visual_debugger.py

import os
import time
import base64


class VisualDebugger:
    def __init__(self, llm_client, model_name="gpt-5"):
        self.llm = llm_client
        self.model_name = model_name
        self.last_completion = None
        self.last_latency = None
        self.last_status = None

        self.expected_behaviors = {
            "Oscillator": "The output (Vout) should be a periodic waveform with stable amplitude.",
            "Schmitt": "The output (Vout) should be a square wave with hysteresis.",
            "Integrator": "The output should be the time integral of the input.",
            "Differentiator": "The output should be the time derivative of the input.",
            "LowPass": "High gain at low frequencies, roll-off at high frequencies.",
            "HighPass": "Attenuation at low frequencies, high gain at high frequencies.",
            "Amplifier": "The output should be an amplified undistorted version of the input.",
            "Opamp": "High open-loop gain and stable DC behavior."
        }

    def _encode_image(self, image_path):
        if not image_path or not os.path.exists(image_path):
            return None
        try:
            with open(image_path, "rb") as image_file:
                return base64.b64encode(image_file.read()).decode("utf-8")
        except:
            return None

    # 新增方法：为了适配 DesignOptimizer 的调用
    def construct_debug_message(self, image_path, error_info, task_description):
        base64_image = self._encode_image(image_path)
        return {
            "has_image": True if base64_image else False,
            "content_text": f"Task: {task_description}\nSimulation Info: {error_info}",
            "content_image": base64_image
        }

    def debug(self, image_path, task_description):
        self.last_completion = None
        self.last_latency = None
        self.last_status = None

        base64_image = self._encode_image(image_path)
        if not base64_image:
            self.last_status = "no_image"
            return "No waveform image found."

        expected_behavior = "The output should match expected circuit behavior."
        for key, desc in self.expected_behaviors.items():
            if key.lower() in str(task_description).lower():
                expected_behavior = desc
                break

        prompt_text = (
            f"Circuit Type: {task_description}\n"
            f"Expected Behavior: {expected_behavior}\n"
            f"Functional verification failed.\n\n"
            f"Analyze the waveform image and provide:\n"
            f"1) Visual analysis\n"
            f"2) Diagnosis\n"
            f"3) Suggested parameter or topology fixes\n"
        )

        t0 = time.time()

        try:
            # 🔹 统一使用 OpenAI 标准接口
            # 不再区分 gpt-5 还是 local，只要 self.llm 是 OpenAI 客户端，调用方式就一致
            response = self.llm.chat.completions.create(
                model=self.model_name,
                messages=[
                    {
                        "role": "user",
                        "content": [
                            {"type": "text", "text": prompt_text},
                            {
                                "type": "image_url",
                                "image_url": {
                                    "url": f"data:image/png;base64,{base64_image}"
                                },
                            },
                        ],
                    }
                ],
                max_tokens=1024,
                temperature=0.01,
            )

            # 获取内容的路径固定为 choices[0].message.content
            content = response.choices[0].message.content

            self.last_latency = time.time() - t0
            self.last_completion = response
            self.last_status = "ok"

            return content

        except Exception as e:
            self.last_latency = time.time() - t0
            self.last_status = "error"
            return f"Visual analysis failed: {str(e)}"