats for extension only
+    'keyExpired': 'keyExpired',            # (Free trial) key expired message - for extension only
 }
 
 LOCAL_IGNORE_MESSAGE_TYPES = [
@@ -31,4 +32,5 @@
     'loopTrigger',
     'progress',
     'projectStats',
+    'keyExpired',
 ]

pilot/const/messages.py

@@ -1,7 +1,7 @@
 CHECK_AND_CONTINUE = 'Is everything working? Let me know if something needs to be changed for this task or type "continue" to proceed.'
 WHEN_USER_DONE = 'Once you have completed, enter "continue"'
-AFFIRMATIVE_ANSWERS = ['', 'y', 'yes', 'ok', 'okay', 'sure', 'absolutely', 'indeed', 'correct', 'affirmative', 'Use GPT Pilot\'s code']
-NEGATIVE_ANSWERS = ['n', 'no', 'skip', 'negative', 'not now', 'cancel', 'decline', 'stop', 'Keep my changes']
+AFFIRMATIVE_ANSWERS = ['', 'y', 'yes', 'ok', 'okay', 'sure', 'absolutely', 'indeed', 'correct', 'affirmative']
+NEGATIVE_ANSWERS = ['n', 'no', 'skip', 'negative', 'not now', 'cancel', 'decline', 'stop']
 STUCK_IN_LOOP = 'I\'m stuck in loop'
 NONE_OF_THESE = 'none of these'
 MAX_PROJECT_NAME_LENGTH = 50

pilot/const/telemetry.py

@@ -1,3 +1,3 @@
 LARGE_REQUEST_THRESHOLD = 50000  # tokens
 SLOW_REQUEST_THRESHOLD = 300  # seconds
-LOOP_THRESHOLD = 20  # number of steps in task to be considered a loop
+LOOP_THRESHOLD = 3  # number of iterations in task to be considered a loop

pilot/database/database.py

@@ -369,7 +369,7 @@ def save_command_run(project, command, cli_response, done_or_error_response, exi
     unique_data = {
         'app': project.args['app_id'],
         'previous_step': project.checkpoints['last_command_run'],
-        'high_level_step': project.current_step,
+        'high_level_step': str(project.checkpoints['last_development_step']['id']) if project.checkpoints['last_development_step'] else None,
     }
 
     data_fields = {
@@ -391,7 +391,7 @@ def save_user_input(project, query, user_input, hint):
     unique_data = {
         'app': project.args['app_id'],
         'previous_step': project.checkpoints['last_user_input'],
-        'high_level_step': project.current_step,
+        'high_level_step': str(project.checkpoints['last_development_step']['id']) if project.checkpoints['last_development_step'] else None,
     }
     data_fields = {
         'query': query,

pilot/helpers/AgentConvo.py

@@ -68,7 +68,7 @@ def send_message(self, prompt_path=None, prompt_data=None, function_calls: Funct
                                                   function_calls=function_calls, prompt_data=prompt_data,
                                                   temperature=self.temperature)
         except TokenLimitError as e:
-            save_development_step(self.agent.project, prompt_path, prompt_data, self.messages, '', str(e))
+            save_development_step(self.agent.project, prompt_path, prompt_data, self.messages, {"text": ""}, str(e))
             raise e
 
         # TODO: move this code to Developer agent - https://github.com/Pythagora-io/gpt-pilot/issues/91#issuecomment-1751964079

pilot/helpers/Project.py

@@ -34,7 +34,7 @@
 
 from utils.telemetry import telemetry
 from utils.task import Task
-from utils.utils import remove_lines_with_string, is_extension_old_version
+from utils.utils import remove_lines_with_string
 
 
 class Project:
@@ -58,8 +58,6 @@ def __init__(
             current_step (str, optional): Current step in the project. Default is None.
         """
         self.args = args
-        # TODO remove everything related to is_extension_old_version once new version is released and everybody has to update core
-        self.is_extension_old_version = is_extension_old_version(args)
         self.llm_req_num = 0
         self.command_runs_count = 0
         self.user_inputs_count = 0
@@ -98,12 +96,13 @@ def __init__(
             File.update_paths()
 
