# Libraries
import json
import os
import re
import util.dazedwrap as dazedwrap
import threading
import time
import traceback
import tiktoken
import openai
from pathlib import Path
from colorama import Fore
from dotenv import load_dotenv
from retry import retry
from tqdm import tqdm

# Open AI
load_dotenv()
if os.getenv("api").replace(" ", "") != "":
    openai.base_url = os.getenv("api")
openai.organization = os.getenv("org")
openai.api_key = os.getenv("key")

# Globals
MODEL = os.getenv("model")
TIMEOUT = int(os.getenv("timeout"))
LANGUAGE = os.getenv("language").capitalize()
PROMPT = Path("prompt.txt").read_text(encoding="utf-8")
VOCAB = Path("vocab.txt").read_text(encoding="utf-8")
THREADS = int(os.getenv("threads"))
LOCK = threading.Lock()
WIDTH = int(os.getenv("width"))
LISTWIDTH = int(os.getenv("listWidth"))
NOTEWIDTH = 70
MAXHISTORY = 10
ESTIMATE = ""
TOKENS = [0, 0]
NAMESLIST = []
FILENAME = None
NAMES = False  # Output a list of all the character names found
BRFLAG = False  # If the game uses <br> instead
FIXTEXTWRAP = True  # Overwrites textwrap
IGNORETLTEXT = False  # Ignores all translated text.
MISMATCH = []  # Lists files that throw a mismatch error (Length of GPT list response is wrong)
PBAR = None

# tqdm Globals
BAR_FORMAT = "{l_bar}{bar:10}{r_bar}{bar:-10b}"
POSITION = 0
LEAVE = False
PBAR = None

# Regex - Need to change this if you want to translate from/to other languages. Default is Japanese Regex
LANGREGEX = r"[一-龠ぁ-ゔァ-ヴーａ-ｚＡ-Ｚ０-９\uFF61-\uFF9F]+"

# Pricing - Depends on the model https://openai.com/pricing
# Batch Size - GPT 3.5 Struggles past 15 lines per request. GPT4 struggles past 50 lines per request
# If you are getting a MISMATCH LENGTH error, lower the batch size.
if "gpt-3.5" in MODEL:
    INPUTAPICOST = 3.00
    OUTPUTAPICOST = 5.00
    BATCHSIZE = 10
    FREQUENCY_PENALTY = 0.2
elif "gpt-4" in MODEL:
    INPUTAPICOST = 2.0
    OUTPUTAPICOST = 8.00
    BATCHSIZE = 30
    FREQUENCY_PENALTY = 0.05
elif "deepseek" in MODEL:
    INPUTAPICOST = 0.27	
    OUTPUTAPICOST = 1.10
    BATCHSIZE = 30
    FREQUENCY_PENALTY = 0.05
else:
    INPUTAPICOST = float(os.getenv("input_cost"))
    OUTPUTAPICOST = float(os.getenv("output_cost"))
    BATCHSIZE = int(os.getenv("batchsize"))
    FREQUENCY_PENALTY = float(os.getenv("frequency_penalty"))


def handleJSON(filename, estimate):
    global ESTIMATE, totalTokens
    ESTIMATE = estimate

    if estimate:
        start = time.time()
        translatedData = openFiles(filename)

        # Print Result
        end = time.time()
        tqdm.write(getResultString(translatedData, end - start, filename))
        with LOCK:
            TOKENS[0] += translatedData[1][0]
            TOKENS[1] += translatedData[1][1]

        return getResultString(["", TOKENS, None], end - start, "TOTAL")

    else:
        try:
            with open("translated/" + filename, "w", encoding="utf-8", newline="\n") as outFile:
                start = time.time()
                translatedData = openFiles(filename)

                # Print Result
                end = time.time()
                json.dump(translatedData[0], outFile, ensure_ascii=False, indent=4)
                tqdm.write(getResultString(translatedData, end - start, filename))
                with LOCK:
                    TOKENS[0] += translatedData[1][0]
                    TOKENS[1] += translatedData[1][1]
        except Exception:
            return "Fail"

    return getResultString(["", TOKENS, None], end - start, "TOTAL")


def openFiles(filename):
    with open("files/" + filename, "r", encoding="UTF-8-sig") as f:
        data = json.load(f)

