import csv
import pandas as pd
import os
from openai import OpenAI
import sys



   
def classify_lines(input_file_path, audit_ai_csv) -> list[str]:
    from scriptAudit.utils import remove_empty_content,remove_leading_numbers, remove_numeric_only_content,remove_emptyline_rows,merge_consecutive_action_lines,merge_consecutive_dialogue_lines
    from scriptAudit.utils import  insert_blank_lines,add_fade_in_out,remove_asterisks, merge_consecutive_action_lines_new,extract_labeled_lines, remove_trailing_speaker

    with open(input_file_path, "r") as f:
        raw_lines = [line.strip() for line in f.readlines() if line.strip()]
        
    chunked_results = []
    client = OpenAI(api_key=os.getenv('openai_key'))
    prompt = (
        "You are a screenplay assistant. For each line below, classify it using one of these labels: "
        "slugline, speaker, dialogue, action, parenthetical, transition. Return each line followed by its label in curly braces.\n\n"
    )
    for i in range(0, len(raw_lines), 20):
        chunk = raw_lines[i:i + 20]
        final_prompt = prompt + "\n".join(chunk)
        response = client.responses.create(
                    model="gpt-4o",
                    input=[
                        {
                            "role": "developer",
                            "content": """You are a screenplay auditor. For each line below, classify it using one of these labels: 
                            slugline, speaker, dialogue, action, parenthetical, transition, special_term, title. Return each line followed by its label in curly braces.\n\n
                            **Examples:**
                            INT. ROOM – NIGHT {slugline}
                            KITCHEN – DAY {slugline}
                            JOHN {speaker}
                            (quietly) {parenthetical}
                            JOHN (O.S.) {speaker}
                            JOHN (angrily) {speaker}
                            I knew you’d come. {dialogue}
                            She turns away from the window. {action}
                            FADE OUT. {transition}
                            THE END {title}
                            (V.O.) {special_term}
                            John CONT'D {speaker}"""
                        },
                        {
                            "role": "user",
                            "content": "I need you to classify the lines below. Please provide the classification in the format: 'line {label}'\n\n" + "\n".join(chunk)
                        }
                    ]
                )
        
        classified = response.output_text.splitlines()
        chunked_results.extend(classified)
        extracted = extract_labeled_lines(chunked_results)
        with open(audit_ai_csv, mode='w', newline='', encoding='utf-8') as f:
            writer = csv.writer(f)
            writer.writerow(["content", "script_element"])
            writer.writerows(extracted)
    print("Classification completed.")
    
    df = pd.read_csv(audit_ai_csv)
    df = remove_empty_content(df)
    df = remove_asterisks(df)
    df = remove_leading_numbers(df)
    df = remove_numeric_only_content(df)
    df = remove_emptyline_rows(df)
    df = remove_trailing_speaker(df)
    # df = merge_consecutive_action_lines(df)
    df = merge_consecutive_action_lines_new(df)
    df = merge_consecutive_dialogue_lines(df)
    df = insert_blank_lines(df)
    
    df = add_fade_in_out(df)

    return df