utils.py

import datetime
# Imports required for Google Sheets integration
import gspread
import random
import time
import functools
from gspread.exceptions import SpreadsheetNotFound, APIError
from oauth2client.service_account import ServiceAccountCredentials
import pandas as pd
import json
import gradio as gr
import os

GSERVICE_ACCOUNT_INFO = {
  "type": "service_account",
  "project_id": "txagent",
  "private_key_id": "cc1a12e427917244a93faf6f19e72b589a685e65",
  "private_key": None,
  "client_email": "shanghua@txagent.iam.gserviceaccount.com",
  "client_id": "108950722202634464257",
  "auth_uri": "https://accounts.google.com/o/oauth2/auth",
  "token_uri": "https://oauth2.googleapis.com/token",
  "auth_provider_x509_cert_url": "https://www.googleapis.com/oauth2/v1/certs",
  "client_x509_cert_url": "https://www.googleapis.com/robot/v1/metadata/x509/shanghua%40txagent.iam.gserviceaccount.com",
  "universe_domain": "googleapis.com"
}
GSHEET_NAME = "TxAgent_data_collection"

GSheet_API_KEY = os.environ.get("GSheets_Shanghua_PrivateKey")
if GSheet_API_KEY is None:
    print("GSheet_API_KEY not found in environment variables. Please set it.")
else:
    GSheet_API_KEY = GSheet_API_KEY.replace("\\n", "\n")
    GSERVICE_ACCOUNT_INFO["private_key"] = GSheet_API_KEY

#Exponential backoff retry decorator
def exponential_backoff_gspread(max_retries=30, max_backoff_sec=64, base_delay_sec=1, target_exception=APIError):
    """
    Decorator to implement exponential backoff for gspread API calls.

    Retries a function call if it raises a specific exception (defaults to APIError)
    that matches the Google Sheets API rate limit error (HTTP 429).

    Args:
        max_retries (int): Maximum number of retry attempts.
        max_backoff_sec (int): Maximum delay between retries in seconds.
        base_delay_sec (int): Initial delay in seconds for the first retry.
        target_exception (Exception): The specific exception type to catch.
    """
    def decorator(func):
        @functools.wraps(func)
        def wrapper(*args, **kwargs):
            retries = 0
            while True: # Loop indefinitely until success or max retries exceeded
                try:
                    # Attempt to execute the wrapped function
                    return func(*args, **kwargs)
                except target_exception as e:
                    # Check if the error is the specific 429 Quota Exceeded error
                    # We parse the string representation as gspread's APIError includes the status code there.
                    error_message = str(e)
                    is_rate_limit_error = "[429]" in error_message and (
                        "Quota exceeded" in error_message or "Too Many Requests" in error_message
                    )

                    if is_rate_limit_error:
                        retries += 1
                        if retries > max_retries:
                            print(f"Max retries ({max_retries}) exceeded for {func.__name__}. Last error: {e}")
                            raise e # Re-raise the last exception after exhausting retries

                        # Calculate exponential backoff delay with random jitter (0-1 second)
                        backoff_delay = min(max_backoff_sec, base_delay_sec * (2 ** (retries - 1)) + random.uniform(0, 1))

                        print(
                            f"Rate limit hit for {func.__name__} (Attempt {retries}/{max_retries}). "
                            f"Retrying in {backoff_delay:.2f} seconds. Error: {e}"
                        )
                        time.sleep(backoff_delay)
                    else:
                        # If it's a different kind of APIError (e.g., 403 Forbidden, 404 Not Found), re-raise immediately.
                        print(f"Non-rate-limit APIError encountered in {func.__name__}: {e}")
                        raise e
                except Exception as e:
                    # Catch any other unexpected exceptions during the function execution
                    print(f"An unexpected error occurred in {func.__name__}: {e}")
                    raise e # Re-raise unexpected errors
        return wrapper
    return decorator

#2) Initialize Google Sheets client
# Define the scopes
scope = [
    "https://spreadsheets.google.com/feeds",
    "https://www.googleapis.com/auth/drive",
]

# Authenticate immediately on import
creds = ServiceAccountCredentials.from_json_keyfile_dict(GSERVICE_ACCOUNT_INFO, scope)
client = gspread.authorize(creds)

@exponential_backoff_gspread(max_retries=30, max_backoff_sec=64)
def read_sheet_to_df(custom_sheet_name=None, sheet_index=0):
    """
    Read all data from a Google Sheet into a pandas DataFrame.

    Parameters:
        custom_sheet_name (str): The name of the Google Sheet to open. If None, uses GSHEET_NAME.
        sheet_index (int): Index of the worksheet within the spreadsheet (default is 0, the first sheet).