        # Map Files
        if ".json" in filename:
            translatedData = parseJSON(data, filename)

        else:
            raise NameError(filename + " Not Supported")

    return translatedData


def getResultString(translatedData, translationTime, filename):
    # File Print String
    totalTokenstring = (
        Fore.YELLOW + "[Input: " + str(translatedData[1][0]) + "]"
        "[Output: "
        + str(translatedData[1][1])
        + "]" "[Cost: ${:,.4f}".format(((translatedData[1][0] / 1000000) * INPUTAPICOST) + ((translatedData[1][1] / 1000000) * OUTPUTAPICOST))
        + "]"
    )
    timeString = Fore.BLUE + "[" + str(round(translationTime, 1)) + "s]"

    if translatedData[2] == None:
        # Success
        return filename + ": " + totalTokenstring + timeString + Fore.GREEN + " \u2713 " + Fore.RESET

    else:
        # Fail
        try:
            raise translatedData[2]
        except Exception as e:
            traceback.print_exc()
            errorString = str(e) + Fore.RED
            return filename + ": " + totalTokenstring + timeString + Fore.RED + " \u2717 " + errorString + Fore.RESET


def parseJSON(data, filename):
    totalTokens = [0, 0]
    totalLines = 0
    totalLines = len(data)
    global LOCK, PBAR

    with tqdm(bar_format=BAR_FORMAT, position=POSITION, total=totalLines, leave=LEAVE) as pbar:
        pbar.desc = filename
        PBAR = pbar
        try:
            result = translateJSON(data, [])
            totalTokens[0] += result[0]
            totalTokens[1] += result[1]
        except Exception as e:
            traceback.print_exc()
            return [data, totalTokens, e]
    return [data, totalTokens, None]


def translateJSON(data, translatedList):
    global LOCK, ESTIMATE, FILENAME, PBAR, MISMATCH
    if translatedList:
        stringList = translatedList[0]
    else:
        stringList = []
    tokens = [0, 0]
    speaker = ""
    i = 0

    while i < len(data):
        speakerKey = "character_nameText"
        messageKey = "m_text"

        # Speaker
        if speakerKey in data[i] and data[i][speakerKey]:
            # Grab and TL
            speaker = data[i][speakerKey]
            response = getSpeaker(speaker)
            speaker = response[0]
            tokens[0] += response[1][0]
            tokens[1] += response[1][1]

            # Set Speaker
            data[i][speakerKey] = speaker

        # Dialogue
        if messageKey in data[i]\
            and data[i][messageKey].strip()\
            and data[i][messageKey] != "a"\
            and data[i][messageKey].replace("\u3000", "").strip() != "":
            jaString = data[i][messageKey]

            # Save Original String
            originalString = jaString

            # If there isn't any Japanese in the text just skip
            if not re.search(LANGREGEX, jaString):
                i += 1
                continue

            # Pass 1
            if not translatedList:
                # Strip Spaces
                jaString = jaString.strip()

                # Remove Textwrap
                # jaString = jaString.replace('\n', ' ')

                if jaString:
                    if speaker:
                        stringList.append(f"[{speaker}]: {jaString}")
                    else:
                        stringList.append(jaString)

            # Pass 2
            else:
                # Get Text
                if stringList:
                    # Grab and Pop
                    translatedText = stringList[0]
                    stringList.pop(0)

                    # Set to None if empty list
                    if len(stringList) <= 0:
                        stringList = None

                    # Remove speaker
                    match = re.search(r'(^\[.+?\]\s?[|:]\s?)', translatedText)
                    if match:
                        translatedText = translatedText.replace(match.group(1), "") 

                    # Escape Quotes
                    translatedText = re.sub(r'(?<!\\)"', r"", translatedText)

                    # Remove characters that may break scripts
                    # translatedText = translatedText.replace("<", "(")
                    # translatedText = translatedText.replace(">", ")")
                    translatedText = translatedText.replace("『", "")
                    translatedText = translatedText.replace("』", "")

                    # Remove GPT ' Quotes
                    if translatedText:
                        if translatedText[0] == "'":
                            translatedText = translatedText[1:]
                        if translatedText[-1] == "'":
                            translatedText = translatedText[:-1]
                    else:
                        print("Warning: Empty Translation for", originalString)

                    # Textwrap
                    # translatedText = dazedwrap.wrapText(translatedText, width=WIDTH)  
                    
                    # Set Data
                    if "『" in data[i][messageKey] and "』" not in translatedText:
                        data[i][messageKey] = data[i][messageKey].replace(originalString, f"『{translatedText}』")
                    else:
                        data[i][messageKey] = data[i][messageKey].replace(originalString, f"{translatedText}")     
        # Next Value
        i += 1       

