自译解报告

feat: 添加wsl2使用windows proxy的方法

.gitattributes

@@ -0,0 +1,5 @@
+*.h linguist-detectable=false
+*.cpp linguist-detectable=false
+*.tex linguist-detectable=false
+*.cs linguist-detectable=false
+*.tps linguist-detectable=false

.gitignore

@@ -131,6 +131,12 @@ dmypy.json
 # Pyre type checker
 .pyre/
 
+.vscode
+
 history
 ssr_conf
 config_private.py
+gpt_log
+private.md
+private_upload
+other_llms

Dockerfile

@@ -1,8 +1,4 @@
-FROM ubuntu:latest
-
-RUN apt-get update && \
-    apt-get install -y python3 python3-pip && \
-    rm -rf /var/lib/apt/lists/*
+FROM python:3.11
 
 RUN echo '[global]' > /etc/pip.conf && \
     echo 'index-url = https://mirrors.aliyun.com/pypi/simple/' >> /etc/pip.conf && \

README.md

@@ -1,8 +1,144 @@
+
+
 # ChatGPT 学术优化
 
-**如果喜欢这个项目，请给它一个Star**
+**如果喜欢这个项目，请给它一个Star；如果你发明了更好用的学术快捷键，欢迎发issue或者pull requests**
 
-## 使用docker
+If you like this project, please give it a Star. If you've come up with more useful academic shortcuts, feel free to open an issue or pull request.
+
+```
+代码中参考了很多其他优秀项目中的设计，主要包括：
+
+# 借鉴项目1：借鉴了ChuanhuChatGPT中读取OpenAI json的方法、记录历史问询记录的方法以及gradio queue的使用技巧
+https://github.com/GaiZhenbiao/ChuanhuChatGPT
+
+# 借鉴项目2：借鉴了mdtex2html中公式处理的方法
+https://github.com/polarwinkel/mdtex2html
+
+项目使用OpenAI的gpt-3.5-turbo模型，期待gpt-4早点放宽门槛😂
+```
+
+> **Note**
+>
+> 1.请注意只有“红颜色”标识的函数插件（按钮）才支持读取文件。目前暂不能完善地支持pdf格式文献的翻译解读，尚不支持word格式文件的读取。
+>
+> 2.本项目中每个文件的功能都在`project_self_analysis.md`详细说明。随着版本的迭代，您也可以随时自行点击相关函数插件，调用GPT重新生成项目的自译解报告。
+> 
+> 3.如果您不太习惯部分中文命名的函数，您可以随时点击相关函数插件，调用GPT一键生成纯英文的项目源代码。
+
+<div align="center">
+    
+功能 | 描述
+--- | ---
+一键润色 | 支持一键润色、一键查找论文语法错误
+一键中英互译 | 一键中英互译
+一键代码解释 | 可以正确显示代码、解释代码
+自定义快捷键 | 支持自定义快捷键
+配置代理服务器 | 支持配置代理服务器
+模块化设计 | 支持自定义高阶的实验性功能
+自我程序剖析 | [实验性功能] 一键读懂本项目的源代码
+程序剖析 | [实验性功能] 一键可以剖析其他Python/C++项目
+读论文 | [实验性功能] 一键解读latex论文全文并生成摘要
+批量注释生成 | [实验性功能] 一键批量生成函数注释
+chat分析报告生成 | [实验性功能] 运行后自动生成总结汇报
+公式显示 | 可以同时显示公式的tex形式和渲染形式
+图片显示 | 可以在markdown中显示图片
+支持GPT输出的markdown表格 | 可以输出支持GPT的markdown表格
+…… | ……
+
+</div>
+
+- 新界面
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/228600410-7d44e34f-63f1-4046-acb8-045cb05da8bb.png" width="700" >
+</div>
+
+
+
+- 所有按钮都通过读取functional.py动态生成，可随意加自定义功能，解放粘贴板
+<div align="center">
+<img src="img/公式.gif" width="700" >
+</div>
+
+- 润色/纠错
+<div align="center">
+<img src="img/润色.gif" width="700" >
+</div>
+
+
+- 支持GPT输出的markdown表格
+<div align="center">
+<img src="img/demo2.jpg" width="500" >
+</div>
+
+- 如果输出包含公式，会同时以tex形式和渲染形式显示，方便复制和阅读
+<div align="center">
+<img src="img/demo.jpg" width="500" >
+</div>
+
+
+- 懒得看项目代码？整个工程直接给chatgpt炫嘴里
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226935232-6b6a73ce-8900-4aee-93f9-733c7e6fef53.png" width="700" >
+</div>
+
+## 直接运行 (Windows, Linux or MacOS)
+
+下载项目
+
+```sh
+git clone https://github.com/binary-husky/chatgpt_academic.git
+cd chatgpt_academic
+```
+
+我们建议将`config.py`复制为`config_private.py`并将后者用作个性化配置文件以避免`config.py`中的变更影响你的使用或不小心将包含你的OpenAI API KEY的`config.py`提交至本项目。
+
+```sh
+cp config.py config_private.py
+```
+
+在`config_private.py`中，配置 海外Proxy 和 OpenAI API KEY
+```
+1. 如果你在国内，需要设置海外代理才能够使用 OpenAI API，你可以通过 config.py 文件来进行设置。
+2. 配置 OpenAI API KEY。你需要在 OpenAI 官网上注册并获取 API KEY。一旦你拿到了 API KEY，在 config.py 文件里配置好即可。
+```
+安装依赖
+
+```sh
+python -m pip install -r requirements.txt
+```
+
+或者，如果你希望使用`conda`
+
+```sh
+conda create -n gptac 'gradio>=3.23' requests
+conda activate gptac
+python3 -m pip install mdtex2html
+```
+
+运行
+
+```sh
+python main.py
+```
+
+测试实验性功能
+```
+- 测试C++项目头文件分析
+    input区域 输入 `./crazy_functions/test_project/cpp/libJPG` ， 然后点击 "[实验] 解析整个C++项目（input输入项目根路径）"
+- 测试给Latex项目写摘要
+    input区域 输入 `./crazy_functions/test_project/latex/attention` ， 然后点击 "[实验] 读tex论文写摘要（input输入项目根路径）"
+- 测试Python项目分析
+    input区域 输入 `./crazy_functions/test_project/python/dqn` ， 然后点击 "[实验] 解析整个py项目（input输入项目根路径）"
+- 测试自我代码解读
+    点击 "[实验] 请解析并解构此项目本身"
+- 测试实验功能模板函数（要求gpt回答历史上的今天发生了什么），您可以根据此函数为模板，实现更复杂的功能
+    点击 "[实验] 实验功能函数模板"
+```
+
+与代理网络有关的issue（网络超时、代理不起作用）汇总到 https://github.com/binary-husky/chatgpt_academic/issues/1
+
+## 使用docker (Linux)
 
 ``` sh
 # 下载项目
@@ -15,12 +151,121 @@ docker build -t gpt-academic .
 # 运行
 docker run --rm -it --net=host gpt-academic
 
+# 测试实验性功能
+## 测试自我代码解读
+点击 "[实验] 请解析并解构此项目本身"
+## 测试实验功能模板函数（要求gpt回答历史上的今天发生了什么），您可以根据此函数为模板，实现更复杂的功能
+点击 "[实验] 实验功能函数模板"
+##（请注意在docker中运行时，需要额外注意程序的文件访问权限问题）
+## 测试C++项目头文件分析
+input区域 输入 ./crazy_functions/test_project/cpp/libJPG ， 然后点击 "[实验] 解析整个C++项目（input输入项目根路径）"
+## 测试给Latex项目写摘要
+input区域 输入 ./crazy_functions/test_project/latex/attention ， 然后点击 "[实验] 读tex论文写摘要（input输入项目根路径）"
+## 测试Python项目分析
+input区域 输入 ./crazy_functions/test_project/python/dqn ， 然后点击 "[实验] 解析整个py项目（input输入项目根路径）"
+
 ```
 
-## 参考项目
+## 使用WSL2（Windows Subsystem for Linux 子系统）
+选择这种方式默认您已经具备一定基本知识，因此不再赘述多余步骤。如果不是这样，您可以从[这里](https://learn.microsoft.com/zh-cn/windows/wsl/about)或GPT处获取更多关于子系统的信息。
+
+WSL2可以配置使用Windows侧的代理上网，前置步骤可以参考[这里](https://www.cnblogs.com/tuilk/p/16287472.html)
+由于Windows相对WSL2的IP会发生变化，我们需要每次启动前先获取这个IP来保证顺利访问，将config.py中设置proxies的部分更改为如下代码：
+```python
+import subprocess
+cmd_get_ip = 'grep -oP  "(\d+\.)+(\d+)" /etc/resolv.conf'
+ip_proxy = subprocess.run(
+        cmd_get_ip, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, shell=True
+        ).stdout.strip() # 获取windows的IP
+proxies = { "http": ip_proxy + ":51837", "https": ip_proxy + ":51837", } # 请自行修改
 ```
-https://github.com/Python-Markdown/markdown
-https://github.com/gradio-app/gradio
-https://github.com/polarwinkel/mdtex2html
-https://github.com/GaiZhenbiao/ChuanhuChatGPT
+在启动main.py后，可以在windows浏览器中访问服务。至此测试、使用与上面其他方法无异。 
+
+
+## 自定义新的便捷按钮（学术快捷键自定义）
+打开functional.py，添加条目如下，然后重启程序即可。（如果按钮已经添加成功并可见，那么前缀、后缀都支持热修改，无需重启程序即可生效。）
+例如
 ```
+"超级英译中": {
+
+    # 前缀，会被加在你的输入之前。例如，用来描述你的要求，例如翻译、解释代码、润色等等
+    "Prefix": "请翻译把下面一段内容成中文，然后用一个markdown表格逐一解释文中出现的专有名词：\n\n", 
+    
+    # 后缀，会被加在你的输入之后。例如，配合前缀可以把你的输入内容用引号圈起来。
+    "Suffix": "",
+    
+},
+```
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226899272-477c2134-ed71-4326-810c-29891fe4a508.png" width="500" >
+</div>
+
+
+如果你发明了更好用的学术快捷键，欢迎发issue或者pull requests！
+
+## 配置代理
+
+在```config.py```中修改端口与代理软件对应
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226571294-37a47cd9-4d40-4c16-97a2-d360845406f7.png" width="500" >
+<img src="https://user-images.githubusercontent.com/96192199/226838985-e5c95956-69c2-4c23-a4dd-cd7944eeb451.png" width="500" >
+</div>
+
+配置完成后，你可以用以下命令测试代理是否工作，如果一切正常，下面的代码将输出你的代理服务器所在地：
+```
+python check_proxy.py
+```
+
+## 兼容性测试
+
+### 图片显示：
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/228737599-bf0a9d9c-1808-4f43-ae15-dfcc7af0f295.png" width="800" >
+</div>
+
+
+### 如果一个程序能够读懂并剖析自己：
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226936850-c77d7183-0749-4c1c-9875-fd4891842d0c.png" width="800" >
+</div>
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226936618-9b487e4b-ab5b-4b6e-84c6-16942102e917.png" width="800" >
+</div>
+
+### 其他任意Python/Cpp项目剖析：
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226935232-6b6a73ce-8900-4aee-93f9-733c7e6fef53.png" width="800" >
+</div>
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/226969067-968a27c1-1b9c-486b-8b81-ab2de8d3f88a.png" width="800" >
+</div>
+
+### Latex论文一键阅读理解与摘要生成
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/227504406-86ab97cd-f208-41c3-8e4a-7000e51cf980.png" width="800" >
+</div>
+
+### 自动报告生成
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/227503770-fe29ce2c-53fd-47b0-b0ff-93805f0c2ff4.png" height="300" >
+<img src="https://user-images.githubusercontent.com/96192199/227504617-7a497bb3-0a2a-4b50-9a8a-95ae60ea7afd.png" height="300" >
+<img src="https://user-images.githubusercontent.com/96192199/227504005-efeaefe0-b687-49d0-bf95-2d7b7e66c348.png" height="300" >
+</div>
+
+### 模块化功能设计
+<div align="center">
+<img src="https://user-images.githubusercontent.com/96192199/227504981-4c6c39c0-ae79-47e6-bffe-0e6442d9da65.png" height="400" >
+<img src="https://user-images.githubusercontent.com/96192199/227504931-19955f78-45cd-4d1c-adac-e71e50957915.png" height="400" >
+</div>
+
+## Todo:
+
+- (Top Priority) 调用另一个开源项目text-generation-webui的web接口，使用其他llm模型
+- 总结大工程源代码时，文本过长、token溢出的问题（目前的方法是直接二分丢弃处理溢出，过于粗暴，有效信息大量丢失）
+- UI不够美观
+

check_proxy.py

@@ -0,0 +1,27 @@
+
+def check_proxy(proxies):
+    import requests
+    proxies_https = proxies['https'] if proxies is not None else '无'
+    try:
+        response = requests.get("https://ipapi.co/json/", proxies=proxies, timeout=4)
+        data = response.json()
+        print(f'查询代理的地理位置，返回的结果是{data}')
+        if 'country_name' in data:
+            country = data['country_name']
+            result = f"代理配置 {proxies_https}, 代理所在地：{country}"
+        elif 'error' in data:
+            result = f"代理配置 {proxies_https}, 代理所在地：未知，IP查询频率受限"
+        print(result)
+        return result
+    except:
+        result = f"代理配置 {proxies_https}, 代理所在地查询超时，代理可能无效"
+        print(result)
+        return result
+
+
+if __name__ == '__main__':
+    import os; os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染
+    from toolbox import get_conf
+    proxies, = get_conf('proxies')
+    check_proxy(proxies)
+    

config.py

@@ -1,11 +1,38 @@
-# my_api_key = "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"
+# API_KEY = "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" 此key无效
 API_KEY = "sk-此处填API秘钥"
 API_URL = "https://api.openai.com/v1/chat/completions"
 
+# 改为True应用代理
 USE_PROXY = False
 if USE_PROXY:
-    proxies = { "http": "socks5h://localhost:11284", "https": "socks5h://localhost:11284", } 
+
+    # 填写格式是 [协议]://  [地址] :[端口] ，
+    # 例如    "socks5h://localhost:11284"
+    # [协议] 常见协议无非socks5h/http，例如 v2*** 和 s** 的默认本地协议是socks5h，cl**h 的默认本地协议是http
+    # [地址] 懂的都懂，不懂就填localhost或者127.0.0.1肯定错不了（localhost意思是代理软件安装在本机上）
+    # [端口] 在代理软件的设置里，不同的代理软件界面不一样，但端口号都应该在最显眼的位置上
+
+    # 代理网络的地址，打开你的科学上网软件查看代理的协议(socks5/http)、地址(localhost)和端口(11284)
+    proxies = { "http": "socks5h://localhost:11284", "https": "socks5h://localhost:11284", }
     print('网络代理状态：运行。')
 else:
     proxies = None
     print('网络代理状态：未配置。无代理状态下很可能无法访问。')
+
+# 发送请求到OpenAI后，等待多久判定为超时
+TIMEOUT_SECONDS = 25
+
+# 网页的端口, -1代表随机端口
+WEB_PORT = -1
+
+# 如果OpenAI不响应（网络卡顿、代理失败、KEY失效），重试的次数限制
+MAX_RETRY = 2
+
+# 选择的OpenAI模型是（gpt4现在只对申请成功的人开放）
+LLM_MODEL = "gpt-3.5-turbo"
+
+# 设置并行使用的线程数
+CONCURRENT_COUNT = 100
+
+# 设置用户名和密码
+AUTHENTICATION = [] # [("username", "password"), ("username2", "password2"), ...]

crazy_functions/test_project/cpp/cppipc/buffer.cpp

@@ -0,0 +1,87 @@
+#include "libipc/buffer.h"
+#include "libipc/utility/pimpl.h"
+
+#include <cstring>
+
+namespace ipc {
+
+bool operator==(buffer const & b1, buffer const & b2) {
+    return (b1.size() == b2.size()) && (std::memcmp(b1.data(), b2.data(), b1.size()) == 0);
+}
+
+bool operator!=(buffer const & b1, buffer const & b2) {
+    return !(b1 == b2);
+}
+
+class buffer::buffer_ : public pimpl<buffer_> {
+public:
+    void*       p_;
+    std::size_t s_;
+    void*       a_;
+    buffer::destructor_t d_;
+
+    buffer_(void* p, std::size_t s, buffer::destructor_t d, void* a)
+        : p_(p), s_(s), a_(a), d_(d) {
+    }
+
+    ~buffer_() {
+        if (d_ == nullptr) return;
+        d_((a_ == nullptr) ? p_ : a_, s_);
+    }
+};
+
+buffer::buffer()
+    : buffer(nullptr, 0, nullptr, nullptr) {
+}
+
+buffer::buffer(void* p, std::size_t s, destructor_t d)
+    : p_(p_->make(p, s, d, nullptr)) {
+}
+
+buffer::buffer(void* p, std::size_t s, destructor_t d, void* additional)
+    : p_(p_->make(p, s, d, additional)) {
+}
+
+buffer::buffer(void* p, std::size_t s)
+    : buffer(p, s, nullptr) {
+}
+
+buffer::buffer(char const & c)
+    : buffer(const_cast<char*>(&c), 1) {
+}
+
+buffer::buffer(buffer&& rhs)
+    : buffer() {
+    swap(rhs);
+}
+
+buffer::~buffer() {
+    p_->clear();
+}
+
+void buffer::swap(buffer& rhs) {
+    std::swap(p_, rhs.p_);
+}
+
+buffer& buffer::operator=(buffer rhs) {
+    swap(rhs);
+    return *this;
+}
+
+bool buffer::empty() const noexcept {
+    return (impl(p_)->p_ == nullptr) || (impl(p_)->s_ == 0);
+}
+
+void* buffer::data() noexcept {
+    return impl(p_)->p_;
+}
+
+void const * buffer::data() const noexcept {
+    return impl(p_)->p_;
+}
+
+std::size_t buffer::size() const noexcept {
+    return impl(p_)->s_;
+}
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/ipc.cpp