    Returns:
        pandas.DataFrame: DataFrame containing the sheet data, with the first row used as headers.
    """

    # Determine which sheet name to use
    if custom_sheet_name is None:
        custom_sheet_name = GSHEET_NAME

    # Open the spreadsheet
    try:
        spreadsheet = client.open(custom_sheet_name)
    except gspread.SpreadsheetNotFound:
        return None

    # Select the desired worksheet
    try:
        worksheet = spreadsheet.get_worksheet(sheet_index)
    except IndexError:
        return None

    # Fetch all data: first row as header, remaining as records
    data = worksheet.get_all_records()

    # Convert to DataFrame
    df = pd.DataFrame(data)

    return df

@exponential_backoff_gspread(max_retries=30, max_backoff_sec=64)
def append_to_sheet(user_data=None, custom_row_dict=None, custom_sheet_name=None, add_header_when_create_sheet=False):
    """
    Append a new row to a Google Sheet. If 'custom_row' is provided, append that row.
    Otherwise, append a default row constructed from the provided user_data.
    Ensures that each value is aligned with the correct column header.
    """
    if custom_sheet_name is None:
        custom_sheet_name = GSHEET_NAME
    
    try:
        # Try to open the spreadsheet by name
        spreadsheet = client.open(custom_sheet_name)
        is_new = False
    except SpreadsheetNotFound:
        # If it doesn't exist, create it
        spreadsheet = client.create(custom_sheet_name)
        # Optionally, share the new spreadsheet with designated emails
        spreadsheet.share('shanghuagao@gmail.com', perm_type='user', role='writer')
        spreadsheet.share('rzhu@college.harvard.edu', perm_type='user', role='writer')
        is_new = True

    print("Spreadsheet ID:", spreadsheet.id)
    # Access the first worksheet
    sheet = spreadsheet.sheet1

    # Check if the sheet has any rows yet
    existing_values = sheet.get_all_values()
    is_empty = (existing_values == [[]]) #indicates empty spreadsheet that was cleared in the past

    # --- Always ensure header row is present and get headers ---
    if (is_new or is_empty) and add_header_when_create_sheet:
        # headers come from the keys of our row dict
        if custom_row_dict is not None:
            headers = list(custom_row_dict.keys())
        else:
            headers = list(user_data.keys())
        sheet.append_row(headers)
    else:
        # Read headers from the first row of the sheet
        headers = sheet.row_values(1) if sheet.row_count > 0 else []

    # --- Build row aligned to headers ---
    if custom_row_dict is not None:
        # Ensure all values are aligned to headers, fill missing with ""
        custom_row = [custom_row_dict.get(header, "") for header in headers]
    else:
        # Construct the default row with a timestamp and user_data fields
        custom_row = [str(datetime.datetime.now()), user_data["question"], user_data["final_answer"], user_data["trace"]]
    
    # Append the custom or default row to the sheet
    sheet.append_row(custom_row)

def format_chat(response, tool_database_labels):
    chat_history = []
    # Keep track of the last assistant message's tool_calls
    last_tool_calls = []

    for msg in response:
        if msg["role"] == "assistant":
            content = msg.get("content", "")
            # Extract tool_calls from this assistant message (if any)
            last_tool_calls = json.loads(msg.get("tool_calls", "[]"))
            # Emit the assistant's main message
            chat_history.append(
                gr.ChatMessage(role="assistant", content=content)
            )

        elif msg["role"] == "tool":
            # For each tool response, we pair it with the corresponding call
            for i, tool_call in enumerate(last_tool_calls):
                name = tool_call.get("name", "")
                args = tool_call.get("arguments", {})

                # Determine icon + title
                database_label = ""
                if name == "Tool_RAG":
                    title = "🧰 Tool RAG"
                else:
                    title = f"🛠️ {name}"
                    for db_label, tool_list in tool_database_labels.items():
                        if name in tool_list:
                            title = f"🛠️ {name}\n(**Info** {db_label} [Click to view])"
                            database_label = " (" + db_label + ")"
                            break

                # Parse and pretty-print the tool response content
                raw = msg.get("content", "")
                try:
                    parsed = json.loads(raw)
                    pretty = json.dumps(parsed)
                except json.JSONDecodeError:
                    pretty = raw

                # Add as a single ChatMessage with metadata.title and metadata.log.
                # Display the arguments as the first part of the content, clearly separated from the response,
                # and display the tool response content as contiguous text.
                chat_history.append(
                    gr.ChatMessage(
                        role="assistant",
                        content=f"Tool Response{database_label}:\n{pretty}",
                        metadata={
                            "title": title,
                            "log":  json.dumps(args),
                            "status": 'done'
                        }
                    )
                )

            # Clear after rendering
            last_tool_calls = []

    # if chat_history:
    #     last_msg = chat_history[-1]
    #     if isinstance(last_msg.content, str) and "[FinalAnswer]" in last_msg.content:
    #         # Find the first assistant message
    #         for msg in chat_history:
    #             if msg.role == "assistant" and isinstance(msg.content, str):
    #                 msg.content = "**Reasoning:**\n" + msg.content
    #                 break
    if chat_history:
        last_msg = chat_history[-1]
        if isinstance(last_msg.content, str) and "[FinalAnswer]" in last_msg.content:
            last_msg.content = last_msg.content.replace("[FinalAnswer]", "\n**Answer:**\n")


    final_answer_messages = [gr.ChatMessage(role="assistant", content=chat_history[-1].content.split("\n**Answer:**\n")[-1].strip())]
    assistant_count = sum(1 for msg in chat_history if msg.role == "assistant")
    if assistant_count == 1:
        # If only one assistant message, show "No reasoning conducted."
        reasoning_messages = [gr.ChatMessage(role="assistant", content="No reasoning was conducted.")]
    else:
        # Include ALL messages in reasoning_messages, including the last one
        reasoning_messages = chat_history.copy()


    return final_answer_messages, reasoning_messages, chat_history