    # EOF
    if not translatedList:
        stringListTL = []

        # String List
        if stringList:
            PBAR.total = len(stringList)
            PBAR.refresh()
            response = translateGPT(stringList, "Reply with the English Translation", True)
            tokens[0] += response[1][0]
            tokens[1] += response[1][1]
            stringListTL = response[0]

            if len(stringList) != len(stringListTL):
                # Mismatch
                with LOCK:
                    if FILENAME not in MISMATCH:
                        MISMATCH.append(FILENAME)

        # Set Strings
        translateJSON(data, [stringListTL])
    return tokens

# Save some money and enter the character before translation
def getSpeaker(speaker):
    match speaker:
        case "ファイン":
            return ["Fine", [0, 0]]
        case "":
            return ["", [0, 0]]
        case _:
            # Find Speaker
            for i in range(len(NAMESLIST)):
                if speaker == NAMESLIST[i][0]:
                    return [NAMESLIST[i][1], [0, 0]]

            # Translate and Store Speaker
            response = translateGPT(
                f"{speaker}",
                "Reply with the " + LANGUAGE + " translation of the NPC name.",
                False,
            )
            response[0] = response[0].title()
            response[0] = response[0].replace("'S", "'s")
            response[0] = response[0].replace("Speaker: ", "")

            # Retry if name doesn't translate for some reason
            if re.search(r"([a-zA-Z？?])", response[0]) == None:
                response = translateGPT(
                    f"{speaker}",
                    "Reply with the " + LANGUAGE + " translation of the NPC name.",
                    False,
                )
                response[0] = response[0].title()
                response[0] = response[0].replace("'S", "'s")

            speakerList = [speaker, response[0]]
            NAMESLIST.append(speakerList)
            return response
    return [speaker, [0, 0]]


def batchList(input_list, batch_size):
    if not isinstance(batch_size, int) or batch_size <= 0:
        raise ValueError("batch_size must be a positive integer")

    return [input_list[i : i + batch_size] for i in range(0, len(input_list), batch_size)]


def parseVocabWithCategories(vocabText):
    """Parse vocabulary text and extract terms with their categories."""
    pairs = []
    seen = set()
    currentCategory = None
    
    for line in vocabText.splitlines():
        line = line.strip()
        if not line or line.startswith('```'):
            continue
        
        # Check if this is a category header
        if line.startswith('#'):
            currentCategory = line
            continue
            
        # Parse vocabulary term
        m = re.match(r'^(.+?)(?:\s?[\(–])', line)  # term is everything before space + '(' or '–'
        if m:
            term = m.group(1)
            if term not in seen:
                pairs.append((term, line, currentCategory))
                seen.add(term)
    
    return pairs


def buildMatchedVocabText(vocabPairs, subbedT):
    """Build formatted vocabulary text with matched terms organized by category."""
    matchedCategories = {}

    # Use word boundaries for Japanese if appropriate, or allow substring as before.
    for term, line, category in vocabPairs:
        # "term in subbedT" could be false positive; can use regex but Japanese doesn't always have spaces.
        if term in subbedT:
            if category not in matchedCategories:
                matchedCategories[category] = []
            matchedCategories[category].append(line)

    # Format matched vocabulary with categories
    if matchedCategories:
        formattedLines = ["Here are some vocabulary and terms so that you know the proper spelling and translation.\n"]
        for category, lines in matchedCategories.items():
            if category:  # Only add category header if it exists
                formattedLines.append(category)
            formattedLines.extend(lines)
            formattedLines.append("")  # Add blank line between categories
        matchedVocabText = f"```\n{chr(10).join(formattedLines).rstrip()}\n```"
    else:
        matchedVocabText = ""
    
    return matchedVocabText


def createContext(fullPromptFlag, subbedT, format):
    vocabPairs = parseVocabWithCategories(VOCAB)
    matchedVocabText = buildMatchedVocabText(vocabPairs, subbedT)