@@ -0,0 +1,701 @@
+
+#include <type_traits>
+#include <cstring>
+#include <algorithm>
+#include <utility>          // std::pair, std::move, std::forward
+#include <atomic>
+#include <type_traits>      // aligned_storage_t
+#include <string>
+#include <vector>
+#include <array>
+#include <cassert>
+
+#include "libipc/ipc.h"
+#include "libipc/def.h"
+#include "libipc/shm.h"
+#include "libipc/pool_alloc.h"
+#include "libipc/queue.h"
+#include "libipc/policy.h"
+#include "libipc/rw_lock.h"
+#include "libipc/waiter.h"
+
+#include "libipc/utility/log.h"
+#include "libipc/utility/id_pool.h"
+#include "libipc/utility/scope_guard.h"
+#include "libipc/utility/utility.h"
+
+#include "libipc/memory/resource.h"
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_array.h"
+
+namespace {
+
+using msg_id_t = std::uint32_t;
+using acc_t    = std::atomic<msg_id_t>;
+
+template <std::size_t DataSize, std::size_t AlignSize>
+struct msg_t;
+
+template <std::size_t AlignSize>
+struct msg_t<0, AlignSize> {
+    msg_id_t     cc_id_;
+    msg_id_t     id_;
+    std::int32_t remain_;
+    bool         storage_;
+};
+
+template <std::size_t DataSize, std::size_t AlignSize>
+struct msg_t : msg_t<0, AlignSize> {
+    std::aligned_storage_t<DataSize, AlignSize> data_ {};
+
+    msg_t() = default;
+    msg_t(msg_id_t cc_id, msg_id_t id, std::int32_t remain, void const * data, std::size_t size)
+        : msg_t<0, AlignSize> {cc_id, id, remain, (data == nullptr) || (size == 0)} {
+        if (this->storage_) {
+            if (data != nullptr) {
+                // copy storage-id
+                *reinterpret_cast<ipc::storage_id_t*>(&data_) =
+                     *static_cast<ipc::storage_id_t const *>(data);
+            }
+        }
+        else std::memcpy(&data_, data, size);
+    }
+};
+
+template <typename T>
+ipc::buff_t make_cache(T& data, std::size_t size) {
+    auto ptr = ipc::mem::alloc(size);
+    std::memcpy(ptr, &data, (ipc::detail::min)(sizeof(data), size));
+    return { ptr, size, ipc::mem::free };
+}
+
+struct cache_t {
+    std::size_t fill_;
+    ipc::buff_t buff_;
+
+    cache_t(std::size_t f, ipc::buff_t && b)
+        : fill_(f), buff_(std::move(b))
+    {}
+
+    void append(void const * data, std::size_t size) {
+        if (fill_ >= buff_.size() || data == nullptr || size == 0) return;
+        auto new_fill = (ipc::detail::min)(fill_ + size, buff_.size());
+        std::memcpy(static_cast<ipc::byte_t*>(buff_.data()) + fill_, data, new_fill - fill_);
+        fill_ = new_fill;
+    }
+};
+
+auto cc_acc() {
+    static ipc::shm::handle acc_h("__CA_CONN__", sizeof(acc_t));
+    return static_cast<acc_t*>(acc_h.get());
+}
+
+IPC_CONSTEXPR_ std::size_t align_chunk_size(std::size_t size) noexcept {
+    return (((size - 1) / ipc::large_msg_align) + 1) * ipc::large_msg_align;
+}
+
+IPC_CONSTEXPR_ std::size_t calc_chunk_size(std::size_t size) noexcept {
+    return ipc::make_align(alignof(std::max_align_t), align_chunk_size(
+           ipc::make_align(alignof(std::max_align_t), sizeof(std::atomic<ipc::circ::cc_t>)) + size));
+}
+
+struct chunk_t {
+    std::atomic<ipc::circ::cc_t> &conns() noexcept {
+        return *reinterpret_cast<std::atomic<ipc::circ::cc_t> *>(this);
+    }
+
+    void *data() noexcept {
+        return reinterpret_cast<ipc::byte_t *>(this)
+             + ipc::make_align(alignof(std::max_align_t), sizeof(std::atomic<ipc::circ::cc_t>));
+    }
+};
+
+struct chunk_info_t {
+    ipc::id_pool<> pool_;
+    ipc::spin_lock lock_;
+
+    IPC_CONSTEXPR_ static std::size_t chunks_mem_size(std::size_t chunk_size) noexcept {
+        return ipc::id_pool<>::max_count * chunk_size;
+    }
+
+    ipc::byte_t *chunks_mem() noexcept {
+        return reinterpret_cast<ipc::byte_t *>(this + 1);
+    }
+
+    chunk_t *at(std::size_t chunk_size, ipc::storage_id_t id) noexcept {
+        if (id < 0) return nullptr;
+        return reinterpret_cast<chunk_t *>(chunks_mem() + (chunk_size * id));
+    }
+};
+
+auto& chunk_storages() {
+    class chunk_handle_t {
+        ipc::shm::handle handle_;
+
+    public:
+        chunk_info_t *get_info(std::size_t chunk_size) {
+            if (!handle_.valid() &&
+                !handle_.acquire( ("__CHUNK_INFO__" + ipc::to_string(chunk_size)).c_str(), 
+                                  sizeof(chunk_info_t) + chunk_info_t::chunks_mem_size(chunk_size) )) {
+                ipc::error("[chunk_storages] chunk_shm.id_info_.acquire failed: chunk_size = %zd\n", chunk_size);
+                return nullptr;
+            }
+            auto info = static_cast<chunk_info_t*>(handle_.get());
+            if (info == nullptr) {
+                ipc::error("[chunk_storages] chunk_shm.id_info_.get failed: chunk_size = %zd\n", chunk_size);
+                return nullptr;
+            }
+            return info;
+        }
+    };
+    static ipc::map<std::size_t, chunk_handle_t> chunk_hs;
+    return chunk_hs;
+}
+
+chunk_info_t *chunk_storage_info(std::size_t chunk_size) {
+    auto &storages = chunk_storages();
+    std::decay_t<decltype(storages)>::iterator it;
+    {
+        static ipc::rw_lock lock;
+        IPC_UNUSED_ std::shared_lock<ipc::rw_lock> guard {lock};
+        if ((it = storages.find(chunk_size)) == storages.end()) {
+            using chunk_handle_t = std::decay_t<decltype(storages)>::value_type::second_type;
+            guard.unlock();
+            IPC_UNUSED_ std::lock_guard<ipc::rw_lock> guard {lock};
+            it = storages.emplace(chunk_size, chunk_handle_t{}).first;
+        }
+    }
+    return it->second.get_info(chunk_size);
+}
+
+std::pair<ipc::storage_id_t, void*> acquire_storage(std::size_t size, ipc::circ::cc_t conns) {
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return {};
+
+    info->lock_.lock();
+    info->pool_.prepare();
+    // got an unique id
+    auto id = info->pool_.acquire();
+    info->lock_.unlock();
+
+    auto chunk = info->at(chunk_size, id);
+    if (chunk == nullptr) return {};
+    chunk->conns().store(conns, std::memory_order_relaxed);
+    return { id, chunk->data() };
+}
+
+void *find_storage(ipc::storage_id_t id, std::size_t size) {
+    if (id < 0) {
+        ipc::error("[find_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
+        return nullptr;
+    }
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return nullptr;
+    return info->at(chunk_size, id)->data();
+}
+
+void release_storage(ipc::storage_id_t id, std::size_t size) {
+    if (id < 0) {
+        ipc::error("[release_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
+        return;
+    }
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return;
+    info->lock_.lock();
+    info->pool_.release(id);
+    info->lock_.unlock();
+}
+
+template <ipc::relat Rp, ipc::relat Rc>
+bool sub_rc(ipc::wr<Rp, Rc, ipc::trans::unicast>, 
+            std::atomic<ipc::circ::cc_t> &/*conns*/, ipc::circ::cc_t /*curr_conns*/, ipc::circ::cc_t /*conn_id*/) noexcept {
+    return true;
+}
+
+template <ipc::relat Rp, ipc::relat Rc>
+bool sub_rc(ipc::wr<Rp, Rc, ipc::trans::broadcast>, 
+            std::atomic<ipc::circ::cc_t> &conns, ipc::circ::cc_t curr_conns, ipc::circ::cc_t conn_id) noexcept {
+    auto last_conns = curr_conns & ~conn_id;
+    for (unsigned k = 0;;) {
+        auto chunk_conns  = conns.load(std::memory_order_acquire);
+        if (conns.compare_exchange_weak(chunk_conns, chunk_conns & last_conns, std::memory_order_release)) {
+            return (chunk_conns & last_conns) == 0;
+        }
+        ipc::yield(k);
+    }
+}
+
+template <typename Flag>
+void recycle_storage(ipc::storage_id_t id, std::size_t size, ipc::circ::cc_t curr_conns, ipc::circ::cc_t conn_id) {
+    if (id < 0) {
+        ipc::error("[recycle_storage] id is invalid: id = %ld, size = %zd\n", (long)id, size);
+        return;
+    }
+    std::size_t chunk_size = calc_chunk_size(size);
+    auto info = chunk_storage_info(chunk_size);
+    if (info == nullptr) return;
+
+    auto chunk = info->at(chunk_size, id);
+    if (chunk == nullptr) return;
+
+    if (!sub_rc(Flag{}, chunk->conns(), curr_conns, conn_id)) {
+        return;
+    }
+    info->lock_.lock();
+    info->pool_.release(id);
+    info->lock_.unlock();
+}
+
+template <typename MsgT>
+bool clear_message(void* p) {
+    auto msg = static_cast<MsgT*>(p);
+    if (msg->storage_) {
+        std::int32_t r_size = static_cast<std::int32_t>(ipc::data_length) + msg->remain_;
+        if (r_size <= 0) {
+            ipc::error("[clear_message] invalid msg size: %d\n", (int)r_size);
+            return true;
+        }
+        release_storage(
+            *reinterpret_cast<ipc::storage_id_t*>(&msg->data_),
+            static_cast<std::size_t>(r_size));
+    }
+    return true;
+}
+
+struct conn_info_head {
+
+    ipc::string name_;
+    msg_id_t    cc_id_; // connection-info id
+    ipc::detail::waiter cc_waiter_, wt_waiter_, rd_waiter_;
+    ipc::shm::handle acc_h_;
+
+    conn_info_head(char const * name)
+        : name_     {name}
+        , cc_id_    {(cc_acc() == nullptr) ? 0 : cc_acc()->fetch_add(1, std::memory_order_relaxed)}
+        , cc_waiter_{("__CC_CONN__" + name_).c_str()}
+        , wt_waiter_{("__WT_CONN__" + name_).c_str()}
+        , rd_waiter_{("__RD_CONN__" + name_).c_str()}
+        , acc_h_    {("__AC_CONN__" + name_).c_str(), sizeof(acc_t)} {
+    }
+
+    void quit_waiting() {
+        cc_waiter_.quit_waiting();
+        wt_waiter_.quit_waiting();
+        rd_waiter_.quit_waiting();
+    }
+
+    auto acc() {
+        return static_cast<acc_t*>(acc_h_.get());
+    }
+
+    auto& recv_cache() {
+        thread_local ipc::unordered_map<msg_id_t, cache_t> tls;
+        return tls;
+    }
+};
+
+template <typename W, typename F>
+bool wait_for(W& waiter, F&& pred, std::uint64_t tm) {
+    if (tm == 0) return !pred();
+    for (unsigned k = 0; pred();) {
+        bool ret = true;
+        ipc::sleep(k, [&k, &ret, &waiter, &pred, tm] {
+            ret = waiter.wait_if(std::forward<F>(pred), tm);
+            k   = 0;
+        });
+        if (!ret) return false; // timeout or fail
+        if (k == 0) break; // k has been reset
+    }
+    return true;
+}
+
+template <typename Policy,
+          std::size_t DataSize  = ipc::data_length,
+          std::size_t AlignSize = (ipc::detail::min)(DataSize, alignof(std::max_align_t))>
+struct queue_generator {
+
+    using queue_t = ipc::queue<msg_t<DataSize, AlignSize>, Policy>;
+
+    struct conn_info_t : conn_info_head {
+        queue_t que_;
+
+        conn_info_t(char const * name)
+            : conn_info_head{name}
+            , que_{("__QU_CONN__" +
+                    ipc::to_string(DataSize) + "__" +
+                    ipc::to_string(AlignSize) + "__" + name).c_str()} {
+        }
+
+        void disconnect_receiver() {
+            bool dis = que_.disconnect();
+            this->quit_waiting();
+            if (dis) {
+                this->recv_cache().clear();
+            }
+        }
+    };
+};
+
+template <typename Policy>
+struct detail_impl {
+
+using policy_t    = Policy;
+using flag_t      = typename policy_t::flag_t;
+using queue_t     = typename queue_generator<policy_t>::queue_t;
+using conn_info_t = typename queue_generator<policy_t>::conn_info_t;
+
+constexpr static conn_info_t* info_of(ipc::handle_t h) noexcept {
+    return static_cast<conn_info_t*>(h);
+}
+
+constexpr static queue_t* queue_of(ipc::handle_t h) noexcept {
+    return (info_of(h) == nullptr) ? nullptr : &(info_of(h)->que_);
+}
+
+/* API implementations */
+
+static void disconnect(ipc::handle_t h) {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        return;
+    }
+    que->shut_sending();
+    assert(info_of(h) != nullptr);
+    info_of(h)->disconnect_receiver();
+}
+
+static bool reconnect(ipc::handle_t * ph, bool start_to_recv) {
+    assert(ph != nullptr);
+    assert(*ph != nullptr);
+    auto que = queue_of(*ph);
+    if (que == nullptr) {
+        return false;
+    }
+    if (start_to_recv) {
+        que->shut_sending();
+        if (que->connect()) { // wouldn't connect twice
+            info_of(*ph)->cc_waiter_.broadcast();
+            return true;
+        }
+        return false;
+    }
+    // start_to_recv == false
+    if (que->connected()) {
+        info_of(*ph)->disconnect_receiver();
+    }
+    return que->ready_sending();
+}
+
+static bool connect(ipc::handle_t * ph, char const * name, bool start_to_recv) {
+    assert(ph != nullptr);
+    if (*ph == nullptr) {
+        *ph = ipc::mem::alloc<conn_info_t>(name);
+    }
+    return reconnect(ph, start_to_recv);
+}
+
+static void destroy(ipc::handle_t h) {
+    disconnect(h);
+    ipc::mem::free(info_of(h));
+}
+
+static std::size_t recv_count(ipc::handle_t h) noexcept {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        return ipc::invalid_value;
+    }
+    return que->conn_count();
+}
+
+static bool wait_for_recv(ipc::handle_t h, std::size_t r_count, std::uint64_t tm) {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        return false;
+    }
+    return wait_for(info_of(h)->cc_waiter_, [que, r_count] {
+        return que->conn_count() < r_count;
+    }, tm);
+}
+
+template <typename F>
+static bool send(F&& gen_push, ipc::handle_t h, void const * data, std::size_t size) {
+    if (data == nullptr || size == 0) {
+        ipc::error("fail: send(%p, %zd)\n", data, size);
+        return false;
+    }
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        ipc::error("fail: send, queue_of(h) == nullptr\n");
+        return false;
+    }
+    if (que->elems() == nullptr) {
+        ipc::error("fail: send, queue_of(h)->elems() == nullptr\n");
+        return false;
+    }
+    if (!que->ready_sending()) {
+        ipc::error("fail: send, que->ready_sending() == false\n");
+        return false;
+    }
+    ipc::circ::cc_t conns = que->elems()->connections(std::memory_order_relaxed);
+    if (conns == 0) {
+        ipc::error("fail: send, there is no receiver on this connection.\n");
+        return false;
+    }
+    // calc a new message id
+    auto acc = info_of(h)->acc();
+    if (acc == nullptr) {
+        ipc::error("fail: send, info_of(h)->acc() == nullptr\n");
+        return false;
+    }
+    auto msg_id   = acc->fetch_add(1, std::memory_order_relaxed);
+    auto try_push = std::forward<F>(gen_push)(info_of(h), que, msg_id);
+    if (size > ipc::large_msg_limit) {
+        auto   dat = acquire_storage(size, conns);
+        void * buf = dat.second;
+        if (buf != nullptr) {
+            std::memcpy(buf, data, size);
+            return try_push(static_cast<std::int32_t>(size) - 
+                            static_cast<std::int32_t>(ipc::data_length), &(dat.first), 0);
+        }
+        // try using message fragment
+        //ipc::log("fail: shm::handle for big message. msg_id: %zd, size: %zd\n", msg_id, size);
+    }
+    // push message fragment
+    std::int32_t offset = 0;
+    for (std::int32_t i = 0; i < static_cast<std::int32_t>(size / ipc::data_length); ++i, offset += ipc::data_length) {
+        if (!try_push(static_cast<std::int32_t>(size) - offset - static_cast<std::int32_t>(ipc::data_length),
+                      static_cast<ipc::byte_t const *>(data) + offset, ipc::data_length)) {
+            return false;
+        }
+    }
+    // if remain > 0, this is the last message fragment
+    std::int32_t remain = static_cast<std::int32_t>(size) - offset;
+    if (remain > 0) {
+        if (!try_push(remain - static_cast<std::int32_t>(ipc::data_length),
+                      static_cast<ipc::byte_t const *>(data) + offset, 
+                      static_cast<std::size_t>(remain))) {
+            return false;
+        }
+    }
+    return true;
+}
+
+static bool send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return send([tm](auto info, auto que, auto msg_id) {
+        return [tm, info, que, msg_id](std::int32_t remain, void const * data, std::size_t size) {
+            if (!wait_for(info->wt_waiter_, [&] {
+                    return !que->push(
+                        [](void*) { return true; },
+                        info->cc_id_, msg_id, remain, data, size);
+                }, tm)) {
+                ipc::log("force_push: msg_id = %zd, remain = %d, size = %zd\n", msg_id, remain, size);
+                if (!que->force_push(
+                        clear_message<typename queue_t::value_t>,
+                        info->cc_id_, msg_id, remain, data, size)) {
+                    return false;
+                }
+            }
+            info->rd_waiter_.broadcast();
+            return true;
+        };
+    }, h, data, size);
+}
+
+static bool try_send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return send([tm](auto info, auto que, auto msg_id) {
+        return [tm, info, que, msg_id](std::int32_t remain, void const * data, std::size_t size) {
+            if (!wait_for(info->wt_waiter_, [&] {
+                    return !que->push(
+                        [](void*) { return true; },
+                        info->cc_id_, msg_id, remain, data, size);
+                }, tm)) {
+                return false;
+            }
+            info->rd_waiter_.broadcast();
+            return true;
+        };
+    }, h, data, size);
+}
+
+static ipc::buff_t recv(ipc::handle_t h, std::uint64_t tm) {
+    auto que = queue_of(h);
+    if (que == nullptr) {
+        ipc::error("fail: recv, queue_of(h) == nullptr\n");
+        return {};
+    }
+    if (!que->connected()) {
+        // hasn't connected yet, just return.
+        return {};
+    }
+    auto& rc = info_of(h)->recv_cache();
+    for (;;) {
+        // pop a new message
+        typename queue_t::value_t msg;
+        if (!wait_for(info_of(h)->rd_waiter_, [que, &msg] {
+                return !que->pop(msg);
+            }, tm)) {
+            // pop failed, just return.
+            return {};
+        }
+        info_of(h)->wt_waiter_.broadcast();
+        if ((info_of(h)->acc() != nullptr) && (msg.cc_id_ == info_of(h)->cc_id_)) {
+            continue; // ignore message to self
+        }
+        // msg.remain_ may minus & abs(msg.remain_) < data_length
+        std::int32_t r_size = static_cast<std::int32_t>(ipc::data_length) + msg.remain_;
+        if (r_size <= 0) {
+            ipc::error("fail: recv, r_size = %d\n", (int)r_size);
+            return {};
+        }
+        std::size_t msg_size = static_cast<std::size_t>(r_size);
+        // large message
+        if (msg.storage_) {
+            ipc::storage_id_t buf_id = *reinterpret_cast<ipc::storage_id_t*>(&msg.data_);
+            void* buf = find_storage(buf_id, msg_size);
+            if (buf != nullptr) {
+                struct recycle_t {
+                    ipc::storage_id_t storage_id;
+                    ipc::circ::cc_t   curr_conns;
+                    ipc::circ::cc_t   conn_id;
+                } *r_info = ipc::mem::alloc<recycle_t>(recycle_t{
+                    buf_id, que->elems()->connections(std::memory_order_relaxed), que->connected_id()
+                });
+                if (r_info == nullptr) {
+                    ipc::log("fail: ipc::mem::alloc<recycle_t>.\n");
+                    return ipc::buff_t{buf, msg_size}; // no recycle
+                } else {
+                    return ipc::buff_t{buf, msg_size, [](void* p_info, std::size_t size) {
+                        auto r_info = static_cast<recycle_t *>(p_info);
+                        IPC_UNUSED_ auto finally = ipc::guard([r_info] {
+                            ipc::mem::free(r_info);
+                        });
+                        recycle_storage<flag_t>(r_info->storage_id, size, r_info->curr_conns, r_info->conn_id);
+                    }, r_info};
+                }
+            } else {
+                ipc::log("fail: shm::handle for large message. msg_id: %zd, buf_id: %zd, size: %zd\n", msg.id_, buf_id, msg_size);
+                continue;
+            }
+        }
+        // find cache with msg.id_
+        auto cac_it = rc.find(msg.id_);
+        if (cac_it == rc.end()) {
+            if (msg_size <= ipc::data_length) {
+                return make_cache(msg.data_, msg_size);
+            }
+            // gc
+            if (rc.size() > 1024) {
+                std::vector<msg_id_t> need_del;
+                for (auto const & pair : rc) {
+                    auto cmp = std::minmax(msg.id_, pair.first);
+                    if (cmp.second - cmp.first > 8192) {
+                        need_del.push_back(pair.first);
+                    }
+                }
+                for (auto id : need_del) rc.erase(id);
+            }
+            // cache the first message fragment
+            rc.emplace(msg.id_, cache_t { ipc::data_length, make_cache(msg.data_, msg_size) });
+        }
+        // has cached before this message
+        else {
+            auto& cac = cac_it->second;
+            // this is the last message fragment
+            if (msg.remain_ <= 0) {
+                cac.append(&(msg.data_), msg_size);
+                // finish this message, erase it from cache
+                auto buff = std::move(cac.buff_);
+                rc.erase(cac_it);
+                return buff;
+            }
+            // there are remain datas after this message
+            cac.append(&(msg.data_), ipc::data_length);
+        }
+    }
+}
+
+static ipc::buff_t try_recv(ipc::handle_t h) {
+    return recv(h, 0);
+}
+
+}; // detail_impl<Policy>
+
+template <typename Flag>
+using policy_t = ipc::policy::choose<ipc::circ::elem_array, Flag>;
+
+} // internal-linkage
+
+namespace ipc {
+
+template <typename Flag>
+ipc::handle_t chan_impl<Flag>::inited() {
+    ipc::detail::waiter::init();
+    return nullptr;
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::connect(ipc::handle_t * ph, char const * name, unsigned mode) {
+    return detail_impl<policy_t<Flag>>::connect(ph, name, mode & receiver);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::reconnect(ipc::handle_t * ph, unsigned mode) {
+    return detail_impl<policy_t<Flag>>::reconnect(ph, mode & receiver);
+}
+
+template <typename Flag>
+void chan_impl<Flag>::disconnect(ipc::handle_t h) {
+    detail_impl<policy_t<Flag>>::disconnect(h);
+}
+
+template <typename Flag>
+void chan_impl<Flag>::destroy(ipc::handle_t h) {
+    detail_impl<policy_t<Flag>>::destroy(h);
+}
+
+template <typename Flag>
+char const * chan_impl<Flag>::name(ipc::handle_t h) {
+    auto info = detail_impl<policy_t<Flag>>::info_of(h);
+    return (info == nullptr) ? nullptr : info->name_.c_str();
+}
+
+template <typename Flag>
+std::size_t chan_impl<Flag>::recv_count(ipc::handle_t h) {
+    return detail_impl<policy_t<Flag>>::recv_count(h);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::wait_for_recv(ipc::handle_t h, std::size_t r_count, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::wait_for_recv(h, r_count, tm);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::send(h, data, size, tm);
+}
+
+template <typename Flag>
+buff_t chan_impl<Flag>::recv(ipc::handle_t h, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::recv(h, tm);
+}
+
+template <typename Flag>
+bool chan_impl<Flag>::try_send(ipc::handle_t h, void const * data, std::size_t size, std::uint64_t tm) {
+    return detail_impl<policy_t<Flag>>::try_send(h, data, size, tm);
+}
+
+template <typename Flag>
+buff_t chan_impl<Flag>::try_recv(ipc::handle_t h) {
+    return detail_impl<policy_t<Flag>>::try_recv(h);
+}
+
+template struct chan_impl<ipc::wr<relat::single, relat::single, trans::unicast  >>;
+// template struct chan_impl<ipc::wr<relat::single, relat::multi , trans::unicast  >>; // TBD
+// template struct chan_impl<ipc::wr<relat::multi , relat::multi , trans::unicast  >>; // TBD
+template struct chan_impl<ipc::wr<relat::single, relat::multi , trans::broadcast>>;
+template struct chan_impl<ipc::wr<relat::multi , relat::multi , trans::broadcast>>;
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/policy.h