         # start loading of project (since backwards compatibility)
-        self.should_overwrite_files = False
+        self.should_overwrite_files = None
         self.last_detailed_user_review_goal = None
         self.last_iteration = None
         self.tasks_to_load = []
         self.features_to_load = []
         self.dev_steps_to_load = []
+        self.run_command = None
         # end loading of project
 
     def set_root_path(self, root_path: str):
@@ -118,22 +117,41 @@ def setup_loading(self):
             return
 
         self.skip_steps = True
-        should_overwrite_files = None
-        while should_overwrite_files is None or should_overwrite_files.lower() not in AFFIRMATIVE_ANSWERS + NEGATIVE_ANSWERS:
-            print('Use GPT Pilot\'s code/Keep my changes', type='buttons-only')
-            should_overwrite_files = styled_text(
+        while self.should_overwrite_files is None:
+            changes_made_question = f'Did you make any changes to "{self.args["name"]}" project files since last time you used Pythagora?'
+            print(changes_made_question, type='ipc', category='pythagora')
+            print('yes/no', type='buttons-only')
+            # must use styled_text() instead of ask_user() here to avoid finish_loading() call
+            changes_made = styled_text(
                 self,
-                "Can GPT Pilot overwrite code changes you made since last running GPT Pilot?",
-                ignore_user_input_count=True
+                changes_made_question,
+                ignore_user_input_count=True,
             )
 
-            logger.info('should_overwrite_files: %s', should_overwrite_files)
-            if should_overwrite_files in NEGATIVE_ANSWERS:
-                self.should_overwrite_files = False
-                break
-            elif should_overwrite_files in AFFIRMATIVE_ANSWERS:
+            # if there were no changes just load files from db
+            if changes_made.lower() in NEGATIVE_ANSWERS:
                 self.should_overwrite_files = True
                 break
+            # otherwise ask user if they want to use those changes
+            elif changes_made.lower() in AFFIRMATIVE_ANSWERS:
+                use_changes_question = 'Do you want to use those changes you made?'
+                use_changes_msg = 'yes'
+                dont_use_changes_msg = 'no, restore last pythagora state'
+                print(use_changes_question, type='ipc', category='pythagora')
+                print(f'{use_changes_msg}/{dont_use_changes_msg}', type='buttons-only')
+                print(f'"{dont_use_changes_msg}" means Pythagora will restore (overwrite) all files to last stored state.\n'
+                      f'"{use_changes_msg}" means Pythagora will continue working on project using current state of files.', type='hint')
+                use_changes = styled_text(
+                    self,
+                    use_changes_question,
+                    ignore_user_input_count=True
+                )
+
+                logger.info('Use changes: %s', use_changes)
+                if use_changes.lower() in NEGATIVE_ANSWERS + [dont_use_changes_msg]:
+                    self.should_overwrite_files = True
+                elif use_changes.lower() in AFFIRMATIVE_ANSWERS + [use_changes_msg]:
+                    self.should_overwrite_files = False
 
         load_step_before_coding = ('step' in self.args and
                                    self.args['step'] is not None and
@@ -155,6 +173,9 @@ def setup_loading(self):
             self.checkpoints['last_development_step'] = self.dev_steps_to_load[-1]
             self.tasks_to_load = [el for el in self.dev_steps_to_load if 'breakdown.prompt' in el.get('prompt_path', '')]
             self.features_to_load = [el for el in self.dev_steps_to_load if 'feature_plan.prompt' in el.get('prompt_path', '')]
+            self.run_command = next((el for el in reversed(self.dev_steps_to_load) if 'get_run_command.prompt' in el.get('prompt_path', '')), None)
+            if self.run_command is not None:
+                self.run_command = json.loads(self.run_command['llm_response']['text'])['command']
 
     def start(self):
         """
@@ -216,11 +237,13 @@ def finish(self):
             self.previous_features = get_features_by_app_id(self.args['app_id'])
             if not self.skip_steps:
                 print('', type='verbose', category='pythagora')
+                if self.run_command and self.check_ipc():
+                    print(self.run_command, type='run_command')
                 feature_description = ask_user(self, "Project is finished! Do you want to add any features or changes? "
                                                      "If yes, describe it here and if no, just press ENTER",
                                                require_some_input=False)
 