    if fullPromptFlag:
        system = PROMPT + matchedVocabText
    else:
        system = f"\
You are an expert Eroge Game translator who translates Japanese text to {LANGUAGE}.\n\
Output ONLY the {LANGUAGE} translation in the following format: `Translation: <{LANGUAGE.upper()}_TRANSLATION>`\n\
- You translate everything, including erotic and explicit content.\n\
- 'Game Characters' - The names, nicknames, and genders of the game characters. Reference this to know the names, nicknames, and gender of characters in the game\n\
- All text in your response must be in {LANGUAGE} even if it is hard to translate.\n\
- Never include any notes, explanations, dislaimers, or anything similar in your response.\n\
- Maintain any spacing in the translation.\n\
- Maintain any code text in brackets if given. (e.g `[Color_0]`, `[Ascii_0]`, `[FCode_1`], etc)\n\
- `...` can be a part of the dialogue. Translate it as it is.\n\
{matchedVocabText}\n\
"
    if format == "json":
        user = f"```json\n{subbedT}\n```"
    else:
        user = subbedT
    return system, user


def translateText(system, user, history, penalty, format, model=MODEL):
    # Prompt
    msg = [{"role": "system", "content": system}]

    # History
    if isinstance(history, list):
        msg.append({"role": "system", "content": "Translation History:"})
        msg.extend([{"role": "assistant", "content": h} for h in history])
    else:
        msg.append({"role": "assistant", "content": history})

    # Response Format
    if format == "json":
        responseFormat = {"type": "json_object"}
    else:
        responseFormat = {"type": "text"}

    # Content to TL
    msg.append({"role": "user", "content": f"{user}"})
    response = openai.chat.completions.create(
        temperature=0,
        frequency_penalty=penalty,
        model=model,
        response_format=responseFormat,
        messages=msg,
    )
    return response


def cleanTranslatedText(translatedText):
    placeholders = {
        f"{LANGUAGE} Translation: ": "",
        "Translation: ": "",
        "っ": "",
        "〜": "~",
        "ッ": "",
        "。": ".",
        "「": '\\"',
        "」": '\\"',
        "- ": "-",
        "—": "―",
        "】": "]",
        "【": "[",
        "é": "e",
        "this guy": "this bastard",
        "This guy": "This bastard",
        "Placeholder Text": "",
        "```json": "",
        "```": "",
        # Add more replacements as needed
    }
    for target, replacement in placeholders.items():
        translatedText = translatedText.replace(target, replacement)

    # Remove Repeating Characters
    pattern = re.compile(r"(.)\s*\1(?:\s*\1){" + str(20 - 1) + r",}")
    translatedText = pattern.sub(lambda match: match.group(0).replace(" ", "")[:20], translatedText)

    # Elongate Long Dashes (Since GPT Ignores them...)
    translatedText = elongateCharacters(translatedText)
    return translatedText


def elongateCharacters(text):
    # Define a pattern to match one character followed by one or more `ー` characters
    # Using a positive lookbehind assertion to capture the preceding character
    pattern = r"(?<=(.))ー+"

    # Define a replacement function that elongates the captured character
    def repl(match):
        char = match.group(1)  # The character before the ー sequence
        count = len(match.group(0)) - 1  # Number of ー characters
        return char * count  # Replace ー sequence with the character repeated

    # Use re.sub() to replace the pattern in the text
    return re.sub(pattern, repl, text)


def extractTranslation(translatedTextList, is_list):
    try:
        translatedTextList = re.sub(r'\\"+\"([^,\n}])', r'\\"\1', translatedTextList)
        translatedTextList = re.sub(r"(?<![\\])\"+(?![\n,])", r'"', translatedTextList)
        line_dict = json.loads(translatedTextList)
        # If it's a batch (i.e., list), extract with tags; otherwise, return the single item.
        string_list = list(line_dict.values())
        if is_list:
            return string_list
        else:
            return string_list[0]

    except Exception as e:
        PBAR.write(f"extractTranslation Error: {e} on String {translatedTextList}")
        return None


def countTokens(system, user, history):
    inputTotalTokens = 0
    outputTotalTokens = 0
    enc = tiktoken.encoding_for_model("gpt-4")

    # Input
    if isinstance(history, list):
        for line in history:
            inputTotalTokens += len(enc.encode(line))
    else:
        inputTotalTokens += len(enc.encode(history))
    inputTotalTokens += len(enc.encode(system))
    inputTotalTokens += len(enc.encode(user))

    # Output
    outputTotalTokens += round(len(enc.encode(user)) * 2.5)

    return [inputTotalTokens, outputTotalTokens]