@@ -0,0 +1,25 @@
+#pragma once
+
+#include <type_traits>
+
+#include "libipc/def.h"
+#include "libipc/prod_cons.h"
+
+#include "libipc/circ/elem_array.h"
+
+namespace ipc {
+namespace policy {
+
+template <template <typename, std::size_t...> class Elems, typename Flag>
+struct choose;
+
+template <typename Flag>
+struct choose<circ::elem_array, Flag> {
+    using flag_t = Flag;
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    using elems_t = circ::elem_array<ipc::prod_cons_impl<flag_t>, DataSize, AlignSize>;
+};
+
+} // namespace policy
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/pool_alloc.cpp

@@ -0,0 +1,17 @@
+#include "libipc/pool_alloc.h"
+
+#include "libipc/memory/resource.h"
+
+namespace ipc {
+namespace mem {
+
+void* pool_alloc::alloc(std::size_t size) {
+    return async_pool_alloc::alloc(size);
+}
+
+void pool_alloc::free(void* p, std::size_t size) {
+    async_pool_alloc::free(p, size);
+}
+
+} // namespace mem
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/prod_cons.h

@@ -0,0 +1,433 @@
+#pragma once
+
+#include <atomic>
+#include <utility>
+#include <cstring>
+#include <type_traits>
+#include <cstdint>
+
+#include "libipc/def.h"
+
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_def.h"
+#include "libipc/utility/log.h"
+#include "libipc/utility/utility.h"
+
+namespace ipc {
+
+////////////////////////////////////////////////////////////////
+/// producer-consumer implementation
+////////////////////////////////////////////////////////////////
+
+template <typename Flag>
+struct prod_cons_impl;
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> rd_; // read index
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_; // write index
+
+    constexpr circ::u2_t cursor() const noexcept {
+        return 0;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        auto cur_wt = circ::index_of(wt_.load(std::memory_order_relaxed));
+        if (cur_wt == circ::index_of(rd_.load(std::memory_order_acquire) - 1)) {
+            return false; // full
+        }
+        std::forward<F>(f)(&(elems[cur_wt].data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    /**
+     * In single-single-unicast, 'force_push' means 'no reader' or 'the only one reader is dead'.
+     * So we could just disconnect all connections of receiver, and return false.
+    */
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(~static_cast<circ::cc_t>(0u));
+        return false;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E* elems) {
+        auto cur_rd = circ::index_of(rd_.load(std::memory_order_relaxed));
+        if (cur_rd == circ::index_of(wt_.load(std::memory_order_acquire))) {
+            return false; // empty
+        }
+        std::forward<F>(f)(&(elems[cur_rd].data_));
+        std::forward<R>(out)(true);
+        rd_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi , trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::single, trans::unicast>> {
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            if (circ::index_of(cur_rd) ==
+                circ::index_of(wt_.load(std::memory_order_acquire))) {
+                return false; // empty
+            }
+            std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+            if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                std::forward<F>(f)(buff);
+                std::forward<R>(out)(true);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi , relat::multi, trans::unicast>>
+     : prod_cons_impl<wr<relat::single, relat::multi, trans::unicast>> {
+
+    using flag_t = std::uint64_t;
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_; // commit index
+
+    template <typename W, typename F, typename E>
+    bool push(W* /*wrapper*/, F&& f, E* elems) {
+        circ::u2_t cur_ct, nxt_ct;
+        for (unsigned k = 0;;) {
+            cur_ct = ct_.load(std::memory_order_relaxed);
+            if (circ::index_of(nxt_ct = cur_ct + 1) ==
+                circ::index_of(rd_.load(std::memory_order_acquire))) {
+                return false; // full
+            }
+            if (ct_.compare_exchange_weak(cur_ct, nxt_ct, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        auto* el = elems + circ::index_of(cur_ct);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        while (1) {
+            auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+            if (cur_ct != wt_.load(std::memory_order_relaxed)) {
+                return true;
+            }
+            if ((~cac_ct) != cur_ct) {
+                return true;
+            }
+            if (!el->f_ct_.compare_exchange_strong(cac_ct, 0, std::memory_order_relaxed)) {
+                return true;
+            }
+            wt_.store(nxt_ct, std::memory_order_release);
+            cur_ct = nxt_ct;
+            nxt_ct = cur_ct + 1;
+            el = elems + circ::index_of(cur_ct);
+        }
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&&, E*) {
+        wrapper->elems()->disconnect_receiver(1);
+        return false;
+    }
+
+    template <typename W, typename F, typename R, 
+              template <std::size_t, std::size_t> class E, std::size_t DS, std::size_t AS>
+    bool pop(W* /*wrapper*/, circ::u2_t& /*cur*/, F&& f, R&& out, E<DS, AS>* elems) {
+        byte_t buff[DS];
+        for (unsigned k = 0;;) {
+            auto cur_rd = rd_.load(std::memory_order_relaxed);
+            auto cur_wt = wt_.load(std::memory_order_acquire);
+            auto id_rd  = circ::index_of(cur_rd);
+            auto id_wt  = circ::index_of(cur_wt);
+            if (id_rd == id_wt) {
+                auto* el = elems + id_wt;
+                auto cac_ct = el->f_ct_.load(std::memory_order_acquire);
+                if ((~cac_ct) != cur_wt) {
+                    return false; // empty
+                }
+                if (el->f_ct_.compare_exchange_weak(cac_ct, 0, std::memory_order_relaxed)) {
+                    wt_.store(cur_wt + 1, std::memory_order_release);
+                }
+                k = 0;
+            }
+            else {
+                std::memcpy(buff, &(elems[circ::index_of(cur_rd)].data_), sizeof(buff));
+                if (rd_.compare_exchange_weak(cur_rd, cur_rd + 1, std::memory_order_release)) {
+                    std::forward<F>(f)(buff);
+                    std::forward<R>(out)(true);
+                    return true;
+                }
+                ipc::yield(k);
+            }
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::single, relat::multi, trans::broadcast>> {
+
+    using rc_t = std::uint64_t;
+
+    enum : rc_t {
+        ep_mask = 0x00000000ffffffffull,
+        ep_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t> rc_ { 0 }; // read-counter
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> wt_;   // write index
+    alignas(cache_line_size) rc_t epoch_ { 0 };             // only one writer
+
+    circ::u2_t cursor() const noexcept {
+        return wt_.load(std::memory_order_acquire);
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch_)) {
+                return false; // has not finished yet
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        epoch_ += ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(wt_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & ep_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, epoch_ | static_cast<rc_t>(cc), std::memory_order_release)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        std::forward<F>(f)(&(el->data_));
+        wt_.fetch_add(1, std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E* elems) {
+        if (cur == cursor()) return false; // acquire
+        auto* el = elems + circ::index_of(cur++);
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & ep_mask) == 0) {
+                std::forward<R>(out)(true);
+                return true;
+            }
+            auto nxt_rc = cur_rc & ~static_cast<rc_t>(wrapper->connected_id());
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)((nxt_rc & ep_mask) == 0);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+template <>
+struct prod_cons_impl<wr<relat::multi, relat::multi, trans::broadcast>> {
+
+    using rc_t   = std::uint64_t;
+    using flag_t = std::uint64_t;
+
+    enum : rc_t {
+        rc_mask = 0x00000000ffffffffull,
+        ep_mask = 0x00ffffffffffffffull,
+        ep_incr = 0x0100000000000000ull,
+        ic_mask = 0xff000000ffffffffull,
+        ic_incr = 0x0000000100000000ull
+    };
+
+    template <std::size_t DataSize, std::size_t AlignSize>
+    struct elem_t {
+        std::aligned_storage_t<DataSize, AlignSize> data_ {};
+        std::atomic<rc_t  > rc_   { 0 }; // read-counter
+        std::atomic<flag_t> f_ct_ { 0 }; // commit flag
+    };
+
+    alignas(cache_line_size) std::atomic<circ::u2_t> ct_;   // commit index
+    alignas(cache_line_size) std::atomic<rc_t> epoch_ { 0 };
+
+    circ::u2_t cursor() const noexcept {
+        return ct_.load(std::memory_order_acquire);
+    }
+
+    constexpr static rc_t inc_rc(rc_t rc) noexcept {
+        return (rc & ic_mask) | ((rc + ic_incr) & ~ic_mask);
+    }
+
+    constexpr static rc_t inc_mask(rc_t rc) noexcept {
+        return inc_rc(rc) & ~rc_mask;
+    }
+
+    template <typename W, typename F, typename E>
+    bool push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.load(std::memory_order_acquire);
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_relaxed);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if ((cc & rem_cc) && ((cur_rc & ~ep_mask) == epoch)) {
+                return false; // has not finished yet
+            }
+            else if (!rem_cc) {
+                auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+                if ((cur_fl != cur_ct) && cur_fl) {
+                    return false; // full
+                }
+            }
+            // consider rem_cc to be 0 here
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed) &&
+                epoch_.compare_exchange_weak(epoch, epoch, std::memory_order_acq_rel)) {
+                break;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename E>
+    bool force_push(W* wrapper, F&& f, E* elems) {
+        E* el;
+        circ::u2_t cur_ct;
+        rc_t epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+        for (unsigned k = 0;;) {
+            circ::cc_t cc = wrapper->elems()->connections(std::memory_order_relaxed);
+            if (cc == 0) return false; // no reader
+            el = elems + circ::index_of(cur_ct = ct_.load(std::memory_order_relaxed));
+            // check all consumers have finished reading this element
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            circ::cc_t rem_cc = cur_rc & rc_mask;
+            if (cc & rem_cc) {
+                ipc::log("force_push: k = %u, cc = %u, rem_cc = %u\n", k, cc, rem_cc);
+                cc = wrapper->elems()->disconnect_receiver(rem_cc); // disconnect all invalid readers
+                if (cc == 0) return false; // no reader
+            }
+            // just compare & exchange
+            if (el->rc_.compare_exchange_weak(
+                        cur_rc, inc_mask(epoch | (cur_rc & ep_mask)) | static_cast<rc_t>(cc), std::memory_order_relaxed)) {
+                if (epoch == epoch_.load(std::memory_order_acquire)) {
+                    break;
+                }
+                else if (push(wrapper, std::forward<F>(f), elems)) {
+                    return true;
+                }
+                epoch = epoch_.fetch_add(ep_incr, std::memory_order_release) + ep_incr;
+            }
+            ipc::yield(k);
+        }
+        // only one thread/process would touch here at one time
+        ct_.store(cur_ct + 1, std::memory_order_release);
+        std::forward<F>(f)(&(el->data_));
+        // set flag & try update wt
+        el->f_ct_.store(~static_cast<flag_t>(cur_ct), std::memory_order_release);
+        return true;
+    }
+
+    template <typename W, typename F, typename R, typename E, std::size_t N>
+    bool pop(W* wrapper, circ::u2_t& cur, F&& f, R&& out, E(& elems)[N]) {
+        auto* el = elems + circ::index_of(cur);
+        auto cur_fl = el->f_ct_.load(std::memory_order_acquire);
+        if (cur_fl != ~static_cast<flag_t>(cur)) {
+            return false; // empty
+        }
+        ++cur;
+        std::forward<F>(f)(&(el->data_));
+        for (unsigned k = 0;;) {
+            auto cur_rc = el->rc_.load(std::memory_order_acquire);
+            if ((cur_rc & rc_mask) == 0) {
+                std::forward<R>(out)(true);
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+                return true;
+            }
+            auto nxt_rc = inc_rc(cur_rc) & ~static_cast<rc_t>(wrapper->connected_id());
+            bool last_one = false;
+            if ((last_one = (nxt_rc & rc_mask) == 0)) {
+                el->f_ct_.store(cur + N - 1, std::memory_order_release);
+            }
+            if (el->rc_.compare_exchange_weak(cur_rc, nxt_rc, std::memory_order_release)) {
+                std::forward<R>(out)(last_one);
+                return true;
+            }
+            ipc::yield(k);
+        }
+    }
+};
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/queue.h

@@ -0,0 +1,216 @@
+#pragma once
+
+#include <type_traits>
+#include <new>
+#include <utility>  // [[since C++14]]: std::exchange
+#include <algorithm>
+#include <atomic>
+#include <tuple>
+#include <thread>
+#include <chrono>
+#include <string>
+#include <cassert>  // assert
+
+#include "libipc/def.h"
+#include "libipc/shm.h"
+#include "libipc/rw_lock.h"
+
+#include "libipc/utility/log.h"
+#include "libipc/platform/detail.h"
+#include "libipc/circ/elem_def.h"
+
+namespace ipc {
+namespace detail {
+
+class queue_conn {
+protected:
+    circ::cc_t connected_ = 0;
+    shm::handle elems_h_;
+
+    template <typename Elems>
+    Elems* open(char const * name) {
+        if (name == nullptr || name[0] == '\0') {
+            ipc::error("fail open waiter: name is empty!\n");
+            return nullptr;
+        }
+        if (!elems_h_.acquire(name, sizeof(Elems))) {
+            return nullptr;
+        }
+        auto elems = static_cast<Elems*>(elems_h_.get());
+        if (elems == nullptr) {
+            ipc::error("fail acquire elems: %s\n", name);
+            return nullptr;
+        }
+        elems->init();
+        return elems;
+    }
+
+    void close() {
+        elems_h_.release();
+    }
+
+public:
+    queue_conn() = default;
+    queue_conn(const queue_conn&) = delete;
+    queue_conn& operator=(const queue_conn&) = delete;
+
+    bool connected() const noexcept {
+        return connected_ != 0;
+    }
+
+    circ::cc_t connected_id() const noexcept {
+        return connected_;
+    }
+
+    template <typename Elems>
+    auto connect(Elems* elems) noexcept
+                         /*needs 'optional' here*/
+     -> std::tuple<bool, bool, decltype(std::declval<Elems>().cursor())> {
+        if (elems == nullptr) return {};
+        // if it's already connected, just return
+        if (connected()) return {connected(), false, 0};
+        connected_ = elems->connect_receiver();
+        return {connected(), true, elems->cursor()};
+    }
+
+    template <typename Elems>
+    bool disconnect(Elems* elems) noexcept {
+        if (elems == nullptr) return false;
+        // if it's already disconnected, just return false
+        if (!connected()) return false;
+        elems->disconnect_receiver(std::exchange(connected_, 0));
+        return true;
+    }
+};
+
+template <typename Elems>
+class queue_base : public queue_conn {
+    using base_t = queue_conn;
+
+public:
+    using elems_t  = Elems;
+    using policy_t = typename elems_t::policy_t;
+
+protected:
+    elems_t * elems_ = nullptr;
+    decltype(std::declval<elems_t>().cursor()) cursor_ = 0;
+    bool sender_flag_ = false;
+
+public:
+    using base_t::base_t;
+
+    queue_base() = default;
+
+    explicit queue_base(char const * name)
+        : queue_base{} {
+        elems_ = open<elems_t>(name);
+    }
+
+    explicit queue_base(elems_t * elems) noexcept
+        : queue_base{} {
+        assert(elems != nullptr);
+        elems_ = elems;
+    }
+
+    /* not virtual */ ~queue_base() {
+        base_t::close();
+    }
+
+    elems_t       * elems()       noexcept { return elems_; }
+    elems_t const * elems() const noexcept { return elems_; }
+
+    bool ready_sending() noexcept {
+        if (elems_ == nullptr) return false;
+        return sender_flag_ || (sender_flag_ = elems_->connect_sender());
+    }
+
+    void shut_sending() noexcept {
+        if (elems_ == nullptr) return;
+        if (!sender_flag_) return;
+        elems_->disconnect_sender();
+    }
+
+    bool connect() noexcept {
+        auto tp = base_t::connect(elems_);
+        if (std::get<0>(tp) && std::get<1>(tp)) {
+            cursor_ = std::get<2>(tp);
+            return true;
+        }
+        return std::get<0>(tp);
+    }
+
+    bool disconnect() noexcept {
+        return base_t::disconnect(elems_);
+    }
+
+    std::size_t conn_count() const noexcept {
+        return (elems_ == nullptr) ? static_cast<std::size_t>(invalid_value) : elems_->conn_count();
+    }
+
+    bool valid() const noexcept {
+        return elems_ != nullptr;
+    }
+
+    bool empty() const noexcept {
+        return !valid() || (cursor_ == elems_->cursor());
+    }
+
+    template <typename T, typename F, typename... P>
+    bool push(F&& prep, P&&... params) {
+        if (elems_ == nullptr) return false;
+        return elems_->push(this, [&](void* p) {
+            if (prep(p)) ::new (p) T(std::forward<P>(params)...);
+        });
+    }
+
+    template <typename T, typename F, typename... P>
+    bool force_push(F&& prep, P&&... params) {
+        if (elems_ == nullptr) return false;
+        return elems_->force_push(this, [&](void* p) {
+            if (prep(p)) ::new (p) T(std::forward<P>(params)...);
+        });
+    }
+
+    template <typename T, typename F>
+    bool pop(T& item, F&& out) {
+        if (elems_ == nullptr) {
+            return false;
+        }
+        return elems_->pop(this, &(this->cursor_), [&item](void* p) {
+            ::new (&item) T(std::move(*static_cast<T*>(p)));
+        }, std::forward<F>(out));
+    }
+};
+
+} // namespace detail
+
+template <typename T, typename Policy>
+class queue final : public detail::queue_base<typename Policy::template elems_t<sizeof(T), alignof(T)>> {
+    using base_t = detail::queue_base<typename Policy::template elems_t<sizeof(T), alignof(T)>>;
+
+public:
+    using value_t = T;
+
+    using base_t::base_t;
+
+    template <typename... P>
+    bool push(P&&... params) {
+        return base_t::template push<T>(std::forward<P>(params)...);
+    }
+
+    template <typename... P>
+    bool force_push(P&&... params) {
+        return base_t::template force_push<T>(std::forward<P>(params)...);
+    }
+
+    bool pop(T& item) {
+        return base_t::pop(item, [](bool) {});
+    }
+
+    template <typename F>
+    bool pop(T& item, F&& out) {
+        return base_t::pop(item, std::forward<F>(out));
+    }
+};
+
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/shm.cpp

