cli.py

from src.connector import ConnectorFactory
from src.alignment import AlignmentFactory
from openai import AsyncOpenAI
import tomllib
from loguru import logger
import asyncio
import aiohttp

# 以二进制读取模式打开文件
with open("config.toml", "rb") as f:
    conf = tomllib.load(f)

concurrency = conf["system"]["concurrency"]

async def get_newB_async(session, oldB: str, oldA: str, newA: str, TOC:str) -> str:
    if oldA.strip() == newA.strip():
        logger.debug("原文未变化")
        return oldB #diff为none：原文没变化直接返回旧译文
    if newA.strip() == "":
        logger.debug("原文被删除")
        return "" #diff为delete：原文被删除直接返回空
    logger.debug(f"oldB: {oldB[0:30]}\n, oldA: {oldA[0:30]}, newA: {newA[0:30]}")

    def get_diff(oldA: str, newA: str) -> str:
        import difflib
        # 生成统一的diff格式
        diff = difflib.unified_diff(
            oldA.splitlines(),
            newA.splitlines(),
            fromfile='oldA',
            tofile='newA',
            lineterm=''
        )
        return '\n'.join(diff)

    logger.debug(f"oldB: {oldB[0:500]}\n, AAdiff: {get_diff(oldA,newA)[0:500]}")

    prompt = f"你将获得中文站旧版文档的一部分、对应的旧版英文文档和最新版英文文档的差异（diff），请根据这些信息生成对应的最新版中文文档；中文站旧版文档若有英文（未翻译）部分的话，就把其翻译成中文；你的输出将直接作用到文档上，所以不要输出其他任何内容：\n中文站旧版文档：\n{oldB}\n\n对应的英文文档diff：\n{get_diff(oldA,newA)}\n\n最新版中文文档：" #（若旧版使用繁体，最新版也需要使用繁体以避免编辑摘要显示太多修改打扰别的维基人）
    # logger.debug("发送至AI ==> "+prompt)

    client = AsyncOpenAI(base_url=conf["OPENAI-api"]["base_url"],max_retries=3)
    for i in range(3):
        try:
            response = await client.chat.completions.create(
                model=conf["OPENAI-api"]["model"],
                messages=[
                {"role": "system", "content": conf["OPENAI-api"]["system_prompt"] },
                {"role": "user", "content": prompt},
            ],
                max_tokens=8192,
                temperature=0.3,  # 控制生成文本的随机性
                stream=False,

            )
            logger.debug("AI返回 <== "+str(response.choices[0].message.content)[0:50])
            return str(response.choices[0].message.content)
        except Exception as e:
            logger.error(f"OpenAI API 调用失败，正在重试 ({i+1}/{3})... 错误信息: {e}")
            await asyncio.sleep(2)  # 等待 2 秒后重试

    logger.error("达到最大重试次数，OpenAI API 调用失败")
    raise  # 重新抛出异常


alignment = AlignmentFactory.create(type=conf["system"]["alignment"],conf=conf["alignment"][conf["system"]["alignment"]])
connector = ConnectorFactory.create(type=conf["system"]["connector"], conf=conf["connector"][conf["system"]["connector"]])


OldABblocks = alignment.split_AB(connector.get_old_A(), connector.get_old_B())
NewAblocks = alignment.split(connector.get_latest_A())

logger.debug(f"OldABblocks.TOC: {OldABblocks.blockIDs}, NewAblocks.TOC: {NewAblocks.blockIDs}")
NewB = [""] * len(NewAblocks.texts)

async def process_block(session, i, NAtext, OldABblocks, NewAblocks):
    logger.debug(f"进度： {i+1}/{len(NewB)}")
    if NewAblocks.blockIDs[i] in OldABblocks.blockIDs:
        oldABIndex = OldABblocks.blockIDs.index(NewAblocks.blockIDs[i])
        return await get_newB_async(session, OldABblocks.oldB[oldABIndex], OldABblocks.oldA[oldABIndex], NAtext ,NewAblocks.toc_to_str())
    else:
        return await get_newB_async(session, "", "", NAtext ,NewAblocks.toc_to_str())

async def main():
    NewB = [""] * len(NewAblocks.texts)
    semaphore = asyncio.Semaphore(concurrency)  # 创建信号量

    async def process_block_with_semaphore(session, i, NAtext, OldABblocks, NewAblocks):
        async with semaphore:  # 获取信号量
            return await process_block(session, i, NAtext, OldABblocks, NewAblocks)

    async with aiohttp.ClientSession() as session:
        tasks = [process_block_with_semaphore(session, i, NAtext, OldABblocks, NewAblocks) for i, NAtext in enumerate(NewAblocks.texts)]
        NewB = await asyncio.gather(*tasks)

    with open("output/newB", "w", encoding="utf-8") as f:
        for i in range(len(NewB)):
            f.write(connector.generate_latest_B(NewB[i]+"\n"))

asyncio.run(main())