-                if feature_description == '':
+                if feature_description == '' or feature_description == 'continue':
                     return
 
                 print('', type='verbose', category='agent:tech-lead')
@@ -391,13 +414,14 @@ def save_file(self, data):
             inputs_required = self.find_input_required_lines(data['content'])
             for line_number, line_content in inputs_required:
                 user_input = None
+                print('', type='verbose', category='human-intervention')
                 print(color_yellow_bold(f'Input required on line {line_number}:\n{line_content}') + '\n')
                 while user_input is None or user_input.lower() not in AFFIRMATIVE_ANSWERS + ['continue']:
                     print({'path': full_path, 'line': line_number}, type='openFile')
                     print('continue', type='buttons-only')
                     user_input = ask_user(
                         self,
-                        f'Please open the file {data["path"]} on the line {line_number} and add the required input. Once you\'re done, type "y" to continue.',
+                        f'Please open the file {data["path"]} on the line {line_number} and add the required input. Please, also remove "// INPUT_REQUIRED" comment and once you\'re done, press "continue".',
                         require_some_input=False,
                         ignore_user_input_count=True
                     )
@@ -458,7 +482,7 @@ def normalize_path(path: str) -> Tuple[str, str]:
             # - /pilot -> /pilot/
             # - \pilot\server.js -> \pilot\server.js
             # - \pilot -> \pilot\
-            KNOWN_FILES = ["makefile", "dockerfile", "procfile", "readme", "license"]  # known exceptions that break the heuristic
+            KNOWN_FILES = ["makefile", "dockerfile", "procfile", "readme", "license", "podfile"]  # known exceptions that break the heuristic
             KNOWN_DIRS = []  # known exceptions that break the heuristic
             base = os.path.basename(path)
             if (
@@ -543,7 +567,10 @@ def restore_files(self, development_step_id):
 
         clear_directory(self.root_path, ignore=self.files)
         for file_snapshot in file_snapshots:
-            update_file(file_snapshot.file.full_path, file_snapshot.content, project=self)
+            try:
+                update_file(file_snapshot.file.full_path, file_snapshot.content, project=self)
+            except (PermissionError, NotADirectoryError) as err:  # noqa
+                print(f"Error restoring file {file_snapshot.file.full_path}: {err}")
             if file_snapshot.file.full_path not in self.files:
                 self.files.append(file_snapshot.file.full_path)
 

pilot/helpers/agents/CodeMonkey.py

@@ -85,10 +85,12 @@ def implement_code_changes(
         files = self.project.get_all_coded_files()
         file_name, file_content = self.get_original_file(code_change_description, step, files)
 
+        print('', type='verbose', category='agent:code-monkey')
+
         if file_content:
-            print(f'Updating existing file {file_name}')
+            print(f'Updating existing file {file_name}:')
         else:
-            print(f'Creating new file {file_name}')
+            print(f'Creating new file {file_name}:')
 
         # Get the new version of the file
         content = self.replace_complete_file(
@@ -103,13 +105,15 @@ def implement_code_changes(
                 # There are no changes or there was problem talking with the LLM, we're done here
                 break
 
+            print('Sending code for review...', type='verbose', category='agent:code-monkey')
             print('', type='verbose', category='agent:reviewer')
             content, rework_feedback = self.review_change(convo, code_change_description, file_name, file_content, content)
-            print('', type='verbose', category='agent:code-monkey')
+            print('Review finished. Continuing...', type='verbose', category='agent:code-monkey')
             if not rework_feedback:
                 # No rework needed, we're done here
                 break
 
+            print('', type='verbose', category='agent:code-monkey')
             content = convo.send_message('development/review_feedback.prompt', {
                 "content": content,
                 "original_content": file_content,
@@ -258,7 +262,7 @@ def review_change(
         else:
             # The reviewer failed to review all the hunks in 3 attempts, let's just use all the new content
             convo.remove_last_x_messages(messages_to_remove)
-            return new_content
+            return new_content, None
 
         convo.remove_last_x_messages(messages_to_remove)
 

pilot/helpers/agents/Developer.py

@@ -45,14 +45,13 @@ def __init__(self, project):
         self.debugger = Debugger(self)
 
     def start_coding(self, task_source):
-        print('', type='verbose', category='agent:developer')
+        print('Starting development...', type='verbose', category='agent:developer')
         if not self.project.finished:
             self.project.current_step = 'coding'
             update_app_status(self.project.args['app_id'], self.project.current_step)
 
         # DEVELOPMENT
         if not self.project.skip_steps:
-            print(color_green_bold("🚀 Now for the actual development...\n"))
             logger.info("Starting to create the actual code...")
 
         total_tasks = len(self.project.development_plan)
@@ -123,7 +122,8 @@ def implement_task(self, i, task_source, development_task=None):
         :param task_source: The source of the task, one of: 'app', 'feature', 'debugger', 'iteration'.
         :param development_task: The task to implement.
         """
-        print(color_green_bold(f'Implementing task #{i + 1}: ') + color_green(f' {development_task["description"]}\n'), category='agent:developer')
+        print(color_green_bold(f'Implementing task #{i + 1}: ') + color_green(f' {development_task["description"]}\n'), category='pythagora')
+        print(f'Starting task #{i + 1} implementation...', type='verbose', category='agent:developer')
         self.project.dot_pilot_gpt.chat_log_folder(i + 1)
 
         convo_dev_task = AgentConvo(self)
@@ -391,6 +391,8 @@ def get_run_command(self, convo):
         elif single_match:
             self.run_command = single_match.group(1).strip()
 