@retry(exceptions=Exception, tries=5, delay=5)
def translateGPT(text, history, fullPromptFlag):
    global PBAR, MISMATCH, FILENAME
    if text:
        with open("log/translationHistory.txt", "a+", encoding="utf-8") as logFile:
            mismatch = False
            totalTokens = [0, 0]
            if isinstance(text, list):
                format = "json"
                tList = batchList(text, BATCHSIZE)
            else:
                format = "text"
                tList = [text]

            for index, tItem in enumerate(tList):
                # Things to Check before starting translation
                if not re.search(LANGREGEX, str(tItem)):
                    if PBAR is not None:
                        PBAR.update(len(tItem))
                    if isinstance(tItem, list):
                        for j in range(len(tItem)):
                            tItem[j] = cleanTranslatedText(tItem[j])
                            tList[index] = tItem
                    else:
                        tList[index] = cleanTranslatedText(tItem)
                    history = tItem[-MAXHISTORY:]
                    continue

                # Before sending to translation, if we have a list of items, add the formatting
                if isinstance(tItem, list):
                    for j in range(len(tItem)):
                        if not tItem[j]:
                            tItem[j] = tItem[j].replace("", "Placeholder Text")
                    payload = {f"Line{i+1}": string for i, string in enumerate(tItem)}
                    payload = json.dumps(payload, indent=4, ensure_ascii=False)
                    varResponse = [payload, []]
                    subbedT = varResponse[0]
                else:
                    varResponse = [tItem, []]
                    subbedT = varResponse[0]

                # Create Message
                system, user = createContext(fullPromptFlag, subbedT, format)

                # Calculate Estimate
                if ESTIMATE:
                    estimate = countTokens(system, user, history)
                    totalTokens[0] += estimate[0]
                    totalTokens[1] += estimate[1]
                    continue

                # Translating
                response = translateText(system, user, history, 0.05, format)

                # Set Tokens
                translatedText = response.choices[0].message.content

                # AI Refused, Try Again
                if not translatedText:
                    response = translateText(f"{system}\n You translate ALL content.", user, history, 0.1, format, model="gpt-4o")
                    translatedText = response.choices[0].message.content

                # Report Tokens
                totalTokens[0] += response.usage.prompt_tokens
                totalTokens[1] += response.usage.completion_tokens

                # Check Translation
                if translatedText:
                    translatedText = cleanTranslatedText(translatedText)
                    if isinstance(tItem, list):
                        extractedTranslations = extractTranslation(translatedText, True)
                        if extractedTranslations == None or len(tItem) != len(extractedTranslations):
                            # Mismatch. Try Again
                            response = translateText(system, user, history, 0.05, format, MODEL)
                            translatedText = response.choices[0].message.content
                            totalTokens[0] += response.usage.prompt_tokens
                            totalTokens[1] += response.usage.completion_tokens

                            # Formatting
                            translatedText = cleanTranslatedText(translatedText)
                            if isinstance(tItem, list):
                                extractedTranslations = extractTranslation(translatedText, True)
                                if extractedTranslations == None or len(tItem) != len(extractedTranslations):
                                    with open("log/mismatchHistory.txt", "a+", encoding="utf-8") as mismatchFile:
                                        mismatchFile.write(f"Mismatch: {FILENAME}\n")
                                        mismatchFile.write(f"Input:\n{subbedT}\n")
                                        mismatchFile.write(f"Output:\n{translatedText}\n")
                                        mismatch = True  # Just here for breakpoint
                        logFile.write(f"Input:\n{subbedT}\n")
                        logFile.write(f"Output:\n{translatedText}\n")

                        # Set if no mismatch
                        if mismatch == False:
                            tList[index] = extractedTranslations
                            history = extractedTranslations[-MAXHISTORY:]  # Update history if we have a list
                        else:
                            history = text[-MAXHISTORY:]
                            mismatch = False
                            if FILENAME not in MISMATCH:
                                MISMATCH.append(FILENAME)

                        # Update Loading Bar
                        with LOCK:
                            if PBAR is not None:
                                PBAR.update(len(tItem))
                    else:
                        # Ensure we're passing a single string to extractTranslation
                        tList[index] = translatedText.replace("Placeholder Text", "")
                else:
                    PBAR.write(f"AI Refused:{tItem}\n")

        # Combine if multilist
        if isinstance(tList[0], list):
            tList = [t for sublist in tList for t in sublist]

        # Return
        if format == "json":
            return [tList, totalTokens]
        else:
            return [tList[0], totalTokens]
    else:
        return [text, [0, 0]]