@@ -0,0 +1,103 @@
+
+#include <string>
+#include <utility>
+
+#include "libipc/shm.h"
+
+#include "libipc/utility/pimpl.h"
+#include "libipc/memory/resource.h"
+
+namespace ipc {
+namespace shm {
+
+class handle::handle_ : public pimpl<handle_> {
+public:
+    shm::id_t id_ = nullptr;
+    void*     m_  = nullptr;
+
+    ipc::string n_;
+    std::size_t s_ = 0;
+};
+
+handle::handle()
+    : p_(p_->make()) {
+}
+
+handle::handle(char const * name, std::size_t size, unsigned mode)
+    : handle() {
+    acquire(name, size, mode);
+}
+
+handle::handle(handle&& rhs)
+    : handle() {
+    swap(rhs);
+}
+
+handle::~handle() {
+    release();
+    p_->clear();
+}
+
+void handle::swap(handle& rhs) {
+    std::swap(p_, rhs.p_);
+}
+
+handle& handle::operator=(handle rhs) {
+    swap(rhs);
+    return *this;
+}
+
+bool handle::valid() const noexcept {
+    return impl(p_)->m_ != nullptr;
+}
+
+std::size_t handle::size() const noexcept {
+    return impl(p_)->s_;
+}
+
+char const * handle::name() const noexcept {
+    return impl(p_)->n_.c_str();
+}
+
+std::int32_t handle::ref() const noexcept {
+    return shm::get_ref(impl(p_)->id_);
+}
+
+void handle::sub_ref() noexcept {
+    shm::sub_ref(impl(p_)->id_);
+}
+
+bool handle::acquire(char const * name, std::size_t size, unsigned mode) {
+    release();
+    impl(p_)->id_ = shm::acquire((impl(p_)->n_ = name).c_str(), size, mode);
+    impl(p_)->m_  = shm::get_mem(impl(p_)->id_, &(impl(p_)->s_));
+    return valid();
+}
+
+std::int32_t handle::release() {
+    if (impl(p_)->id_ == nullptr) return -1;
+    return shm::release(detach());
+}
+
+void* handle::get() const {
+    return impl(p_)->m_;
+}
+
+void handle::attach(id_t id) {
+    if (id == nullptr) return;
+    release();
+    impl(p_)->id_ = id;
+    impl(p_)->m_  = shm::get_mem(impl(p_)->id_, &(impl(p_)->s_));
+}
+
+id_t handle::detach() {
+    auto old = impl(p_)->id_;
+    impl(p_)->id_ = nullptr;
+    impl(p_)->m_  = nullptr;
+    impl(p_)->s_  = 0;
+    impl(p_)->n_.clear();
+    return old;
+}
+
+} // namespace shm
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/waiter.h

@@ -0,0 +1,83 @@
+#pragma once
+
+#include <utility>
+#include <string>
+#include <mutex>
+#include <atomic>
+
+#include "libipc/def.h"
+#include "libipc/mutex.h"
+#include "libipc/condition.h"
+#include "libipc/platform/detail.h"
+
+namespace ipc {
+namespace detail {
+
+class waiter {
+    ipc::sync::condition cond_;
+    ipc::sync::mutex     lock_;
+    std::atomic<bool>    quit_ {false};
+
+public:
+    static void init();
+
+    waiter() = default;
+    waiter(char const *name) {
+        open(name);
+    }
+
+    ~waiter() {
+        close();
+    }
+
+    bool valid() const noexcept {
+        return cond_.valid() && lock_.valid();
+    }
+
+    bool open(char const *name) noexcept {
+        quit_.store(false, std::memory_order_relaxed);
+        if (!cond_.open((std::string{"_waiter_cond_"} + name).c_str())) {
+            return false;
+        }
+        if (!lock_.open((std::string{"_waiter_lock_"} + name).c_str())) {
+            cond_.close();
+            return false;
+        }
+        return valid();
+    }
+
+    void close() noexcept {
+        cond_.close();
+        lock_.close();
+    }
+
+    template <typename F>
+    bool wait_if(F &&pred, std::uint64_t tm = ipc::invalid_value) noexcept {
+        IPC_UNUSED_ std::lock_guard<ipc::sync::mutex> guard {lock_};
+        while ([this, &pred] {
+                    return !quit_.load(std::memory_order_relaxed)
+                        && std::forward<F>(pred)();
+                }()) {
+            if (!cond_.wait(lock_, tm)) return false;
+        }
+        return true;
+    }
+
+    bool notify() noexcept {
+        std::lock_guard<ipc::sync::mutex>{lock_}; // barrier
+        return cond_.notify(lock_);
+    }
+
+    bool broadcast() noexcept {
+        std::lock_guard<ipc::sync::mutex>{lock_}; // barrier
+        return cond_.broadcast(lock_);
+    }
+
+    bool quit_waiting() {
+        quit_.store(true, std::memory_order_release);
+        return broadcast();
+    }
+};
+
+} // namespace detail
+} // namespace ipc

crazy_functions/test_project/cpp/cppipc/来源

@@ -0,0 +1,3 @@
+https://github.com/mutouyun/cpp-ipc
+
+A high-performance inter-process communication library using shared memory on Linux/Windows.

crazy_functions/test_project/cpp/libJPG/jpgd.h

@@ -0,0 +1,316 @@
+// jpgd.h - C++ class for JPEG decompression.
+// Public domain, Rich Geldreich <richgel99@gmail.com>
+#ifndef JPEG_DECODER_H
+#define JPEG_DECODER_H
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <setjmp.h>
+
+namespace jpgd
+{
+  typedef unsigned char  uint8;
+  typedef   signed short int16;
+  typedef unsigned short uint16;
+  typedef unsigned int   uint;
+  typedef   signed int   int32;
+
+  // Loads a JPEG image from a memory buffer or a file.
+  // req_comps can be 1 (grayscale), 3 (RGB), or 4 (RGBA).
+  // On return, width/height will be set to the image's dimensions, and actual_comps will be set to the either 1 (grayscale) or 3 (RGB).
+  // Notes: For more control over where and how the source data is read, see the decompress_jpeg_image_from_stream() function below, or call the jpeg_decoder class directly.
+  // Requesting a 8 or 32bpp image is currently a little faster than 24bpp because the jpeg_decoder class itself currently always unpacks to either 8 or 32bpp.
+// BEGIN EPIC MOD
+//unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps);
+  unsigned char *decompress_jpeg_image_from_memory(const unsigned char *pSrc_data, int src_data_size, int *width, int *height, int *actual_comps, int req_comps, int format);
+// END EPIC MOD
+  unsigned char *decompress_jpeg_image_from_file(const char *pSrc_filename, int *width, int *height, int *actual_comps, int req_comps);
+
+  // Success/failure error codes.
+  enum jpgd_status
+  {
+    JPGD_SUCCESS = 0, JPGD_FAILED = -1, JPGD_DONE = 1,
+    JPGD_BAD_DHT_COUNTS = -256, JPGD_BAD_DHT_INDEX, JPGD_BAD_DHT_MARKER, JPGD_BAD_DQT_MARKER, JPGD_BAD_DQT_TABLE, 
+    JPGD_BAD_PRECISION, JPGD_BAD_HEIGHT, JPGD_BAD_WIDTH, JPGD_TOO_MANY_COMPONENTS, 
+    JPGD_BAD_SOF_LENGTH, JPGD_BAD_VARIABLE_MARKER, JPGD_BAD_DRI_LENGTH, JPGD_BAD_SOS_LENGTH,
+    JPGD_BAD_SOS_COMP_ID, JPGD_W_EXTRA_BYTES_BEFORE_MARKER, JPGD_NO_ARITHMITIC_SUPPORT, JPGD_UNEXPECTED_MARKER,
+    JPGD_NOT_JPEG, JPGD_UNSUPPORTED_MARKER, JPGD_BAD_DQT_LENGTH, JPGD_TOO_MANY_BLOCKS,
+    JPGD_UNDEFINED_QUANT_TABLE, JPGD_UNDEFINED_HUFF_TABLE, JPGD_NOT_SINGLE_SCAN, JPGD_UNSUPPORTED_COLORSPACE,
+    JPGD_UNSUPPORTED_SAMP_FACTORS, JPGD_DECODE_ERROR, JPGD_BAD_RESTART_MARKER, JPGD_ASSERTION_ERROR,
+    JPGD_BAD_SOS_SPECTRAL, JPGD_BAD_SOS_SUCCESSIVE, JPGD_STREAM_READ, JPGD_NOTENOUGHMEM
+  };
+    
+  // Input stream interface.
+  // Derive from this class to read input data from sources other than files or memory. Set m_eof_flag to true when no more data is available.
+  // The decoder is rather greedy: it will keep on calling this method until its internal input buffer is full, or until the EOF flag is set.
+  // It the input stream contains data after the JPEG stream's EOI (end of image) marker it will probably be pulled into the internal buffer.
+  // Call the get_total_bytes_read() method to determine the actual size of the JPEG stream after successful decoding.
+  class jpeg_decoder_stream
+  {
+  public:
+    jpeg_decoder_stream() { }
+    virtual ~jpeg_decoder_stream() { }
+
+    // The read() method is called when the internal input buffer is empty.
+    // Parameters:
+    // pBuf - input buffer
+    // max_bytes_to_read - maximum bytes that can be written to pBuf
+    // pEOF_flag - set this to true if at end of stream (no more bytes remaining)
+    // Returns -1 on error, otherwise return the number of bytes actually written to the buffer (which may be 0).
+    // Notes: This method will be called in a loop until you set *pEOF_flag to true or the internal buffer is full.
+    virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag) = 0;
+  };
+
+  // stdio FILE stream class.
+  class jpeg_decoder_file_stream : public jpeg_decoder_stream
+  {
+    jpeg_decoder_file_stream(const jpeg_decoder_file_stream &);
+    jpeg_decoder_file_stream &operator =(const jpeg_decoder_file_stream &);
+
+    FILE *m_pFile;
+    bool m_eof_flag, m_error_flag;
+
+  public:
+    jpeg_decoder_file_stream();
+    virtual ~jpeg_decoder_file_stream();
+    
+    bool open(const char *Pfilename);
+    void close();
+
+    virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag);
+  };
+
+  // Memory stream class.
+  class jpeg_decoder_mem_stream : public jpeg_decoder_stream
+  {
+    const uint8 *m_pSrc_data;
+    uint m_ofs, m_size;
+
+  public:
+    jpeg_decoder_mem_stream() : m_pSrc_data(NULL), m_ofs(0), m_size(0) { }
+    jpeg_decoder_mem_stream(const uint8 *pSrc_data, uint size) : m_pSrc_data(pSrc_data), m_ofs(0), m_size(size) { }
+
+    virtual ~jpeg_decoder_mem_stream() { }
+
+    bool open(const uint8 *pSrc_data, uint size);
+    void close() { m_pSrc_data = NULL; m_ofs = 0; m_size = 0; }
+    
+    virtual int read(uint8 *pBuf, int max_bytes_to_read, bool *pEOF_flag);
+  };
+
+  // Loads JPEG file from a jpeg_decoder_stream.
+  unsigned char *decompress_jpeg_image_from_stream(jpeg_decoder_stream *pStream, int *width, int *height, int *actual_comps, int req_comps);
+
+  enum 
+  { 
+    JPGD_IN_BUF_SIZE = 8192, JPGD_MAX_BLOCKS_PER_MCU = 10, JPGD_MAX_HUFF_TABLES = 8, JPGD_MAX_QUANT_TABLES = 4, 
+    JPGD_MAX_COMPONENTS = 4, JPGD_MAX_COMPS_IN_SCAN = 4, JPGD_MAX_BLOCKS_PER_ROW = 8192, JPGD_MAX_HEIGHT = 16384, JPGD_MAX_WIDTH = 16384 
+  };
+          
+  typedef int16 jpgd_quant_t;
+  typedef int16 jpgd_block_t;
+
+  class jpeg_decoder
+  {
+  public:
+    // Call get_error_code() after constructing to determine if the stream is valid or not. You may call the get_width(), get_height(), etc.
+    // methods after the constructor is called. You may then either destruct the object, or begin decoding the image by calling begin_decoding(), then decode() on each scanline.
+    jpeg_decoder(jpeg_decoder_stream *pStream);
+
+    ~jpeg_decoder();
+
+    // Call this method after constructing the object to begin decompression.
+    // If JPGD_SUCCESS is returned you may then call decode() on each scanline.
+    int begin_decoding();
+
+    // Returns the next scan line.
+    // For grayscale images, pScan_line will point to a buffer containing 8-bit pixels (get_bytes_per_pixel() will return 1). 
+    // Otherwise, it will always point to a buffer containing 32-bit RGBA pixels (A will always be 255, and get_bytes_per_pixel() will return 4).
+    // Returns JPGD_SUCCESS if a scan line has been returned.
+    // Returns JPGD_DONE if all scan lines have been returned.
+    // Returns JPGD_FAILED if an error occurred. Call get_error_code() for a more info.
+    int decode(const void** pScan_line, uint* pScan_line_len);
+    
+    inline jpgd_status get_error_code() const { return m_error_code; }
+
+    inline int get_width() const { return m_image_x_size; }
+    inline int get_height() const { return m_image_y_size; }
+
+    inline int get_num_components() const { return m_comps_in_frame; }
+
+    inline int get_bytes_per_pixel() const { return m_dest_bytes_per_pixel; }
+    inline int get_bytes_per_scan_line() const { return m_image_x_size * get_bytes_per_pixel(); }
+
+    // Returns the total number of bytes actually consumed by the decoder (which should equal the actual size of the JPEG file).
+    inline int get_total_bytes_read() const { return m_total_bytes_read; }
+    
+  private:
+    jpeg_decoder(const jpeg_decoder &);
+    jpeg_decoder &operator =(const jpeg_decoder &);
+
+    typedef void (*pDecode_block_func)(jpeg_decoder *, int, int, int);
+
+    struct huff_tables
+    {
+      bool ac_table;
+      uint  look_up[256];
+      uint  look_up2[256];
+      uint8 code_size[256];
+      uint  tree[512];
+    };
+
+    struct coeff_buf
+    {
+      uint8 *pData;
+      int block_num_x, block_num_y;
+      int block_len_x, block_len_y;
+      int block_size;
+    };
+
+    struct mem_block
+    {
+      mem_block *m_pNext;
+      size_t m_used_count;
+      size_t m_size;
+      char m_data[1];
+    };
+
+    jmp_buf m_jmp_state;
+    mem_block *m_pMem_blocks;
+    int m_image_x_size;
+    int m_image_y_size;
+    jpeg_decoder_stream *m_pStream;
+    int m_progressive_flag;
+    uint8 m_huff_ac[JPGD_MAX_HUFF_TABLES];
+    uint8* m_huff_num[JPGD_MAX_HUFF_TABLES];      // pointer to number of Huffman codes per bit size
+    uint8* m_huff_val[JPGD_MAX_HUFF_TABLES];      // pointer to Huffman codes per bit size
+    jpgd_quant_t* m_quant[JPGD_MAX_QUANT_TABLES]; // pointer to quantization tables
+    int m_scan_type;                              // Gray, Yh1v1, Yh1v2, Yh2v1, Yh2v2 (CMYK111, CMYK4114 no longer supported)
+    int m_comps_in_frame;                         // # of components in frame
+    int m_comp_h_samp[JPGD_MAX_COMPONENTS];       // component's horizontal sampling factor
+    int m_comp_v_samp[JPGD_MAX_COMPONENTS];       // component's vertical sampling factor
+    int m_comp_quant[JPGD_MAX_COMPONENTS];        // component's quantization table selector
+    int m_comp_ident[JPGD_MAX_COMPONENTS];        // component's ID
+    int m_comp_h_blocks[JPGD_MAX_COMPONENTS];
+    int m_comp_v_blocks[JPGD_MAX_COMPONENTS];
+    int m_comps_in_scan;                          // # of components in scan
+    int m_comp_list[JPGD_MAX_COMPS_IN_SCAN];      // components in this scan
+    int m_comp_dc_tab[JPGD_MAX_COMPONENTS];       // component's DC Huffman coding table selector
+    int m_comp_ac_tab[JPGD_MAX_COMPONENTS];       // component's AC Huffman coding table selector
+    int m_spectral_start;                         // spectral selection start
+    int m_spectral_end;                           // spectral selection end
+    int m_successive_low;                         // successive approximation low
+    int m_successive_high;                        // successive approximation high
+    int m_max_mcu_x_size;                         // MCU's max. X size in pixels
+    int m_max_mcu_y_size;                         // MCU's max. Y size in pixels
+    int m_blocks_per_mcu;
+    int m_max_blocks_per_row;
+    int m_mcus_per_row, m_mcus_per_col;
+    int m_mcu_org[JPGD_MAX_BLOCKS_PER_MCU];
+    int m_total_lines_left;                       // total # lines left in image
+    int m_mcu_lines_left;                         // total # lines left in this MCU
+    int m_real_dest_bytes_per_scan_line;
+    int m_dest_bytes_per_scan_line;               // rounded up
+    int m_dest_bytes_per_pixel;                   // 4 (RGB) or 1 (Y)
+    huff_tables* m_pHuff_tabs[JPGD_MAX_HUFF_TABLES];
+    coeff_buf* m_dc_coeffs[JPGD_MAX_COMPONENTS];
+    coeff_buf* m_ac_coeffs[JPGD_MAX_COMPONENTS];
+    int m_eob_run;
+    int m_block_y_mcu[JPGD_MAX_COMPONENTS];
+    uint8* m_pIn_buf_ofs;
+    int m_in_buf_left;
+    int m_tem_flag;
+    bool m_eof_flag;
+    uint8 m_in_buf_pad_start[128];
+    uint8 m_in_buf[JPGD_IN_BUF_SIZE + 128];
+    uint8 m_in_buf_pad_end[128];
+    int m_bits_left;
+    uint m_bit_buf;
+    int m_restart_interval;
+    int m_restarts_left;
+    int m_next_restart_num;
+    int m_max_mcus_per_row;
+    int m_max_blocks_per_mcu;
+    int m_expanded_blocks_per_mcu;
+    int m_expanded_blocks_per_row;
+    int m_expanded_blocks_per_component;
+    bool  m_freq_domain_chroma_upsample;
+    int m_max_mcus_per_col;
+    uint m_last_dc_val[JPGD_MAX_COMPONENTS];
+    jpgd_block_t* m_pMCU_coefficients;
+    int m_mcu_block_max_zag[JPGD_MAX_BLOCKS_PER_MCU];
+    uint8* m_pSample_buf;
+    int m_crr[256];
+    int m_cbb[256];
+    int m_crg[256];
+    int m_cbg[256];
+    uint8* m_pScan_line_0;
+    uint8* m_pScan_line_1;
+    jpgd_status m_error_code;
+    bool m_ready_flag;
+    int m_total_bytes_read;
+
+    void free_all_blocks();
+    // BEGIN EPIC MOD
+    UE_NORETURN void stop_decoding(jpgd_status status);
+    // END EPIC MOD
+    void *alloc(size_t n, bool zero = false);
+    void word_clear(void *p, uint16 c, uint n);
+    void prep_in_buffer();
+    void read_dht_marker();
+    void read_dqt_marker();
+    void read_sof_marker();
+    void skip_variable_marker();
+    void read_dri_marker();
+    void read_sos_marker();
+    int next_marker();
+    int process_markers();
+    void locate_soi_marker();
+    void locate_sof_marker();
+    int locate_sos_marker();
+    void init(jpeg_decoder_stream * pStream);
+    void create_look_ups();
+    void fix_in_buffer();
+    void transform_mcu(int mcu_row);
+    void transform_mcu_expand(int mcu_row);
+    coeff_buf* coeff_buf_open(int block_num_x, int block_num_y, int block_len_x, int block_len_y);
+    inline jpgd_block_t *coeff_buf_getp(coeff_buf *cb, int block_x, int block_y);
+    void load_next_row();
+    void decode_next_row();
+    void make_huff_table(int index, huff_tables *pH);
+    void check_quant_tables();
+    void check_huff_tables();
+    void calc_mcu_block_order();
+    int init_scan();
+    void init_frame();
+    void process_restart();
+    void decode_scan(pDecode_block_func decode_block_func);
+    void init_progressive();
+    void init_sequential();
+    void decode_start();
+    void decode_init(jpeg_decoder_stream * pStream);
+    void H2V2Convert();
+    void H2V1Convert();
+    void H1V2Convert();
+    void H1V1Convert();
+    void gray_convert();
+    void expanded_convert();
+    void find_eoi();
+    inline uint get_char();
+    inline uint get_char(bool *pPadding_flag);
+    inline void stuff_char(uint8 q);
+    inline uint8 get_octet();
+    inline uint get_bits(int num_bits);
+    inline uint get_bits_no_markers(int numbits);
+    inline int huff_decode(huff_tables *pH);
+    inline int huff_decode(huff_tables *pH, int& extrabits);
+    static inline uint8 clamp(int i);
+    static void decode_block_dc_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_dc_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_ac_first(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+    static void decode_block_ac_refine(jpeg_decoder *pD, int component_id, int block_x, int block_y);
+  };
+  
+} // namespace jpgd
+
+#endif // JPEG_DECODER_H

