import numpy as np
import re


def split_markdown_by_title(markdown_file):
    with open(markdown_file, 'r', encoding='utf-8') as f:
        content = f.read()

    re_str = "# cola|# mnli|# mrpc|# qnli|# qqp|# rte|# sst2|# wnli|# mmlu|# squad_v2|# iwslt|# un_multi|# math"
    
    datasets = ["# cola", "# mnli", "# mrpc", "# qnli", "# qqp", "# rte", "# sst2", "# wnli",
                "# mmlu", "# squad_v2", "# iwslt", "# un_multi", "# math"]
    
    # re_str = "# cola|# mnli|# mrpc|# qnli|# qqp|# rte|# sst2|# wnli"
    # datasets = ["# cola", "# mnli", "# mrpc", "# qnli", "# qqp", "# rte", "# sst2", "# wnli"]
    primary_sections = re.split(re_str, content)[1:]
    assert len(primary_sections) == len(datasets)

    all_sections_dict = {}

    for dataset, primary_section in zip(datasets, primary_sections):
        re_str = "## "
        results = re.split(re_str, primary_section)
        keywords = ["10 prompts", "bertattack", "checklist", "deepwordbug", "stresstest",
                    "textfooler", "textbugger", "translation"]

        secondary_sections_dict = {}
        for res in results:
            for keyword in keywords:
                if keyword in res.lower():
                    secondary_sections_dict[keyword] = res
                    break

        all_sections_dict[dataset] = secondary_sections_dict

    return all_sections_dict
# def prompts_understanding(sections_dict):
#     for dataset in sections_dict.keys():
#         # print(dataset)
#         for title in sections_dict[dataset].keys():
#             if title == "10 prompts":
#                 prompts = sections_dict[dataset][title].split("\n")
#                 num = 0
#                 task_prompts_acc = []
#                 role_prompts_acc = []
#                 for prompt in prompts:
#                     if "Acc: " not in prompt:
#                         continue
#                     else:
#                         import re
#                         num += 1
#                         match = re.search(r'Acc: (\d+\.\d+)%', prompt)
#                         if match:
#                             number = float(match.group(1))
#                         if num <= 10:
#                             task_prompts_acc.append(number)
#                         else:
#                             role_prompts_acc.append(number)

#                 print(task_prompts_acc)
#                 print(role_prompts_acc)
import os
def list_files(directory):
    files = [os.path.join(directory, d) for d in os.listdir(directory) if not os.path.isdir(os.path.join(directory, d))]
    return files

def convert_model_name(attack):
    attack_name = {
        "T5": "t5",
        "UL2": "ul2",
        "Vicuna": "vicuna",
        "ChatGPT": "chatgpt",
    }
    return attack_name[attack]

def convert_attack_name(attack):
    attack_name = {
        "BertAttack": "bertattack",
        "CheckList": "checklist",
        "DeepWordBug": "deepwordbug",
        "StressTest": "stresstest",
        "TextFooler": "textfooler",
        "TextBugger": "textbugger",
        "Semantic": "translation",
    }
    return attack_name[attack]

def convert_dataset_name(dataset):
    dataset_name = {
        "CoLA": "# cola",
        "MNLI": "# mnli",
        "MRPC": "# mrpc",
        "QNLI": "# qnli",
        "QQP": "# qqp",
        "RTE": "# rte",
        "SST-2": "# sst2",
        "WNLI": "# wnli",
        "MMLU": "# mmlu",
        "SQuAD V2": "# squad_v2",
        "IWSLT": "# iwslt",
        "UN Multi": "# un_multi",
        "Math": "# math",
        "Avg": "Avg",
    }
    return dataset_name[dataset]


def retrieve(model_name, dataset_name, attack_name, prompt_type):
    model_name = convert_model_name(model_name)
    dataset_name = convert_dataset_name(dataset_name)
    attack_name = convert_attack_name(attack_name)

    if "zero" in prompt_type:
        shot = "zeroshot"
    else:
        shot = "fewshot"
    
    if "task" in prompt_type:
        prompt_type = "task"
    else:
        prompt_type = "role"
    
    directory_path = "./db"
    md_dir = os.path.join(directory_path, model_name + "_" + shot + ".md")
    sections_dict = split_markdown_by_title(md_dir)
    results = {}
    for cur_dataset in sections_dict.keys():
        if cur_dataset == dataset_name:
            dataset_dict = sections_dict[cur_dataset]
            for cur_attack in dataset_dict.keys():
                
                if cur_attack == attack_name:

                    if attack_name == "translation":
                        prompts_dict = dataset_dict[attack_name].split("\n")
            
                        for prompt_summary in prompts_dict:
                            if "acc: " not in prompt_summary:
                                continue
                            
                            prompt = prompt_summary.split("prompt: ")[1]
                            
                            import re
                        
                            match_atk = re.search(r'acc: (\d+\.\d+)%', result)
                            number_atk = float(match_atk.group(1))
                            results[prompt] = number_atk
  
                        sorted_results = sorted(results.items(), key=lambda item: item[1])[:6]
      

                        return sorted_results

                    elif attack_name in ["bertattack", "checklist", "deepwordbug", "stresstest", "textfooler", "textbugger"]:

                        prompts_dict = dataset_dict[attack_name].split("\n")
                        num = 0
                        

                        for prompt_summary in prompts_dict:
                            if "Attacked prompt: " not in prompt_summary:
                                continue

                            num += 1
                            import re
                            match_origin = re.search(r'Original acc: (\d+\.\d+)%', prompt_summary)
                            match_atk = re.search(r'attacked acc: (\d+\.\d+)%', prompt_summary)
                        if match_origin and match_atk:
                            number_origin = float(match_origin.group(1))
                            number_atk = float(match_atk.group(1))
                    
                    # print(model_shot, dataset, title, len(summary[attack][dataset]), num)
    
    # for atk in summary.keys():
    #     for dataset in summary[atk].keys():
    #         # if atk == "translation":
    #         print(atk, dataset, len(summary[atk][dataset]))
    #             # print(summary[atk][dataset][:10])
    
    output_dict = {}

    sorted_atk_name = ["TextBugger",  "DeepWordBug", "TextFooler", "BertAttack", "CheckList", "StressTest", "Semantic"]
    sorted_dataset_name = ["SST-2", "CoLA", "QQP", "MRPC", "MNLI", "QNLI", "RTE",  "WNLI", "MMLU", "SQuAD V2", "IWSLT", "UN Multi", "Math"]
    
    for atk in sorted_atk_name:
        output_dict[atk] = {}
        for dataset in sorted_dataset_name:
            output_dict[atk][dataset] = ""
    
    for sorted_atk in sorted_atk_name:
        for attack, dataset_drop_rates in summary.items():
            # attack = convert_attack_name(attack)
            if convert_attack_name(attack) == sorted_atk:
                for sorted_dataset in sorted_dataset_name:
                    for dataset, drop_rates in dataset_drop_rates.items():
                        if convert_dataset_name(dataset) == sorted_dataset:
                            if len(drop_rates) > 0:
                                output_dict[sorted_atk][sorted_dataset] = "{:.2f}".format(sum(drop_rates)/len(drop_rates)) + "\scriptsize{$\pm$" + "{:.2f}".format(np.std(drop_rates)) + "}"
                            else:
                                output_dict[sorted_atk][sorted_dataset] = "-"

                total_drop_rate = summary[attack]["Avg"]
                output_dict[sorted_atk]["Avg"] = "{:.2f}".format(np.mean(total_drop_rate)) + "\scriptsize{$\pm$" + "{:.2f}".format(np.std(total_drop_rate)) + "}"