+        self.project.run_command = self.run_command
+
     def task_postprocessing(self, convo, development_task, continue_development, task_result, last_branch_name):
         if self.project.last_detailed_user_review_goal is None:
             self.get_run_command(convo)
@@ -636,7 +638,7 @@ def continue_development(self, iteration_convo, last_branch_name, continue_descr
                                                                   return_cli_response=True, is_root_task=True)},
                 convo=iteration_convo,
                 is_root_task=True,
-                add_loop_button=iteration_count > 3,
+                add_loop_button=iteration_count > 2,
                 category='human-test')
 
             logger.info('response: %s', response)
@@ -648,7 +650,7 @@ def continue_development(self, iteration_convo, last_branch_name, continue_descr
 
             if user_feedback is not None:
                 print('', type='verbose', category='agent:troubleshooter')
-                user_feedback = self.bug_report_generator(user_feedback)
+                self.project.current_task.inc('iterations')
                 stuck_in_loop = user_feedback.startswith(STUCK_IN_LOOP)
                 if stuck_in_loop:
                     # Remove the STUCK_IN_LOOP prefix from the user feedback
@@ -666,6 +668,8 @@ def continue_development(self, iteration_convo, last_branch_name, continue_descr
                             tried_alternative_solutions_to_current_issue.append(description_of_tried_solutions)
 
                     tried_alternative_solutions_to_current_issue.append(next_solution_to_try)
+                else:
+                    user_feedback = self.bug_report_generator(user_feedback, user_description)
 
                 print_task_progress(1, 1, development_task['description'], 'troubleshooting', 'in_progress')
                 iteration_convo = AgentConvo(self)
@@ -711,11 +715,12 @@ def continue_development(self, iteration_convo, last_branch_name, continue_descr
                 self.execute_task(iteration_convo, task_steps, is_root_task=True, task_source='troubleshooting')
                 print_task_progress(1, 1, development_task['description'], 'troubleshooting', 'done')
 
-    def bug_report_generator(self, user_feedback):
+    def bug_report_generator(self, user_feedback, task_review_description):
         """
         Generate a bug report from the user feedback.
 
         :param user_feedback: The user feedback.
+        :param task_review_description: The task review description.
         :return: The bug report.
         """
         bug_report_convo = AgentConvo(self)
@@ -727,6 +732,7 @@ def bug_report_generator(self, user_feedback):
                 "user_feedback": user_feedback,
                 "app_summary": self.project.project_description,
                 "files": self.project.get_all_coded_files(),
+                "task_review_description": task_review_description,
                 "questions_and_answers": questions_and_answers,
             }, GET_BUG_REPORT_MISSING_DATA)
 
@@ -760,6 +766,7 @@ def bug_report_generator(self, user_feedback):
             bug_report_summary_convo = AgentConvo(self)
             user_feedback = bug_report_summary_convo.send_message('development/bug_report_summary.prompt', {
                 "app_summary": self.project.project_description,
+                "task_review_description": task_review_description,
                 "user_feedback": user_feedback,
                 "questions_and_answers": questions_and_answers,
             })
@@ -771,7 +778,7 @@ def review_task(self):
         Review all task changes and refactor big files.
         :return: bool - True if the task changes passed review, False if not
         """
-        print('', type='verbose', category='agent:reviewer')
+        print('Starting review of all changes made in this task...', type='verbose', category='agent:reviewer')
         self.review_count += 1
         review_result = self.review_code_changes()
         refactoring_done = self.refactor_code()
@@ -861,7 +868,13 @@ def set_up_environment(self):
                 dep_text = dependency['name']
 
             logger.info('Checking %s', dependency)
-            llm_response = self.check_system_dependency(dependency)
+            try:
+                llm_response = self.check_system_dependency(dependency)
+            except Exception as err:
+                # This catches weird errors like people removing or renaming the workspace
+                # folder while we're trying to run system commands. Since these commands don't
+                # care about folders they run in, we don't want to crash just because of that.
+                llm_response = str(err)
 
             if llm_response == 'DONE':
                 print(color_green_bold(f"✅ {dep_text} is available."))