crazy_functions/test_project/cpp/libJPG/jpge.h

@@ -0,0 +1,172 @@
+
+// jpge.h - C++ class for JPEG compression.
+// Public domain, Rich Geldreich <richgel99@gmail.com>
+// Alex Evans: Added RGBA support, linear memory allocator.
+#ifndef JPEG_ENCODER_H
+#define JPEG_ENCODER_H
+
+#include <stdint.h>
+
+namespace jpge
+{
+  typedef unsigned char  uint8;
+  typedef signed short   int16;
+  typedef signed int     int32;
+  typedef unsigned short uint16;
+  typedef unsigned int   uint32;
+  typedef unsigned int   uint;
+
+  // JPEG chroma subsampling factors. Y_ONLY (grayscale images) and H2V2 (color images) are the most common.
+  enum subsampling_t { Y_ONLY = 0, H1V1 = 1, H2V1 = 2, H2V2 = 3 };
+
+  // JPEG compression parameters structure.
+  struct params
+  {
+    inline params() : m_quality(85), m_subsampling(H2V2), m_no_chroma_discrim_flag(false), m_two_pass_flag(false) { }
+
+    inline bool check_valid() const
+    {
+      if ((m_quality < 1) || (m_quality > 100)) return false;
+      if ((uint)m_subsampling > (uint)H2V2) return false;
+      return true;
+    }
+
+    // Quality: 1-100, higher is better. Typical values are around 50-95.
+    int m_quality;
+
+    // m_subsampling:
+    // 0 = Y (grayscale) only
+    // 1 = YCbCr, no subsampling (H1V1, YCbCr 1x1x1, 3 blocks per MCU)
+    // 2 = YCbCr, H2V1 subsampling (YCbCr 2x1x1, 4 blocks per MCU)
+    // 3 = YCbCr, H2V2 subsampling (YCbCr 4x1x1, 6 blocks per MCU-- very common)
+    subsampling_t m_subsampling;
+
+    // Disables CbCr discrimination - only intended for testing.
+    // If true, the Y quantization table is also used for the CbCr channels.
+    bool m_no_chroma_discrim_flag;
+
+    bool m_two_pass_flag;
+  };
+  
+  // Writes JPEG image to a file. 
+  // num_channels must be 1 (Y) or 3 (RGB), image pitch must be width*num_channels.
+  bool compress_image_to_jpeg_file(const char *pFilename, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params = params());
+
+  // Writes JPEG image to memory buffer. 
+  // On entry, buf_size is the size of the output buffer pointed at by pBuf, which should be at least ~1024 bytes. 
+  // If return value is true, buf_size will be set to the size of the compressed data.
+  bool compress_image_to_jpeg_file_in_memory(void *pBuf, int64_t &buf_size, int64_t width, int64_t height, int64_t num_channels, const uint8 *pImage_data, const params &comp_params = params());
+    
+  // Output stream abstract class - used by the jpeg_encoder class to write to the output stream. 
+  // put_buf() is generally called with len==JPGE_OUT_BUF_SIZE bytes, but for headers it'll be called with smaller amounts.
+  class output_stream
+  {
+  public:
+    virtual ~output_stream() { };
+    virtual bool put_buf(const void* Pbuf, int64_t len) = 0;
+    template<class T> inline bool put_obj(const T& obj) { return put_buf(&obj, sizeof(T)); }
+  };
+    
+  // Lower level jpeg_encoder class - useful if more control is needed than the above helper functions.
+  class jpeg_encoder
+  {
+  public:
+    jpeg_encoder();
+    ~jpeg_encoder();
+
+    // Initializes the compressor.
+    // pStream: The stream object to use for writing compressed data.
+    // params - Compression parameters structure, defined above.
+    // width, height  - Image dimensions.
+    // channels - May be 1, or 3. 1 indicates grayscale, 3 indicates RGB source data.
+    // Returns false on out of memory or if a stream write fails.
+    bool init(output_stream *pStream, int64_t width, int64_t height, int64_t src_channels, const params &comp_params = params());
+    
+    const params &get_params() const { return m_params; }
+    
+    // Deinitializes the compressor, freeing any allocated memory. May be called at any time.
+    void deinit();
+
+    uint get_total_passes() const { return m_params.m_two_pass_flag ? 2 : 1; }
+    inline uint get_cur_pass() { return m_pass_num; }
+
+    // Call this method with each source scanline.
+    // width * src_channels bytes per scanline is expected (RGB or Y format).
+    // You must call with NULL after all scanlines are processed to finish compression.
+    // Returns false on out of memory or if a stream write fails.
+    bool process_scanline(const void* pScanline);
+        
+  private:
+    jpeg_encoder(const jpeg_encoder &);
+    jpeg_encoder &operator =(const jpeg_encoder &);
+
+    typedef int32 sample_array_t;
+        
+    output_stream *m_pStream;
+    params m_params;
+    uint8 m_num_components;
+    uint8 m_comp_h_samp[3], m_comp_v_samp[3];
+    int m_image_x, m_image_y, m_image_bpp, m_image_bpl;
+    int m_image_x_mcu, m_image_y_mcu;
+    int m_image_bpl_xlt, m_image_bpl_mcu;
+    int m_mcus_per_row;
+    int m_mcu_x, m_mcu_y;
+    uint8 *m_mcu_lines[16];
+    uint8 m_mcu_y_ofs;
+    sample_array_t m_sample_array[64];
+    int16 m_coefficient_array[64];
+    int32 m_quantization_tables[2][64];
+    uint m_huff_codes[4][256];
+    uint8 m_huff_code_sizes[4][256];
+    uint8 m_huff_bits[4][17];
+    uint8 m_huff_val[4][256];
+    uint32 m_huff_count[4][256];
+    int m_last_dc_val[3];
+    enum { JPGE_OUT_BUF_SIZE = 2048 };
+    uint8 m_out_buf[JPGE_OUT_BUF_SIZE];
+    uint8 *m_pOut_buf;
+    uint m_out_buf_left;
+    uint32 m_bit_buffer;
+    uint m_bits_in;
+    uint8 m_pass_num;
+    bool m_all_stream_writes_succeeded;
+        
+    void optimize_huffman_table(int table_num, int table_len);
+    void emit_byte(uint8 i);
+    void emit_word(uint i);
+    void emit_marker(int marker);
+    void emit_jfif_app0();
+    void emit_dqt();
+    void emit_sof();
+    void emit_dht(uint8 *bits, uint8 *val, int index, bool ac_flag);
+    void emit_dhts();
+    void emit_sos();
+    void emit_markers();
+    void compute_huffman_table(uint *codes, uint8 *code_sizes, uint8 *bits, uint8 *val);
+    void compute_quant_table(int32 *dst, int16 *src);
+    void adjust_quant_table(int32 *dst, int32 *src);
+    void first_pass_init();
+    bool second_pass_init();
+    bool jpg_open(int p_x_res, int p_y_res, int src_channels);
+    void load_block_8_8_grey(int x);
+    void load_block_8_8(int x, int y, int c);
+    void load_block_16_8(int x, int c);
+    void load_block_16_8_8(int x, int c);
+    void load_quantized_coefficients(int component_num);
+    void flush_output_buffer();
+    void put_bits(uint bits, uint len);
+    void code_coefficients_pass_one(int component_num);
+    void code_coefficients_pass_two(int component_num);
+    void code_block(int component_num);
+    void process_mcu_row();
+    bool terminate_pass_one();
+    bool terminate_pass_two();
+    bool process_end_of_image();
+    void load_mcu(const void* src);
+    void clear();
+    void init();
+  };
+
+} // namespace jpge
+
+#endif // JPEG_ENCODER

crazy_functions/test_project/cpp/libJPG/来源

@@ -0,0 +1,3 @@
+jpge.h - C++ class for JPEG compression.
+Public domain, Rich Geldreich <richgel99@gmail.com>
+Alex Evans: Added RGBA support, linear memory allocator.

crazy_functions/test_project/latex/attention/background.tex

@@ -0,0 +1,58 @@
+The goal of reducing sequential computation also forms the foundation of the Extended Neural GPU \citep{extendedngpu}, ByteNet \citep{NalBytenet2017} and ConvS2S \citep{JonasFaceNet2017}, all of which use convolutional neural networks as basic building block, computing hidden representations in parallel for all input and output positions. In these models, the number of operations required to relate signals from two arbitrary input or output positions grows in the distance between positions, linearly for ConvS2S and logarithmically for ByteNet. This makes it more difficult to learn dependencies between distant positions \citep{hochreiter2001gradient}. In the Transformer this is reduced to a constant number of operations, albeit at the cost of reduced effective resolution due to averaging attention-weighted positions, an effect we counteract with Multi-Head Attention as described in section~\ref{sec:attention}. 
+
+Self-attention, sometimes called intra-attention is an attention mechanism relating different positions of a single sequence in order to compute a representation of the sequence. Self-attention has been used successfully in a variety of tasks including reading comprehension, abstractive summarization, textual entailment and learning task-independent sentence representations \citep{cheng2016long, decomposableAttnModel, paulus2017deep, lin2017structured}.
+
+End-to-end memory networks are based on a recurrent attention mechanism instead of sequence-aligned recurrence and have been shown to perform well on simple-language question answering and language modeling tasks \citep{sukhbaatar2015}.
+
+To the best of our knowledge, however, the Transformer is the first transduction model relying entirely on self-attention to compute representations of its input and output without using sequence-aligned RNNs or convolution.
+In the following sections, we will describe the Transformer, motivate self-attention and discuss its advantages over models such as \citep{neural_gpu, NalBytenet2017} and \citep{JonasFaceNet2017}.
+
+
+%\citep{JonasFaceNet2017} report new SOTA on machine translation for English-to-German (EnDe), Enlish-to-French (EnFr) and English-to-Romanian language pairs. 
+
+%For example,! in MT, we must draw information from both input and previous output words to translate an output word accurately. An attention layer \citep{bahdanau2014neural} can connect a very large number of positions at low computation cost, making it an essential ingredient in competitive recurrent models for machine translation.
+
+%A natural question to ask then is, "Could we replace recurrence with attention?". \marginpar{Don't know if it's the most natural question to ask given the previous statements. Also, need to say that the complexity table summarizes these statements} Such a model would be blessed with the computational efficiency of attention and the power of cross-positional communication. In this work, show that pure attention models work remarkably well for MT, achieving new SOTA results on EnDe and EnFr, and can be trained in under $2$ days on xyz architecture. 
+
+%After the seminal models introduced in \citep{sutskever14, bahdanau2014neural, cho2014learning}, recurrent models have become the dominant solution for both sequence modeling and sequence-to-sequence transduction. Many efforts such as \citep{wu2016google,luong2015effective,jozefowicz2016exploring} have pushed the boundaries of machine translation (MT) and language modeling with recurrent endoder-decoder and recurrent language models. Recent effort \citep{shazeer2017outrageously} has successfully combined the power of conditional computation with sequence models to train very large models for MT, pushing SOTA at lower computational cost.
+
+%Recurrent models compute a vector of hidden states $h_t$, for each time step $t$ of computation. $h_t$ is a function of both the input at time $t$ and the previous hidden state $h_t$. This dependence on the previous hidden state precludes processing all timesteps at once, instead requiring long sequences of sequential operations.  In practice, this results in greatly reduced computational efficiency, as on modern computing hardware, a single operation on a large batch is much faster than a large number of operations on small batches.  The problem gets worse at longer sequence lengths. Although sequential computation is not a severe bottleneck at inference time, as autoregressively generating each output requires all previous outputs, the inability to compute scores at all output positions at once hinders us from rapidly training our models over large datasets. Although impressive work such as \citep{Kuchaiev2017Factorization} is able to significantly accelerate the training of LSTMs with factorization tricks, we are still bound by the linear dependence on sequence length.
+
+%If the model could compute hidden states at each time step using only the inputs and outputs,  it would be liberated from the dependence on results from previous time steps during training. This line of thought is the foundation of recent efforts such as the Markovian neural GPU \citep{neural_gpu}, ByteNet \citep{NalBytenet2017} and ConvS2S \citep{JonasFaceNet2017}, all of which use convolutional neural networks as a building block to compute hidden representations simultaneously for all timesteps, resulting in $O(1)$ sequential time complexity. \citep{JonasFaceNet2017} report new SOTA on machine translation for English-to-German (EnDe), Enlish-to-French (EnFr) and English-to-Romanian language pairs. 
+
+%A crucial component for accurate sequence prediction is modeling cross-positional communication. For example, in MT, we must draw information from both input and previous output words to translate an output word accurately. An attention layer \citep{bahdanau2014neural} can connect a very large number of positions at a low computation cost, also $O(1)$ sequential time complexity, making it an essential ingredient in recurrent encoder-decoder architectures for MT. A natural question to ask then is, "Could we replace recurrence with attention?". \marginpar{Don't know if it's the most natural question to ask given the previous statements. Also, need to say that the complexity table summarizes these statements} Such a model would be blessed with the computational efficiency of attention and the power of cross-positional communication. In this work, show that pure attention models work remarkably well for MT, achieving new SOTA results on EnDe and EnFr, and can be trained in under $2$ days on xyz architecture. 
+
+
+
+%Note: Facebook model is no better than RNNs in this regard, since it requires a number of layers proportional to the distance you want to communicate.  Bytenet is more promising, since it requires a logarithmnic number of layers (does bytenet have SOTA results)?   
+
+%Note: An attention  layer can connect a very large number of positions at a low computation cost in O(1) sequential operations.  This is why encoder-decoder attention has been so successful in seq-to-seq models so far.  It is only natural, then, to also use attention to connect the timesteps of the same sequence.
+
+%Note: I wouldn't say that long sequences are not a problem during inference.  It would be great if we could infer with no long sequences.  We could just say later on that, while our training graph is constant-depth, our model still requires sequential operations in the decoder part during inference due to the autoregressive nature of the model.   
+
+%\begin{table}[h!]
+%\caption{Attention models are quite efficient for cross-positional communications when sequence length is smaller than channel depth. $n$ represents the sequence length and $d$ represents the channel depth.}
+%\label{tab:op_complexities}
+%\begin{center}
+%\vspace{-5pt}
+%\scalebox{0.75}{
+
+%\begin{tabular}{l|c|c|c}
+%\hline \hline
+%Layer Type & Receptive & Complexity & Sequential  \\
+%           & Field     &            & Operations  \\
+%\hline
+%Pointwise Feed-Forward & $1$ & $O(n \cdot d^2)$ & $O(1)$ \\
+%\hline
+%Recurrent & $n$ & $O(n \cdot d^2)$ & $O(n)$ \\
+%\hline
+%Convolutional & $r$ & $O(r \cdot n \cdot d^2)$ & $O(1)$ \\
+%\hline
+%Convolutional (separable) & $r$ & $O(r \cdot n \cdot d + n %\cdot d^2)$ & $O(1)$ \\
+%\hline
+%Attention & $r$ & $O(r \cdot n \cdot d)$ & $O(1)$ \\
+%\hline \hline
+%\end{tabular}
+%}
+%\end{center}
+%\end{table}

crazy_functions/test_project/latex/attention/introduction.tex

@@ -0,0 +1,18 @@
+Recurrent neural networks, long short-term memory \citep{hochreiter1997} and gated recurrent \citep{gruEval14} neural networks in particular, have been firmly established as state of the art approaches in sequence modeling and transduction problems such as language modeling and machine translation \citep{sutskever14, bahdanau2014neural, cho2014learning}. Numerous efforts have since continued to push the boundaries of recurrent language models and encoder-decoder architectures \citep{wu2016google,luong2015effective,jozefowicz2016exploring}.
+
+Recurrent models typically factor computation along the symbol positions of the input and output sequences. Aligning the positions to steps in computation time, they generate a sequence of hidden states $h_t$, as a function of the previous hidden state $h_{t-1}$ and the input for position $t$. This inherently sequential nature precludes parallelization within training examples, which becomes critical at longer sequence lengths, as memory constraints limit batching across examples.
+%\marginpar{not sure if the memory constraints are understandable here}
+Recent work has achieved significant improvements in computational efficiency through factorization tricks \citep{Kuchaiev2017Factorization} and conditional computation \citep{shazeer2017outrageously}, while also improving model performance in case of the latter. The fundamental constraint of sequential computation, however, remains.
+
+%\marginpar{@all: there is work on analyzing what attention really does in seq2seq models, couldn't find it right away} 
+
+Attention mechanisms have become an integral part of compelling sequence modeling and transduction models in various tasks, allowing modeling of dependencies without regard to their distance in the input or output sequences \citep{bahdanau2014neural, structuredAttentionNetworks}. In all but a few cases \citep{decomposableAttnModel}, however, such attention mechanisms are used in conjunction with a recurrent network.
+
+%\marginpar{not sure if "cross-positional communication" is understandable without explanation}
+%\marginpar{insert exact training times and stats for the model that reaches sota earliest, maybe even a single GPU model?}
+
+In this work we propose the Transformer, a model architecture eschewing recurrence and instead relying entirely on an attention mechanism to draw global dependencies between input and output. The Transformer allows for significantly more parallelization and can reach a new state of the art in translation quality after being trained for as little as twelve hours on eight P100 GPUs. 
+%\marginpar{you removed the constant number of repetitions part. I wrote it because I wanted to make it clear that the model does not only perform attention once, while it's also not recurrent. I thought that might be important to get across early.}
+
+% Just a standard paragraph with citations, rewrite.
+%After the seminal papers of \citep{sutskever14}, \citep{bahdanau2014neural}, and \citep{cho2014learning}, recurrent models have become the dominant solution for both sequence modeling and sequence-to-sequence transduction. Many efforts such as \citep{wu2016google,luong2015effective,jozefowicz2016exploring} have pushed the boundaries of machine translation and language modeling with recurrent sequence models. Recent effort \citep{shazeer2017outrageously} has combined the power of conditional computation with sequence models to train very large models for machine translation, pushing SOTA at lower computational cost. Recurrent models compute a vector of hidden states $h_t$, for each time step $t$ of computation. $h_t$ is a function of both the input at time $t$ and the previous hidden state $h_t$. This dependence on the previous hidden state encumbers recurrnet models to process multiple inputs at once, and their time complexity is a linear function of the length of the input and output, both during training and inference. [What I want to say here is that although this is fine during decoding, at training time, we are given both input and output and this linear nature does not allow the RNN to process all inputs and outputs simultaneously and haven't been used on datasets that are the of the scale of the web. What's the largest dataset we have ? . Talk about Nividia and possibly other's effors to speed up things, and possibly other efforts that alleviate this, but are still limited by it's comptuational nature]. Rest of the intro: What if you could construct the state based on the actual inputs and outputs, then you could construct them all at once. This has been the foundation of many promising recent efforts, bytenet,facenet (Also talk about quasi rnn here). Now we talk about attention!! Along with cell architectures such as long short-term meory (LSTM) \citep{hochreiter1997}, and gated recurrent units (GRUs) \citep{cho2014learning}, attention has emerged as an essential ingredient in successful sequence models, in particular for machine translation. In recent years, many, if not all, state-of-the-art (SOTA) results in machine translation have been achieved with attention-based sequence models \citep{wu2016google,luong2015effective,jozefowicz2016exploring}. Talk about the neon work on how it played with attention to do self attention! Then talk about what we do.

