[bitbucket-chatgpt-codereview ]Feature Improvement Request

[윤주오]

1. Request for additional logic to split into multiple batches to call and combine the results when GPT token is exceeded

2.  Added EXCLUX_FILES

i'm using it very well. thank you

example (from gpt) 

schema = {

    'OPENAI_API_KEY': {'type': 'string', 'required': True},

    'BITBUCKET_ACCESS_TOKEN': {'type': 'string', 'required': True},

    'MODEL': {'type': 'string', 'required': True, 'allowed': ['gpt-4-turbo-preview', 'gpt-3.5-turbo-0125']},

    'ORGANIZATION': {'type': 'string', 'required': False},

    'MESSAGE': {'type': 'string', 'required': False},

    'FILES_TO_REVIEW': {'type': 'string', 'required': False},

    'EXCLUDE_FILES': {'type': 'string', 'required': False},  # 제외할 파일 리스트 추가

    'CHATGPT_COMPLETION_FILEPATH': {'type': 'string', 'required': False},

    'CHATGPT_CLIENT_FILEPATH': {'type': 'string', 'required': False},

    'CHATGPT_PROMPT_MAX_TOKENS': {'type': 'integer', 'required': False, 'default': 0},

    'DEBUG': {'type': 'boolean', 'required': False, 'default': False},

}

class ChatGPTCodereviewPipe(Pipe):

    MAX_TOKENS = 4096  # 모델에 따라 최대 토큰 수 설정

    def __init__(self, *args, **kwargs):

        super().__init__(*args, **kwargs)

        self.auth_method_bitbucket = self.resolve_auth()

        # Bitbucket

        self.workspace = os.getenv('BITBUCKET_WORKSPACE')

        self.repo_slug = os.getenv('BITBUCKET_REPO_SLUG')

        self.bitbucket_client = BitbucketApiService(

            self.auth_method_bitbucket, self.workspace, self.repo_slug)

        # ChatGPT

        self.open_api_key = self.get_variable('OPENAI_API_KEY')

        self.organization = self.get_variable('ORGANIZATION')

        self.model = self.get_variable('MODEL')

        self.user_message_content = self.get_variable('MESSAGE')

        self.files_to_review = self.get_variable("FILES_TO_REVIEW")

        self.exclude_files = self.get_variable("EXCLUDE_FILES")  # 제외할 파일 변수

        self.completion_parameters_payload_file = self.get_variable('CHATGPT_COMPLETION_FILEPATH')

        self.chatgpt_parameters_payload_file = self.get_variable('CHATGPT_CLIENT_FILEPATH')

        self.chat_gpt_client = None

    def get_diffs_to_review(self, pull_request_id):

        diffs_text = self.bitbucket_client.get_pull_request_diffs(pull_request_id)

        files_to_review = []

        if self.files_to_review and self.files_to_review.split(','):

            files_to_review = self.files_to_review.split(',')

        exclude_files = []

        if self.exclude_files and self.exclude_files.split(','):

            exclude_files = self.exclude_files.split(',')

        diffs = self.bitbucket_client.fetch_diffs(diffs_text, files_to_review, self.bitbucket_client.DIFF_DELIMITER)

        # 제외할 파일 필터링

        if exclude_files:

            diffs = [diff for diff in diffs if not any(exclude_file in diff for exclude_file in exclude_files)]

        return diffs

    def split_messages_into_batches(self, messages, model, max_tokens):

        """메시지를 최대 토큰 수에 맞게 여러 배치로 분할"""

        batches = []

        current_batch = []

        current_tokens = 0

        for message in messages:

            message_tokens = self.chat_gpt_client.num_tokens_from_messages([message], model)

            if current_tokens + message_tokens > max_tokens:

                batches.append(current_batch)

                current_batch = []

                current_tokens = 0

            current_batch.append(message)

            current_tokens += message_tokens

        if current_batch:

            batches.append(current_batch)

        return batches

    def get_suggestions(self, diffs_to_review):

        messages = []

        default_messages_system = {

            "role": "system",

            "content": DEFAULT_SYSTEM_PROMPT_FOR_CODE_REVIEW

        }

        messages.append(default_messages_system)

        if self.user_message_content:

            messages.append({"role": "system", "content": self.user_message_content})

        default_messages_diffs = {

            "role": "user",

            "content": str(diffs_to_review)

        }

        messages.append(default_messages_diffs)

        # 메시지를 여러 배치로 분할

        message_batches = self.split_messages_into_batches(messages, self.model, self.MAX_TOKENS)

        all_suggestions = {}

        for batch in message_batches:

            self.log_info(f"Processing ChatGPT batch with {len(batch)} messages...")

            completion_params = {

                'model': self.model,

                'messages': batch,

            }

            # get payload with params for completion

            if self.completion_parameters_payload_file:

                users_completion_params = self.load_yaml(self.completion_parameters_payload_file)

                self.log_info(f"ChatGPT configuration: completion parameters: {users_completion_params}")

                completion_params.update(users_completion_params)

            self.log_info(f"ChatGPT configuration: messages: {batch}")

            self.log_info("Processing ChatGPT...")

            start_time = time.time()

            completion = None

            try:

                completion = self.chat_gpt_client.create_completion(**completion_params)

            except BadRequestError as error:

                self.fail(f"{str(error)}")

            end_time = time.time()

            self.log_debug(completion)

            self.log_info(f"Processing ChatGPT takes: {round(end_time - start_time)} seconds")

            self.log_info(f'ChatGPT completion tokens: {completion.usage}')

            raw_suggestions = completion.choices[0].message.content

            self.log_debug(raw_suggestions)

            try:

                suggestions = self.chat_gpt_client.fetch_json(raw_suggestions)

                all_suggestions.update(suggestions)

            except json.JSONDecodeError as error:

                self.fail(str(error))

            self.log_debug(all_suggestions)

        return all_suggestions

@Oleksandr Kyrdan

[Oleksandr Kyrdan]

Hi @윤주오 

Thanks for your contribution to the community and feedback!

We'll discuss this ideas with the team and notify you.

Also, pipe are open source and your are always welcome to make PR directly to the bitbucket-chatgpt-codereview pipe's repository.

Best regards,
Oleksandr Kyrdan