crazy_functions/test_project/latex/attention/model_architecture.tex

@@ -0,0 +1,155 @@
+
+\begin{figure}
+  \centering
+  \includegraphics[scale=0.6]{Figures/ModalNet-21}
+  \caption{The Transformer - model architecture.}
+  \label{fig:model-arch}
+\end{figure}
+
+% Although the primary workhorse of our model is attention, 
+%Our model maintains the encoder-decoder structure that is common to many so-called sequence-to-sequence models \citep{bahdanau2014neural,sutskever14}.  As in all such architectures, the encoder computes a representation of the input sequence, and the decoder consumes these representations along with the output tokens to autoregressively produce the output sequence.  Where, traditionally, the encoder and decoder contain stacks of recurrent or convolutional layers, our encoder and decoder stacks are composed of attention layers and position-wise feed-forward layers (Figure~\ref{fig:model-arch}).  The following sections describe the gross architecture and these particular components in detail.
+
+Most competitive neural sequence transduction models have an encoder-decoder structure \citep{cho2014learning,bahdanau2014neural,sutskever14}. Here, the encoder maps an input sequence of symbol representations $(x_1, ..., x_n)$ to a sequence of continuous representations $\mathbf{z} = (z_1, ..., z_n)$. Given $\mathbf{z}$, the decoder then generates an output sequence $(y_1,...,y_m)$ of symbols one element at a time. At each step the model is auto-regressive \citep{graves2013generating}, consuming the previously generated symbols as additional input when generating the next.
+
+The Transformer follows this overall architecture using stacked self-attention and point-wise, fully connected layers for both the encoder and decoder, shown in the left and right halves of Figure~\ref{fig:model-arch}, respectively.
+
+\subsection{Encoder and Decoder Stacks}
+
+\paragraph{Encoder:}The encoder is composed of a stack of $N=6$ identical layers. Each layer has two sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position-wise fully connected feed-forward network.   We employ a residual connection \citep{he2016deep} around each of the two sub-layers, followed by layer normalization \cite{layernorm2016}.  That is, the output of each sub-layer is $\mathrm{LayerNorm}(x + \mathrm{Sublayer}(x))$, where $\mathrm{Sublayer}(x)$ is the function implemented by the sub-layer itself.  To facilitate these residual connections, all sub-layers in the model, as well as the embedding layers, produce outputs of dimension $\dmodel=512$.
+
+\paragraph{Decoder:}The decoder is also composed of a stack of $N=6$ identical layers.  In addition to the two sub-layers in each encoder layer, the decoder inserts a third sub-layer, which performs multi-head attention over the output of the encoder stack.  Similar to the encoder, we employ residual connections around each of the sub-layers, followed by layer normalization.  We also modify the self-attention sub-layer in the decoder stack to prevent positions from attending to subsequent positions.  This masking, combined with fact that the output embeddings are offset by one position, ensures that the predictions for position $i$ can depend only on the known outputs at positions less than $i$.
+
+% In our model (Figure~\ref{fig:model-arch}), the encoder and decoder are composed of stacks of alternating self-attention layers (for cross-positional communication) and position-wise feed-forward layers (for in-place computation).  In addition, the decoder stack contains encoder-decoder attention layers.  Since attention is agnostic to the distances between words, our model requires a "positional encoding" to be added to the encoder and decoder input. The following sections describe all of these components in detail.
+
+\subsection{Attention} \label{sec:attention}
+An attention function can be described as mapping a query and a set of key-value pairs to an output, where the query, keys, values, and output are all vectors.  The output is computed as a weighted sum of the values, where the weight assigned to each value is computed by a compatibility function of the query with the corresponding key.
+
+\subsubsection{Scaled Dot-Product Attention} \label{sec:scaled-dot-prod}
+
+% \begin{figure}
+%   \centering
+%   \includegraphics[scale=0.6]{Figures/ModalNet-19}
+%   \caption{Scaled Dot-Product Attention.}
+%   \label{fig:multi-head-att}
+% \end{figure}
+
+We call our particular attention "Scaled Dot-Product Attention" (Figure~\ref{fig:multi-head-att}).   The input consists of queries and keys of dimension $d_k$, and values of dimension $d_v$.  We compute the dot products of the query with all keys, divide each by $\sqrt{d_k}$, and apply a softmax function to obtain the weights on the values.
+
+In practice, we compute the attention function on a set of queries simultaneously, packed together into a matrix $Q$.   The keys and values are also packed together into matrices $K$ and $V$.  We compute the matrix of outputs as:
+
+\begin{equation}
+   \mathrm{Attention}(Q, K, V) = \mathrm{softmax}(\frac{QK^T}{\sqrt{d_k}})V
+\end{equation}
+
+The two most commonly used attention functions are additive attention \citep{bahdanau2014neural}, and dot-product (multiplicative) attention.  Dot-product attention is identical to our algorithm, except for the scaling factor of $\frac{1}{\sqrt{d_k}}$. Additive attention computes the compatibility function using a feed-forward network with a single hidden layer.  While the two are similar in theoretical complexity, dot-product attention is much faster and more space-efficient in practice, since it can be implemented using highly optimized matrix multiplication code. 
+
+%We scale the dot products by $1/\sqrt{d_k}$ to limit the magnitude of the dot products, which works well in practice. Otherwise, we found applying the softmax to often result in weights very close to 0 or 1, and hence minuscule gradients.
+
+% Already described in the subsequent section
+%When used as part of decoder self-attention, an optional mask function is applied just before the softmax to prevent positions from attending to subsequent positions.   This mask simply sets the logits corresponding to all illegal connections (those outside of the lower triangle) to $-\infty$.
+
+%\paragraph{Comparison to Additive Attention: } We choose dot product attention over additive attention \citep{bahdanau2014neural} since it can be computed using highly optimized matrix multiplication code.  This optimization is particularly important to us, as we employ many attention layers in our model.
+
+While for small values of $d_k$ the two mechanisms perform similarly, additive attention outperforms dot product attention without scaling for larger values of $d_k$ \citep{DBLP:journals/corr/BritzGLL17}. We suspect that for large values of $d_k$, the dot products grow large in magnitude, pushing the softmax function into regions where it has extremely small gradients  \footnote{To illustrate why the dot products get large, assume that the components of $q$ and $k$ are independent random variables with mean $0$ and variance $1$.  Then their dot product, $q \cdot k = \sum_{i=1}^{d_k} q_ik_i$, has mean $0$ and variance $d_k$.}. To counteract this effect, we scale the dot products by $\frac{1}{\sqrt{d_k}}$.
+
+
+%We suspect this to be caused by the dot products growing too large in magnitude to result in useful gradients after applying the softmax function.  To counteract this, we scale the dot product by $1/\sqrt{d_k}$.
+
+
+\subsubsection{Multi-Head Attention} \label{sec:multihead}
+
+\begin{figure}
+\begin{minipage}[t]{0.5\textwidth}
+  \centering
+  Scaled Dot-Product Attention \\
+  \vspace{0.5cm}
+  \includegraphics[scale=0.6]{Figures/ModalNet-19}
+\end{minipage}
+\begin{minipage}[t]{0.5\textwidth}
+  \centering 
+  Multi-Head Attention \\
+  \vspace{0.1cm}
+  \includegraphics[scale=0.6]{Figures/ModalNet-20}  
+\end{minipage}
+
+
+  % \centering
+
+  \caption{(left) Scaled Dot-Product Attention. (right) Multi-Head Attention consists of several attention layers running in parallel.}
+  \label{fig:multi-head-att}
+\end{figure}
+
+Instead of performing a single attention function with $\dmodel$-dimensional keys, values and queries, we found it beneficial to linearly project the queries, keys and values $h$ times with different, learned linear projections to $d_k$, $d_k$ and $d_v$ dimensions, respectively.
+On each of these projected versions of queries, keys and values we then perform the attention function in parallel, yielding $d_v$-dimensional output values. These are concatenated and once again projected, resulting in the final values, as depicted in Figure~\ref{fig:multi-head-att}.
+
+Multi-head attention allows the model to jointly attend to information from different representation subspaces at different positions. With a single attention head, averaging inhibits this.
+
+\begin{align*}
+    \mathrm{MultiHead}(Q, K, V) &= \mathrm{Concat}(\mathrm{head_1}, ..., \mathrm{head_h})W^O\\
+%    \mathrm{where} \mathrm{head_i} &= \mathrm{Attention}(QW_Q_i^{\dmodel \times d_q}, KW_K_i^{\dmodel \times d_k}, VW^V_i^{\dmodel \times d_v})\\
+    \text{where}~\mathrm{head_i} &= \mathrm{Attention}(QW^Q_i, KW^K_i, VW^V_i)\\
+\end{align*}
+
+Where the projections are parameter matrices $W^Q_i \in \mathbb{R}^{\dmodel \times d_k}$, $W^K_i \in \mathbb{R}^{\dmodel \times d_k}$, $W^V_i \in \mathbb{R}^{\dmodel \times d_v}$ and $W^O \in \mathbb{R}^{hd_v \times \dmodel}$.
+
+
+%find it better (and no more expensive) to have multiple parallel attention layers (each over the full set of positions) with proportionally lower-dimensional keys, values and queries.  We call this "Multi-Head Attention" (Figure~\ref{fig:multi-head-att}).  The keys, values, and queries for each of these parallel attention layers are computed by learned linear transformations of the inputs to the multi-head attention.  We use different linear transformations across different parallel attention layers.  The output of the parallel attention layers are concatenated, and then passed through a final learned linear transformation. 
+
+In this work we employ $h=8$ parallel attention layers, or heads. For each of these we use $d_k=d_v=\dmodel/h=64$.
+Due to the reduced dimension of each head, the total computational cost is similar to that of single-head attention with full dimensionality.
+
+\subsubsection{Applications of Attention in our Model}
+
+The Transformer uses multi-head attention in three different ways: 
+\begin{itemize}
+ \item In "encoder-decoder attention" layers, the queries come from the previous decoder layer, and the memory keys and values come from the output of the encoder.   This allows every position in the decoder to attend over all positions in the input sequence.  This mimics the typical encoder-decoder attention mechanisms in sequence-to-sequence models such as \citep{wu2016google, bahdanau2014neural,JonasFaceNet2017}.
+
+ \item The encoder contains self-attention layers.  In a self-attention layer all of the keys, values and queries come from the same place, in this case, the output of the previous layer in the encoder.   Each position in the encoder can attend to all positions in the previous layer of the encoder.
+
+ \item Similarly, self-attention layers in the decoder allow each position in the decoder to attend to all positions in the decoder up to and including that position.  We need to prevent leftward information flow in the decoder to preserve the auto-regressive property.  We implement this inside of scaled dot-product attention by masking out (setting to $-\infty$) all values in the input of the softmax which correspond to illegal connections.  See Figure~\ref{fig:multi-head-att}.
+
+\end{itemize}
+
+\subsection{Position-wise Feed-Forward Networks}\label{sec:ffn}
+
+In addition to attention sub-layers, each of the layers in our encoder and decoder contains a fully connected feed-forward network, which is applied to each position separately and identically.  This consists of two linear transformations with a ReLU activation in between.
+
+\begin{equation}
+   \mathrm{FFN}(x)=\max(0, xW_1 + b_1) W_2 + b_2
+\end{equation}
+
+While the linear transformations are the same across different positions, they use different parameters from layer to layer. Another way of describing this is as two convolutions with kernel size 1.  The dimensionality of input and output is $\dmodel=512$, and the inner-layer has dimensionality $d_{ff}=2048$.
+
+
+
+%In the appendix, we describe how the position-wise feed-forward network can also be seen as a form of attention.
+
+%from Jakob: The number of operations required for the model to relate signals from two arbitrary input or output positions grows in the distance between positions in input or output, linearly for ConvS2S and logarithmically for ByteNet, making it harder to learn dependencies between these positions \citep{hochreiter2001gradient}. In the transformer this is reduced to a constant number of operations, albeit at the cost of effective resolution caused by averaging attention-weighted positions, an effect we aim to counteract with multi-headed attention.
+
+
+%Figure~\ref{fig:simple-att} presents a simple attention function, $A$, with a single head, that forms the basis of our multi-head attention. $A$ takes a query key vector $\kq$, matrices of memory keys $\km$ and memory values $\vm$ ,and produces a query value vector $\vq$ as 
+%\begin{equation*} \label{eq:attention}
+%    A(\kq, \km, \vm) = {\vm}^T (Softmax(\km \kq).
+%\end{equation*}
+%We linearly transform $\kq,\,\km$, and $\vm$ with learned matrices ${\Wkq \text{,} \, \Wkm}$, and ${\Wvm}$ before calling the attention function, and transform the output query with $\Wvq$ before handing it to the feed forward layer. Each attention layer has it's own set of transformation matrices, which are shared across all query positions. $A$ is applied in parallel for each query position, and is implemented very efficiently as a batch of matrix multiplies. The self-attention and encoder-decoder attention layers use $A$, but with different arguments. For example, in encdoder self-attention, queries in encoder layer $i$ attention to memories in encoder layer $i-1$. To ensure that decoder self-attention layers do not look at future words, we add $- \inf$ to the softmax logits in positions $j+1$ to query length for query position $l$.  
+
+%In simple attention, the query value is a weighted combination of the memory values where the attention weights sum to one. Although this function performs well in practice, the constraint on attention weights can restrict the amount of information that flows from memories to queries because the query cannot focus on multiple memory positions at once, which might be desirable when translating long sequences. \marginpar{@usz, could you think of an example of this ?} We remedy this by maintaining multiple attention heads at each query position that attend to all memory positions in parallel, with a different set of parameters  per attention head $h$. 
+%\marginpar{}
+
+\subsection{Embeddings and Softmax}
+Similarly to other sequence transduction models, we use learned embeddings to convert the input tokens and output tokens to vectors of dimension $\dmodel$.  We also use the usual learned linear transformation and softmax function to convert the decoder output to predicted next-token probabilities.  In our model, we share the same weight matrix between the two embedding layers and the pre-softmax linear transformation, similar to \citep{press2016using}.   In the embedding layers, we multiply those weights by $\sqrt{\dmodel}$.
+
+
+\subsection{Positional Encoding}
+Since our model contains no recurrence and no convolution, in order for the model to make use of the order of the sequence, we must inject some information about the relative or absolute position of the tokens in the sequence.  To this end, we add "positional encodings" to the input embeddings at the bottoms of the encoder and decoder stacks.  The positional encodings have the same dimension $\dmodel$ as the embeddings, so that the two can be summed.   There are many choices of positional encodings, learned and fixed \citep{JonasFaceNet2017}.
+
+In this work, we use sine and cosine functions of different frequencies:
+
+\begin{align*}
+    PE_{(pos,2i)} = sin(pos / 10000^{2i/\dmodel}) \\
+    PE_{(pos,2i+1)} = cos(pos / 10000^{2i/\dmodel})
+\end{align*}
+
+where $pos$ is the position and $i$ is the dimension.  That is, each dimension of the positional encoding corresponds to a sinusoid.  The wavelengths form a geometric progression from $2\pi$ to $10000 \cdot 2\pi$.  We chose this function because we hypothesized it would allow the model to easily learn to attend by relative positions, since for any fixed offset $k$, $PE_{pos+k}$ can be represented as a linear function of $PE_{pos}$.
+
+We also experimented with using learned positional embeddings \citep{JonasFaceNet2017} instead, and found that the two versions produced nearly identical results (see Table~\ref{tab:variations} row (E)).  We chose the sinusoidal version because it may allow the model to extrapolate to sequence lengths longer than the ones encountered during training.

crazy_functions/test_project/latex/attention/parameter_attention.tex

@@ -0,0 +1,45 @@
+\pagebreak
+\section*{Two Feed-Forward Layers = Attention over Parameters}\label{sec:parameter_attention}
+
+In addition to attention layers, our model contains position-wise feed-forward networks (Section \ref{sec:ffn}), which consist of two linear transformations with a ReLU activation in between.  In fact, these networks too can be seen as a form of attention.  Compare the formula for such a network with the formula for a simple dot-product attention layer (biases and scaling factors omitted):
+
+\begin{align*}
+    FFN(x, W_1, W_2) = ReLU(xW_1)W_2 \\
+    A(q, K, V) = Softmax(qK^T)V
+\end{align*}
+
+Based on the similarity of these formulae, the two-layer feed-forward network can be seen as a kind of attention, where the keys and values are the rows of the trainable parameter matrices $W_1$ and $W_2$, and where we use ReLU instead of Softmax in the compatibility function.
+
+%the compatablity function is $compat(q, k_i) = ReLU(q \cdot k_i)$ instead of $Softmax(qK_T)_i$.
+
+Given this similarity, we experimented with replacing the position-wise feed-forward networks with attention layers similar to the ones we use everywhere else our model. The multi-head-attention-over-parameters sublayer is identical to the multi-head attention described in \ref{sec:multihead}, except that the "keys" and "values" inputs to each attention head are trainable model parameters, as opposed to being linear projections of a previous layer.  These parameters are scaled up by a factor of $\sqrt{d_{model}}$ in order to be more similar to activations.
+
+In our first experiment, we replaced each position-wise feed-forward network with a multi-head-attention-over-parameters sublayer with $h_p=8$ heads, key-dimensionality $d_{pk}=64$, and value-dimensionality $d_{pv}=64$, using $n_p=1536$ key-value pairs for each attention head.  The sublayer has a total of $2097152$ parameters, including the parameters in the query projection and the output projection.  This matches the number of parameters in the position-wise feed-forward network that we replaced.  While the theoretical amount of computation is also the same, in practice, the attention version caused the step times to be about 30\% longer.
+
+In our second experiment, we used $h_p=8$ heads, and $n_p=512$ key-value pairs for each attention head, again matching the total number of parameters in the base model.
+
+Results for the first experiment were slightly worse than for the base model, and results for the second experiment were slightly better, see Table~\ref{tab:parameter_attention}.
+
+\begin{table}[h]
+\caption{Replacing the position-wise feed-forward networks with multihead-attention-over-parameters produces similar results to the base model.  All metrics are on the English-to-German translation development set, newstest2013.}
+\label{tab:parameter_attention}
+\begin{center}
+\vspace{-2mm}
+%\scalebox{1.0}{
+\begin{tabular}{c|cccccc|cccc}
+\hline\rule{0pt}{2.0ex}
+ & \multirow{2}{*}{$\dmodel$} & \multirow{2}{*}{$\dff$} &
+\multirow{2}{*}{$h_p$} & \multirow{2}{*}{$d_{pk}$} & \multirow{2}{*}{$d_{pv}$} &
+ \multirow{2}{*}{$n_p$} &
+ PPL & BLEU & params & training\\
+ & & & & & &  & (dev) & (dev) & $\times10^6$ & time \\
+\hline\rule{0pt}{2.0ex}
+base & 512 & 2048 & & & & & 4.92 & 25.8 & 65 & 12 hours\\
+\hline\rule{0pt}{2.0ex}
+AOP$_1$ & 512 & & 8 & 64 & 64 & 1536 & 4.92& 25.5  & 65 & 16 hours\\
+AOP$_2$ & 512 & & 16 & 64 & 64 & 512 & \textbf{4.86} & \textbf{25.9}  & 65 & 16 hours \\
+\hline
+\end{tabular}
+%}
+\end{center}
+\end{table}

crazy_functions/test_project/latex/attention/来源

@@ -0,0 +1,8 @@
+chatgpt的老祖宗《Attention is all you need》
+
+Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin
+
+真实的摘要如下
+The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.8 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.
+
+https://arxiv.org/abs/1706.03762

crazy_functions/test_project/python/dqn/__init__.py

@@ -0,0 +1,2 @@
+from stable_baselines3.dqn.dqn import DQN
+from stable_baselines3.dqn.policies import CnnPolicy, MlpPolicy

crazy_functions/test_project/python/dqn/dqn.py

@@ -0,0 +1,245 @@
+from typing import Any, Dict, List, Optional, Tuple, Type, Union
+
+import gym
+import numpy as np
+import torch as th
+from torch.nn import functional as F
+
+from stable_baselines3.common import logger
+from stable_baselines3.common.off_policy_algorithm import OffPolicyAlgorithm
+from stable_baselines3.common.preprocessing import maybe_transpose
+from stable_baselines3.common.type_aliases import GymEnv, MaybeCallback, Schedule
+from stable_baselines3.common.utils import get_linear_fn, is_vectorized_observation, polyak_update
+from stable_baselines3.dqn.policies import DQNPolicy
+
+
+class DQN(OffPolicyAlgorithm):
+    """
+    Deep Q-Network (DQN)
+
+    Paper: https://arxiv.org/abs/1312.5602, https://www.nature.com/articles/nature14236
+    Default hyperparameters are taken from the nature paper,
+    except for the optimizer and learning rate that were taken from Stable Baselines defaults.
+
+    :param policy: The policy model to use (MlpPolicy, CnnPolicy, ...)
+    :param env: The environment to learn from (if registered in Gym, can be str)
+    :param learning_rate: The learning rate, it can be a function
+        of the current progress remaining (from 1 to 0)
+    :param buffer_size: size of the replay buffer
+    :param learning_starts: how many steps of the model to collect transitions for before learning starts
+    :param batch_size: Minibatch size for each gradient update
+    :param tau: the soft update coefficient ("Polyak update", between 0 and 1) default 1 for hard update
+    :param gamma: the discount factor
+    :param train_freq: Update the model every ``train_freq`` steps. Alternatively pass a tuple of frequency and unit
+        like ``(5, "step")`` or ``(2, "episode")``.
+    :param gradient_steps: How many gradient steps to do after each rollout (see ``train_freq``)
+        Set to ``-1`` means to do as many gradient steps as steps done in the environment
+        during the rollout.
+    :param optimize_memory_usage: Enable a memory efficient variant of the replay buffer
+        at a cost of more complexity.
+        See https://github.com/DLR-RM/stable-baselines3/issues/37#issuecomment-637501195
+    :param target_update_interval: update the target network every ``target_update_interval``
+        environment steps.
+    :param exploration_fraction: fraction of entire training period over which the exploration rate is reduced
+    :param exploration_initial_eps: initial value of random action probability
+    :param exploration_final_eps: final value of random action probability
+    :param max_grad_norm: The maximum value for the gradient clipping
+    :param tensorboard_log: the log location for tensorboard (if None, no logging)
+    :param create_eval_env: Whether to create a second environment that will be
+        used for evaluating the agent periodically. (Only available when passing string for the environment)
+    :param policy_kwargs: additional arguments to be passed to the policy on creation
+    :param verbose: the verbosity level: 0 no output, 1 info, 2 debug
+    :param seed: Seed for the pseudo random generators
+    :param device: Device (cpu, cuda, ...) on which the code should be run.
+        Setting it to auto, the code will be run on the GPU if possible.
+    :param _init_setup_model: Whether or not to build the network at the creation of the instance
+    """
+
+    def __init__(
+        self,
+        policy: Union[str, Type[DQNPolicy]],
+        env: Union[GymEnv, str],
+        learning_rate: Union[float, Schedule] = 1e-4,
+        buffer_size: int = 1000000,
+        learning_starts: int = 50000,
+        batch_size: Optional[int] = 32,
+        tau: float = 1.0,
+        gamma: float = 0.99,
+        train_freq: Union[int, Tuple[int, str]] = 4,
+        gradient_steps: int = 1,
+        optimize_memory_usage: bool = False,
+        target_update_interval: int = 10000,
+        exploration_fraction: float = 0.1,
+        exploration_initial_eps: float = 1.0,
+        exploration_final_eps: float = 0.05,
+        max_grad_norm: float = 10,
+        tensorboard_log: Optional[str] = None,
+        create_eval_env: bool = False,
+        policy_kwargs: Optional[Dict[str, Any]] = None,
+        verbose: int = 0,
+        seed: Optional[int] = None,
+        device: Union[th.device, str] = "auto",
+        _init_setup_model: bool = True,
+    ):
+
+        super(DQN, self).__init__(
+            policy,
+            env,
+            DQNPolicy,
+            learning_rate,
+            buffer_size,
+            learning_starts,
+            batch_size,
+            tau,
+            gamma,
+            train_freq,
+            gradient_steps,
+            action_noise=None,  # No action noise
+            policy_kwargs=policy_kwargs,
+            tensorboard_log=tensorboard_log,
+            verbose=verbose,
+            device=device,
+            create_eval_env=create_eval_env,
+            seed=seed,
+            sde_support=False,
+            optimize_memory_usage=optimize_memory_usage,
+            supported_action_spaces=(gym.spaces.Discrete,),
+        )
+
+        self.exploration_initial_eps = exploration_initial_eps
+        self.exploration_final_eps = exploration_final_eps
+        self.exploration_fraction = exploration_fraction
+        self.target_update_interval = target_update_interval
+        self.max_grad_norm = max_grad_norm
+        # "epsilon" for the epsilon-greedy exploration
+        self.exploration_rate = 0.0
+        # Linear schedule will be defined in `_setup_model()`
+        self.exploration_schedule = None
+        self.q_net, self.q_net_target = None, None
+
+        if _init_setup_model:
+            self._setup_model()
+
+    def _setup_model(self) -> None:
+        super(DQN, self)._setup_model()
+        self._create_aliases()
+        self.exploration_schedule = get_linear_fn(
+            self.exploration_initial_eps, self.exploration_final_eps, self.exploration_fraction
+        )
+
+    def _create_aliases(self) -> None:
+        self.q_net = self.policy.q_net
+        self.q_net_target = self.policy.q_net_target
+
+    def _on_step(self) -> None:
+        """
+        Update the exploration rate and target network if needed.
+        This method is called in ``collect_rollouts()`` after each step in the environment.
+        """
+        if self.num_timesteps % self.target_update_interval == 0:
+            polyak_update(self.q_net.parameters(), self.q_net_target.parameters(), self.tau)
+
+        self.exploration_rate = self.exploration_schedule(self._current_progress_remaining)
+        logger.record("rollout/exploration rate", self.exploration_rate)
+
+    def train(self, gradient_steps: int, batch_size: int = 100) -> None:
+        # Update learning rate according to schedule
+        self._update_learning_rate(self.policy.optimizer)
+
+        losses = []
+        for _ in range(gradient_steps):
+            # Sample replay buffer
+            replay_data = self.replay_buffer.sample(batch_size, env=self._vec_normalize_env)
+
+            with th.no_grad():
+                # Compute the next Q-values using the target network
+                next_q_values = self.q_net_target(replay_data.next_observations)
+                # Follow greedy policy: use the one with the highest value
+                next_q_values, _ = next_q_values.max(dim=1)
+                # Avoid potential broadcast issue
+                next_q_values = next_q_values.reshape(-1, 1)
+                # 1-step TD target
+                target_q_values = replay_data.rewards + (1 - replay_data.dones) * self.gamma * next_q_values
+
+            # Get current Q-values estimates
+            current_q_values = self.q_net(replay_data.observations)
+
+            # Retrieve the q-values for the actions from the replay buffer
+            current_q_values = th.gather(current_q_values, dim=1, index=replay_data.actions.long())
+
+            # Compute Huber loss (less sensitive to outliers)
+            loss = F.smooth_l1_loss(current_q_values, target_q_values)
+            losses.append(loss.item())
+
+            # Optimize the policy
+            self.policy.optimizer.zero_grad()
+            loss.backward()
+            # Clip gradient norm
+            th.nn.utils.clip_grad_norm_(self.policy.parameters(), self.max_grad_norm)
+            self.policy.optimizer.step()
+
+        # Increase update counter
+        self._n_updates += gradient_steps
+
+        logger.record("train/n_updates", self._n_updates, exclude="tensorboard")
+        logger.record("train/loss", np.mean(losses))
+
+    def predict(
+        self,
+        observation: np.ndarray,
+        state: Optional[np.ndarray] = None,
+        mask: Optional[np.ndarray] = None,
+        deterministic: bool = False,
+    ) -> Tuple[np.ndarray, Optional[np.ndarray]]:
+        """
+        Overrides the base_class predict function to include epsilon-greedy exploration.
+
+        :param observation: the input observation
+        :param state: The last states (can be None, used in recurrent policies)
+        :param mask: The last masks (can be None, used in recurrent policies)
+        :param deterministic: Whether or not to return deterministic actions.
+        :return: the model's action and the next state
+            (used in recurrent policies)
+        """
+        if not deterministic and np.random.rand() < self.exploration_rate:
+            if is_vectorized_observation(maybe_transpose(observation, self.observation_space), self.observation_space):
+                n_batch = observation.shape[0]
+                action = np.array([self.action_space.sample() for _ in range(n_batch)])
+            else:
+                action = np.array(self.action_space.sample())
+        else:
+            action, state = self.policy.predict(observation, state, mask, deterministic)
+        return action, state
+
+    def learn(
+        self,
+        total_timesteps: int,
+        callback: MaybeCallback = None,
+        log_interval: int = 4,
+        eval_env: Optional[GymEnv] = None,
+        eval_freq: int = -1,
+        n_eval_episodes: int = 5,
+        tb_log_name: str = "DQN",
+        eval_log_path: Optional[str] = None,
+        reset_num_timesteps: bool = True,
+    ) -> OffPolicyAlgorithm:
+
+        return super(DQN, self).learn(
+            total_timesteps=total_timesteps,
+            callback=callback,
+            log_interval=log_interval,
+            eval_env=eval_env,
+            eval_freq=eval_freq,
+            n_eval_episodes=n_eval_episodes,
+            tb_log_name=tb_log_name,
+            eval_log_path=eval_log_path,
+            reset_num_timesteps=reset_num_timesteps,
+        )
+
+    def _excluded_save_params(self) -> List[str]:
+        return super(DQN, self)._excluded_save_params() + ["q_net", "q_net_target"]
+
+    def _get_torch_save_params(self) -> Tuple[List[str], List[str]]:
+        state_dicts = ["policy", "policy.optimizer"]
+
+        return state_dicts, []

crazy_functions/test_project/python/dqn/policies.py

@@ -0,0 +1,237 @@
+from typing import Any, Dict, List, Optional, Type
+
+import gym
+import torch as th
+from torch import nn
+
+from stable_baselines3.common.policies import BasePolicy, register_policy
+from stable_baselines3.common.torch_layers import BaseFeaturesExtractor, FlattenExtractor, NatureCNN, create_mlp
+from stable_baselines3.common.type_aliases import Schedule
+
+
+class QNetwork(BasePolicy):
+    """
+    Action-Value (Q-Value) network for DQN
+
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param net_arch: The specification of the policy and value networks.
+    :param activation_fn: Activation function
+    :param normalize_images: Whether to normalize images or not,
+         dividing by 255.0 (True by default)
+    """
+
+    def __init__(
+        self,
+        observation_space: gym.spaces.Space,
+        action_space: gym.spaces.Space,
+        features_extractor: nn.Module,
+        features_dim: int,
+        net_arch: Optional[List[int]] = None,
+        activation_fn: Type[nn.Module] = nn.ReLU,
+        normalize_images: bool = True,
+    ):
+        super(QNetwork, self).__init__(
+            observation_space,
+            action_space,
+            features_extractor=features_extractor,
+            normalize_images=normalize_images,
+        )
+
+        if net_arch is None:
+            net_arch = [64, 64]
+
+        self.net_arch = net_arch
+        self.activation_fn = activation_fn
+        self.features_extractor = features_extractor
+        self.features_dim = features_dim
+        self.normalize_images = normalize_images
+        action_dim = self.action_space.n  # number of actions
+        q_net = create_mlp(self.features_dim, action_dim, self.net_arch, self.activation_fn)
+        self.q_net = nn.Sequential(*q_net)
+
+    def forward(self, obs: th.Tensor) -> th.Tensor:
+        """
+        Predict the q-values.
+
+        :param obs: Observation
+        :return: The estimated Q-Value for each action.
+        """
+        return self.q_net(self.extract_features(obs))
+
+    def _predict(self, observation: th.Tensor, deterministic: bool = True) -> th.Tensor:
+        q_values = self.forward(observation)
+        # Greedy action
+        action = q_values.argmax(dim=1).reshape(-1)
+        return action
+
+    def _get_constructor_parameters(self) -> Dict[str, Any]:
+        data = super()._get_constructor_parameters()
+
+        data.update(
+            dict(
+                net_arch=self.net_arch,
+                features_dim=self.features_dim,
+                activation_fn=self.activation_fn,
+                features_extractor=self.features_extractor,
+            )
+        )
+        return data
+
+
+class DQNPolicy(BasePolicy):
+    """
+    Policy class with Q-Value Net and target net for DQN
+
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param lr_schedule: Learning rate schedule (could be constant)
+    :param net_arch: The specification of the policy and value networks.
+    :param activation_fn: Activation function
+    :param features_extractor_class: Features extractor to use.
+    :param features_extractor_kwargs: Keyword arguments
+        to pass to the features extractor.
+    :param normalize_images: Whether to normalize images or not,
+         dividing by 255.0 (True by default)
+    :param optimizer_class: The optimizer to use,
+        ``th.optim.Adam`` by default
+    :param optimizer_kwargs: Additional keyword arguments,
+        excluding the learning rate, to pass to the optimizer
+    """
+
+    def __init__(
+        self,
+        observation_space: gym.spaces.Space,
+        action_space: gym.spaces.Space,
+        lr_schedule: Schedule,
+        net_arch: Optional[List[int]] = None,
+        activation_fn: Type[nn.Module] = nn.ReLU,
+        features_extractor_class: Type[BaseFeaturesExtractor] = FlattenExtractor,
+        features_extractor_kwargs: Optional[Dict[str, Any]] = None,
+        normalize_images: bool = True,
+        optimizer_class: Type[th.optim.Optimizer] = th.optim.Adam,
+        optimizer_kwargs: Optional[Dict[str, Any]] = None,
+    ):
+        super(DQNPolicy, self).__init__(
+            observation_space,
+            action_space,
+            features_extractor_class,
+            features_extractor_kwargs,
+            optimizer_class=optimizer_class,
+            optimizer_kwargs=optimizer_kwargs,
+        )
+
+        if net_arch is None:
+            if features_extractor_class == FlattenExtractor:
+                net_arch = [64, 64]
+            else:
+                net_arch = []
+
+        self.net_arch = net_arch
+        self.activation_fn = activation_fn
+        self.normalize_images = normalize_images
+
+        self.net_args = {
+            "observation_space": self.observation_space,
+            "action_space": self.action_space,
+            "net_arch": self.net_arch,
+            "activation_fn": self.activation_fn,
+            "normalize_images": normalize_images,
+        }
+
+        self.q_net, self.q_net_target = None, None
+        self._build(lr_schedule)
+
+    def _build(self, lr_schedule: Schedule) -> None:
+        """
+        Create the network and the optimizer.
+
+        :param lr_schedule: Learning rate schedule
+            lr_schedule(1) is the initial learning rate
+        """
+
+        self.q_net = self.make_q_net()
+        self.q_net_target = self.make_q_net()
+        self.q_net_target.load_state_dict(self.q_net.state_dict())
+
+        # Setup optimizer with initial learning rate
+        self.optimizer = self.optimizer_class(self.parameters(), lr=lr_schedule(1), **self.optimizer_kwargs)
+
+    def make_q_net(self) -> QNetwork:
+        # Make sure we always have separate networks for features extractors etc
+        net_args = self._update_features_extractor(self.net_args, features_extractor=None)
+        return QNetwork(**net_args).to(self.device)
+
+    def forward(self, obs: th.Tensor, deterministic: bool = True) -> th.Tensor:
+        return self._predict(obs, deterministic=deterministic)
+
+    def _predict(self, obs: th.Tensor, deterministic: bool = True) -> th.Tensor:
+        return self.q_net._predict(obs, deterministic=deterministic)
+
+    def _get_constructor_parameters(self) -> Dict[str, Any]:
+        data = super()._get_constructor_parameters()
+
+        data.update(
+            dict(
+                net_arch=self.net_args["net_arch"],
+                activation_fn=self.net_args["activation_fn"],
+                lr_schedule=self._dummy_schedule,  # dummy lr schedule, not needed for loading policy alone
+                optimizer_class=self.optimizer_class,
+                optimizer_kwargs=self.optimizer_kwargs,
+                features_extractor_class=self.features_extractor_class,
+                features_extractor_kwargs=self.features_extractor_kwargs,
+            )
+        )
+        return data
+
+
+MlpPolicy = DQNPolicy
+
+
+class CnnPolicy(DQNPolicy):
+    """
+    Policy class for DQN when using images as input.
+
+    :param observation_space: Observation space
+    :param action_space: Action space
+    :param lr_schedule: Learning rate schedule (could be constant)
+    :param net_arch: The specification of the policy and value networks.
+    :param activation_fn: Activation function
+    :param features_extractor_class: Features extractor to use.
+    :param normalize_images: Whether to normalize images or not,
+         dividing by 255.0 (True by default)
+    :param optimizer_class: The optimizer to use,
+        ``th.optim.Adam`` by default
+    :param optimizer_kwargs: Additional keyword arguments,
+        excluding the learning rate, to pass to the optimizer
+    """
+
+    def __init__(
+        self,
+        observation_space: gym.spaces.Space,
+        action_space: gym.spaces.Space,
+        lr_schedule: Schedule,
+        net_arch: Optional[List[int]] = None,
+        activation_fn: Type[nn.Module] = nn.ReLU,
+        features_extractor_class: Type[BaseFeaturesExtractor] = NatureCNN,
+        features_extractor_kwargs: Optional[Dict[str, Any]] = None,
+        normalize_images: bool = True,
+        optimizer_class: Type[th.optim.Optimizer] = th.optim.Adam,
+        optimizer_kwargs: Optional[Dict[str, Any]] = None,
+    ):
+        super(CnnPolicy, self).__init__(
+            observation_space,
+            action_space,
+            lr_schedule,
+            net_arch,
+            activation_fn,
+            features_extractor_class,
+            features_extractor_kwargs,
+            normalize_images,
+            optimizer_class,
+            optimizer_kwargs,
+        )
+
+
+register_policy("MlpPolicy", MlpPolicy)
+register_policy("CnnPolicy", CnnPolicy)

crazy_functions/test_project/python/dqn/来源

@@ -0,0 +1,2 @@
+github stablebaseline3
+https://github.com/DLR-RM/stable-baselines3

crazy_functions/test_project/其他测试

@@ -0,0 +1,27 @@
+"In practice, we found that a high-entropy initial state is more likely to increase the speed of training.
+The entropy is calculated by:
+$$H=-\sum_{k= 1}^{n_k} p(k) \cdot \log p(k), p(k)=\frac{|A_k|}{|\mathcal{A}|}$$
+where $H$ is the entropy, $|A_k|$ is the number of agent nodes in $k$-th cluster, $|\mathcal{A}|$ is the total number of agents.
+To ensure the Cooperation Graph initialization has higher entropy,
+we will randomly generate multiple initial states, 
+rank by their entropy and then pick the one with maximum $H$."
+
+```
+FROM ubuntu:latest
+
+RUN apt-get update && \
+    apt-get install -y python3 python3-pip && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN echo '[global]' > /etc/pip.conf && \
+    echo 'index-url = https://mirrors.aliyun.com/pypi/simple/' >> /etc/pip.conf && \
+    echo 'trusted-host = mirrors.aliyun.com' >> /etc/pip.conf
+
+RUN pip3 install gradio requests[socks] mdtex2html
+
+COPY . /gpt
+WORKDIR /gpt
+
+
+CMD ["python3", "main.py"]
+```

crazy_functions/代码重写为全英文_多线程.py

@@ -0,0 +1,75 @@
+import threading
+from predict import predict_no_ui_long_connection
+from toolbox import CatchException, write_results_to_file
+
+
+
+@CatchException
+def 全项目切换英文(txt, top_p, temperature, chatbot, history, sys_prompt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    # 集合文件
+    import time, glob, os
+    os.makedirs('gpt_log/generated_english_version', exist_ok=True)
+    os.makedirs('gpt_log/generated_english_version/crazy_functions', exist_ok=True)
+    file_manifest = [f for f in glob.glob('./*.py') if ('test_project' not in f) and ('gpt_log' not in f)] + \
+                    [f for f in glob.glob('./crazy_functions/*.py') if ('test_project' not in f) and ('gpt_log' not in f)]
+    i_say_show_user_buffer = []
+
+    # 随便显示点什么防止卡顿的感觉
+    for index, fp in enumerate(file_manifest):
+        # if 'test_project' in fp: continue
+        with open(fp, 'r', encoding='utf-8') as f:
+            file_content = f.read()
+        i_say_show_user =f'[{index}/{len(file_manifest)}] 接下来请将以下代码中包含的所有中文转化为英文，只输出代码: {os.path.abspath(fp)}'
+        i_say_show_user_buffer.append(i_say_show_user)
+        chatbot.append((i_say_show_user, "[Local Message] 等待多线程操作，中间过程不予显示."))
+        yield chatbot, history, '正常'
+
+    # 任务函数
+    mutable_return = [None for _ in file_manifest]
+    def thread_worker(fp,index):
+        with open(fp, 'r', encoding='utf-8') as f:
+            file_content = f.read()
+        i_say = f'接下来请将以下代码中包含的所有中文转化为英文，只输出代码，文件名是{fp}，文件代码是 ```{file_content}```'
+        # ** gpt request **
+        gpt_say = predict_no_ui_long_connection(inputs=i_say, top_p=top_p, temperature=temperature, history=history, sys_prompt=sys_prompt)
+        mutable_return[index] = gpt_say
+
+    # 所有线程同时开始执行任务函数
+    handles = [threading.Thread(target=thread_worker, args=(fp,index)) for index, fp in enumerate(file_manifest)]
+    for h in handles:
+        h.daemon = True
+        h.start()
+    chatbot.append(('开始了吗？', f'多线程操作已经开始'))
+    yield chatbot, history, '正常'
+
+    # 循环轮询各个线程是否执行完毕
+    cnt = 0
+    while True:
+        time.sleep(1)
+        th_alive = [h.is_alive() for h in handles]
+        if not any(th_alive): break
+        stat = ['执行中' if alive else '已完成' for alive in th_alive]
+        stat_str = '|'.join(stat)
+        cnt += 1
+        chatbot[-1] = (chatbot[-1][0], f'多线程操作已经开始，完成情况: {stat_str}' + ''.join(['.']*(cnt%4)))
+        yield chatbot, history, '正常'
+
+    # 把结果写入文件
+    for index, h in enumerate(handles):
+        h.join() # 这里其实不需要join了，肯定已经都结束了
+        fp = file_manifest[index]
+        gpt_say = mutable_return[index]
+        i_say_show_user = i_say_show_user_buffer[index]
+
+        where_to_relocate = f'gpt_log/generated_english_version/{fp}'
+        with open(where_to_relocate, 'w+', encoding='utf-8') as f: f.write(gpt_say.lstrip('```').rstrip('```'))
+        chatbot.append((i_say_show_user, f'[Local Message] 已完成{os.path.abspath(fp)}的转化，\n\n存入{os.path.abspath(where_to_relocate)}'))
+        history.append(i_say_show_user); history.append(gpt_say)
+        yield chatbot, history, '正常'
+        time.sleep(1)
+
+    # 备份一个文件
+    res = write_results_to_file(history)
+    chatbot.append(("生成一份任务执行报告", res))
+    yield chatbot, history, '正常'

crazy_functions/批量总结PDF文档.py

@@ -0,0 +1,99 @@
+from predict import predict_no_ui
+from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down
+fast_debug = False
+
+
+def 解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt):
+    import time, glob, os, fitz
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        with fitz.open(fp) as doc:
+            file_content = ""
+            for page in doc:
+                file_content += page.get_text()
+            print(file_content)
+
+        prefix = "接下来请你逐文件分析下面的论文文件，概括其内容" if index==0 else ""
+        i_say = prefix + f'请对下面的文章片段用中文做一个概述，文件名是{os.path.relpath(fp, project_folder)}，文章内容是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的文章片段做一个概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        print('[1] yield chatbot, history')
+        yield chatbot, history, '正常'
+
+        if not fast_debug: 
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[])   # 带超时倒计时
+
+            print('[2] end gpt req')
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            print('[3] yield chatbot, history')
+            yield chatbot, history, msg
+            print('[4] next')
+            if not fast_debug: time.sleep(2)
+
+    all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
+    i_say = f'根据以上你自己的分析，对全文进行概括，用学术性语言写一段中文摘要，然后再写一段英文摘要（包括{all_file}）。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield chatbot, history, '正常'
+
+    if not fast_debug: 
+        msg = '正常'
+        # ** gpt request **
+        gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say, chatbot, top_p, temperature, history=history)   # 带超时倒计时
+
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say); history.append(gpt_say)
+        yield chatbot, history, msg
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield chatbot, history, msg
+
+
+@CatchException
+def 批量总结PDF文档(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    import glob, os
+
+    # 基本信息：功能、贡献者
+    chatbot.append([
+        "函数插件功能？",
+        "批量总结PDF文档。函数插件贡献者: ValeriaWong"])
+    yield chatbot, history, '正常'
+
+    # 尝试导入依赖，如果缺少依赖，则给出安装建议
+    try:
+        import fitz
+    except:
+        report_execption(chatbot, history, 
+            a = f"解析项目: {txt}", 
+            b = f"导入软件依赖失败。使用该模块需要额外依赖，安装方法```pip install --upgrade pymupdf```。")
+        yield chatbot, history, '正常'
+        return
+
+    # 清空历史，以免输入溢出
+    history = []
+
+    # 检测输入参数，如没有给定输入参数，直接退出
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield chatbot, history, '正常'
+        return
+
+    # 搜索需要处理的文件清单
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.pdf', recursive=True)] # + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    
+    # 如果没找到任何文件
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex或.pdf文件: {txt}")
+        yield chatbot, history, '正常'
+        return
+
+    # 开始正式执行任务
+    yield from 解析PDF(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)

crazy_functions/生成函数注释.py

@@ -0,0 +1,57 @@
+from predict import predict_no_ui
+from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down
+fast_debug = False
+
+
+def 生成函数注释(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt):
+    import time, glob, os
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8') as f:
+            file_content = f.read()
+
+        i_say = f'请对下面的程序文件做一个概述，并对文件中的所有函数生成注释，使用markdown表格输出结果，文件名是{os.path.relpath(fp, project_folder)}，文件内容是 ```{file_content}```'
+        i_say_show_user = f'[{index}/{len(file_manifest)}] 请对下面的程序文件做一个概述，并对文件中的所有函数生成注释: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        print('[1] yield chatbot, history')
+        yield chatbot, history, '正常'
+
+        if not fast_debug: 
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[])   # 带超时倒计时
+
+            print('[2] end gpt req')
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            print('[3] yield chatbot, history')
+            yield chatbot, history, msg
+            print('[4] next')
+            if not fast_debug: time.sleep(2)
+
+    if not fast_debug: 
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield chatbot, history, msg
+
+
+
+@CatchException
+def 批量生成函数注释(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield chatbot, history, '正常'
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.py', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)]
+
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield chatbot, history, '正常'
+        return
+    yield from 生成函数注释(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)

crazy_functions/解析项目源代码.py

@@ -0,0 +1,149 @@
+from predict import predict_no_ui
+from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down
+fast_debug = False
+
+def 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt):
+    import time, glob, os
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8') as f:
+            file_content = f.read()
+
+        prefix = "接下来请你逐文件分析下面的工程" if index==0 else ""
+        i_say = prefix + f'请对下面的程序文件做一个概述文件名是{os.path.relpath(fp, project_folder)}，文件代码是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的程序文件做一个概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        yield chatbot, history, '正常'
+
+        if not fast_debug: 
+            msg = '正常'
+
+            # ** gpt request **
+            gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[])   # 带超时倒计时
+
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            yield chatbot, history, msg
+            if not fast_debug: time.sleep(2)
+
+    all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
+    i_say = f'根据以上你自己的分析，对程序的整体功能和构架做出概括。然后用一张markdown表格整理每个文件的功能（包括{all_file}）。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield chatbot, history, '正常'
+
+    if not fast_debug: 
+        msg = '正常'
+        # ** gpt request **
+        gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say, chatbot, top_p, temperature, history=history)   # 带超时倒计时
+        
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say); history.append(gpt_say)
+        yield chatbot, history, msg
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield chatbot, history, msg
+
+
+
+
+@CatchException
+def 解析项目本身(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    import time, glob, os
+    file_manifest = [f for f in glob.glob('*.py')]
+    for index, fp in enumerate(file_manifest):
+        # if 'test_project' in fp: continue
+        with open(fp, 'r', encoding='utf-8') as f:
+            file_content = f.read()
+
+        prefix = "接下来请你分析自己的程序构成，别紧张，" if index==0 else ""
+        i_say = prefix + f'请对下面的程序文件做一个概述文件名是{fp}，文件代码是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的程序文件做一个概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        yield chatbot, history, '正常'
+
+        if not fast_debug: 
+            # ** gpt request **
+            # gpt_say = predict_no_ui(inputs=i_say, top_p=top_p, temperature=temperature)
+            gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[])   # 带超时倒计时
+
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            yield chatbot, history, '正常'
+            time.sleep(2)
+
+    i_say = f'根据以上你自己的分析，对程序的整体功能和构架做出概括。然后用一张markdown表格整理每个文件的功能（包括{file_manifest}）。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield chatbot, history, '正常'
+
+    if not fast_debug: 
+        # ** gpt request **
+        # gpt_say = predict_no_ui(inputs=i_say, top_p=top_p, temperature=temperature, history=history)
+        gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say, chatbot, top_p, temperature, history=history)   # 带超时倒计时
+
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say); history.append(gpt_say)
+        yield chatbot, history, '正常'
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield chatbot, history, '正常'
+
+@CatchException
+def 解析一个Python项目(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield chatbot, history, '正常'
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.py', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何python文件: {txt}")
+        yield chatbot, history, '正常'
+        return
+    yield from 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
+
+
+@CatchException
+def 解析一个C项目的头文件(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield chatbot, history, '正常'
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.h', recursive=True)] # + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.h头文件: {txt}")
+        yield chatbot, history, '正常'
+        return
+    yield from 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
+
+@CatchException
+def 解析一个C项目(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield chatbot, history, '正常'
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.h', recursive=True)]  + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.h头文件: {txt}")
+        yield chatbot, history, '正常'
+        return
+    yield from 解析源代码(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)
+

crazy_functions/读文章写摘要.py

@@ -0,0 +1,70 @@
+from predict import predict_no_ui
+from toolbox import CatchException, report_execption, write_results_to_file, predict_no_ui_but_counting_down
+fast_debug = False
+
+
+def 解析Paper(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt):
+    import time, glob, os
+    print('begin analysis on:', file_manifest)
+    for index, fp in enumerate(file_manifest):
+        with open(fp, 'r', encoding='utf-8') as f:
+            file_content = f.read()
+
+        prefix = "接下来请你逐文件分析下面的论文文件，概括其内容" if index==0 else ""
+        i_say = prefix + f'请对下面的文章片段用中文做一个概述，文件名是{os.path.relpath(fp, project_folder)}，文章内容是 ```{file_content}```'
+        i_say_show_user = prefix + f'[{index}/{len(file_manifest)}] 请对下面的文章片段做一个概述: {os.path.abspath(fp)}'
+        chatbot.append((i_say_show_user, "[Local Message] waiting gpt response."))
+        print('[1] yield chatbot, history')
+        yield chatbot, history, '正常'
+
+        if not fast_debug: 
+            msg = '正常'
+            # ** gpt request **
+            gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[])   # 带超时倒计时
+
+            print('[2] end gpt req')
+            chatbot[-1] = (i_say_show_user, gpt_say)
+            history.append(i_say_show_user); history.append(gpt_say)
+            print('[3] yield chatbot, history')
+            yield chatbot, history, msg
+            print('[4] next')
+            if not fast_debug: time.sleep(2)
+
+    all_file = ', '.join([os.path.relpath(fp, project_folder) for index, fp in enumerate(file_manifest)])
+    i_say = f'根据以上你自己的分析，对全文进行概括，用学术性语言写一段中文摘要，然后再写一段英文摘要（包括{all_file}）。'
+    chatbot.append((i_say, "[Local Message] waiting gpt response."))
+    yield chatbot, history, '正常'
+
+    if not fast_debug: 
+        msg = '正常'
+        # ** gpt request **
+        gpt_say = yield from predict_no_ui_but_counting_down(i_say, i_say, chatbot, top_p, temperature, history=history)   # 带超时倒计时
+
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say); history.append(gpt_say)
+        yield chatbot, history, msg
+        res = write_results_to_file(history)
+        chatbot.append(("完成了吗？", res))
+        yield chatbot, history, msg
+
+
+
+@CatchException
+def 读文章写摘要(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    import glob, os
+    if os.path.exists(txt):
+        project_folder = txt
+    else:
+        if txt == "": txt = '空空如也的输入栏'
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到本地项目或无权访问: {txt}")
+        yield chatbot, history, '正常'
+        return
+    file_manifest = [f for f in glob.glob(f'{project_folder}/**/*.tex', recursive=True)] # + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.cpp', recursive=True)] + \
+                    # [f for f in glob.glob(f'{project_folder}/**/*.c', recursive=True)]
+    if len(file_manifest) == 0:
+        report_execption(chatbot, history, a = f"解析项目: {txt}", b = f"找不到任何.tex文件: {txt}")
+        yield chatbot, history, '正常'
+        return
+    yield from 解析Paper(file_manifest, project_folder, top_p, temperature, chatbot, history, systemPromptTxt)

crazy_functions/高级功能函数模板.py

@@ -0,0 +1,25 @@
+from predict import predict_no_ui_long_connection
+from toolbox import CatchException, report_execption, write_results_to_file
+import datetime
+
+@CatchException
+def 高阶功能模板函数(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+    history = []    # 清空历史，以免输入溢出
+    chatbot.append(("这是什么功能？", "[Local Message] 请注意，您正在调用一个函数模板，该函数面向希望实现更多有趣功能的开发者，它可以作为创建新功能函数的模板。为了做到简单易读，该函数只有25行代码，不会实时反馈文字流或心跳，请耐心等待程序输出完成。另外您若希望分享新的功能模组，请不吝PR！"))
+    yield chatbot, history, '正常'  # 由于请求gpt需要一段时间，我们先及时地做一次状态显示
+
+    for i in range(5):
+        currentMonth = (datetime.date.today() + datetime.timedelta(days=i)).month
+        currentDay = (datetime.date.today() + datetime.timedelta(days=i)).day
+        i_say = f'历史中哪些事件发生在{currentMonth}月{currentDay}日？列举两条并发送相关图片。发送图片时，请使用Markdown，将Unsplash API中的PUT_YOUR_QUERY_HERE替换成描述改事件的三个最重要的单词。'
+        chatbot.append((i_say, "[Local Message] waiting gpt response."))
+        yield chatbot, history, '正常'  # 由于请求gpt需要一段时间，我们先及时地做一次状态显示
+
+        # history = [] 每次询问不携带之前的询问历史
+        gpt_say = predict_no_ui_long_connection(
+            inputs=i_say, top_p=top_p, temperature=temperature, history=[], 
+            sys_prompt="当你想发送一张照片时，请使用Markdown, 并且不要有反斜线, 不要用代码块。使用 Unsplash API (https://source.unsplash.com/1280x720/? < PUT_YOUR_QUERY_HERE >)。") # 请求gpt，需要一段时间
+
+        chatbot[-1] = (i_say, gpt_say)
+        history.append(i_say);history.append(gpt_say)
+        yield chatbot, history, '正常'  # 显示

functional.py

@@ -1,67 +1,54 @@
-# """
-# 'primary' for main call-to-action, 
-# 'secondary' for a more subdued style, 
-# 'stop' for a stop button.
-# """
-
+# 'primary' 颜色对应 theme.py 中的 primary_hue
+# 'secondary' 颜色对应 theme.py 中的 neutral_hue
+# 'stop' 颜色对应 theme.py 中的 color_er
+# 默认按钮颜色是 secondary
+from toolbox import clear_line_break
 
 def get_functionals():
     return {
         "英语学术润色": {
-            "Prefix": "Below is a paragraph from an academic paper. Polish the writing to meet the academic style, \
-improve the spelling, grammar, clarity, concision and overall readability. When neccessary, rewrite the whole sentence. \
-Furthermore, list all modification and explain the reasons to do so in markdown table.\n\n",
-            "Button": None,
-            "Suffix": "",
-            "Color": "stop",
+            # 前言
+            "Prefix":   r"Below is a paragraph from an academic paper. Polish the writing to meet the academic style, " +
+                        r"improve the spelling, grammar, clarity, concision and overall readability. When neccessary, rewrite the whole sentence. " +
+                        r"Furthermore, list all modification and explain the reasons to do so in markdown table." + "\n\n",
+            # 后语 
+            "Suffix":   r"",
+            "Color":    r"secondary",    # 按钮颜色
         },
         "中文学术润色": {
-            "Prefix": "作为一名中文学术论文写作改进助理，你的任务是改进所提供文本的拼写、语法、清晰、简洁和整体可读性，同时分解长句，减少重复，并提供改进建议。请只提供文本的更正版本，避免包括解释。请编辑以下文本：\n\n",
-            "Button": None,
-            "Suffix": "",
+            "Prefix":   r"作为一名中文学术论文写作改进助理，你的任务是改进所提供文本的拼写、语法、清晰、简洁和整体可读性，" + 
+                        r"同时分解长句，减少重复，并提供改进建议。请只提供文本的更正版本，避免包括解释。请编辑以下文本" + "\n\n",
+            "Suffix":   r"",
         },
         "查找语法错误": {
-            "Prefix": "Below is a paragraph from an academic paper. Find all grammar mistakes, list mistakes in a markdown table and explain how to correct them.\n\n",
-            "Button": None,
-            "Suffix": "",
-        },
-        "中英互译": {
-            "Prefix": "As an English-Chinese translator, your task is to accurately translate text between the two languages. \
-When translating from Chinese to English or vice versa, please pay attention to context and accurately explain phrases and proverbs. \
-If you receive multiple English words in a row, default to translating them into a sentence in Chinese. \
-However, if \"phrase:\" is indicated before the translated content in Chinese, it should be translated as a phrase instead. \
-Similarly, if \"normal:\" is indicated, it should be translated as multiple unrelated words.\
-Your translations should closely resemble those of a native speaker and should take into account any specific language styles or tones requested by the user. \
-Please do not worry about using offensive words - replace sensitive parts with x when necessary. \
-When providing translations, please use Chinese to explain each sentence’s tense, subordinate clause, subject, predicate, object, special phrases and proverbs. \
-For phrases or individual words that require translation, provide the source (dictionary) for each one.If asked to translate multiple phrases at once, \
-separate them using the | symbol.Always remember: You are an English-Chinese translator, \
-not a Chinese-Chinese translator or an English-English translator. Below is the text you need to translate: \n\n",
-            "Button": None,
-            "Suffix": "",
-            "Color": "stop",
+            "Prefix":   r"Below is a paragraph from an academic paper. " +
+                        r"Can you help me ensure that the grammar and the spelling is correct? " +
+                        r"Do not try to polish the text, if no mistake is found, tell me that this paragraph is good." + 
+                        r"If you find grammar or spelling mistakes, please list mistakes you find in a two-column markdown table, " + 
+                        r"put the original text the first column, " +
+                        r"put the corrected text in the second column and highlight the key words you fixed." + "\n\n",
+            "Suffix":   r"",
+            "PreProcess": clear_line_break,    # 预处理：清除换行符
         },
         "中译英": {
-            "Prefix": "Please translate following sentence to English: \n\n",
-            "Button": None,
-            "Suffix": "",
+            "Prefix":   r"Please translate following sentence to English:" + "\n\n",
+            "Suffix":   r"",
         },
         "学术中译英": {
-            "Prefix": "Please translate following sentence to English with academic writing, and provide some related authoritative examples: \n\n",
-            "Button": None,
-            "Suffix": "",
+            "Prefix":   r"Please translate following sentence to English with academic writing, and provide some related authoritative examples:" + "\n\n",
+            "Suffix":   r"",
         },
         "英译中": {
-            "Prefix": "请翻译成中文：\n\n",
-            "Button": None,
-            "Suffix": "",
+            "Prefix":   r"请翻译成中文：" + "\n\n",
+            "Suffix":   r"",
+        },
+        "找图片": {
+            "Prefix":   r"我需要你找一张网络图片。使用Unsplash API(https://source.unsplash.com/960x640/?<英语关键词>)获取图片URL，" +
+                        r"然后请使用Markdown格式封装，并且不要有反斜线，不要用代码块。现在，请按以下描述给我发送图片：" + "\n\n",
+            "Suffix":   r"",
         },
         "解释代码": {
-            "Prefix": "请解释以下代码：\n```\n",
-            "Button": None,
-            "Suffix": "\n```\n",
-            "Color": "stop",
+            "Prefix":   r"请解释以下代码：" + "\n```\n",
+            "Suffix":   "\n```\n",
         },
     }
-
-

functional_crazy.py

@@ -0,0 +1,73 @@
+from functools import HotReload # HotReload 的意思是热更新，修改函数插件后，不需要重启程序，代码直接生效
+
+# UserVisibleLevel是过滤器参数。
+# 由于UI界面空间有限，所以通过这种方式决定UI界面中显示哪些插件
+# 默认函数插件 VisibleLevel 是 0
+# 当 UserVisibleLevel >= 函数插件的 VisibleLevel 时，该函数插件才会被显示出来
+UserVisibleLevel = 1
+
+
+def get_crazy_functionals():
+    from crazy_functions.读文章写摘要 import 读文章写摘要
+    from crazy_functions.生成函数注释 import 批量生成函数注释
+    from crazy_functions.解析项目源代码 import 解析项目本身
+    from crazy_functions.解析项目源代码 import 解析一个Python项目
+    from crazy_functions.解析项目源代码 import 解析一个C项目的头文件
+    from crazy_functions.解析项目源代码 import 解析一个C项目
+    from crazy_functions.高级功能函数模板 import 高阶功能模板函数
+    from crazy_functions.代码重写为全英文_多线程 import 全项目切换英文
+
+    function_plugins = {
+        "请解析并解构此项目本身": {
+            # HotReload 的意思是热更新，修改函数插件后，不需要重启程序，代码直接生效
+            "Function": 解析项目本身
+        },
+        "解析整个py项目": {
+            "Color": "stop",    # 按钮颜色
+            "Function": 解析一个Python项目
+        },
+        "解析整个C++项目头文件": {
+            "Color": "stop",    # 按钮颜色
+            "Function": 解析一个C项目的头文件
+        },
+        "解析整个C++项目": {
+            "Color": "stop",    # 按钮颜色
+            "Function": 解析一个C项目
+        },
+        "读tex论文写摘要": {
+            "Color": "stop",    # 按钮颜色
+            "Function": 读文章写摘要
+        },
+        "批量生成函数注释": {
+            "Color": "stop",    # 按钮颜色
+            "Function": 批量生成函数注释
+        },
+        "[多线程demo] 把本项目源代码切换成全英文": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Function": HotReload(全项目切换英文)
+        },
+        "[函数插件模板demo] 历史上的今天": {
+            # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+            "Function": HotReload(高阶功能模板函数)
+        },
+    }
+
+    # VisibleLevel=1 经过测试，但功能未达到理想状态
+    if UserVisibleLevel >= 1:
+        from crazy_functions.批量总结PDF文档 import 批量总结PDF文档
+        function_plugins.update({
+            "[仅供开发调试] 批量总结PDF文档": {
+                "Color": "stop",
+                # HotReload 的意思是热更新，修改函数插件代码后，不需要重启程序，代码直接生效
+                "Function": HotReload(批量总结PDF文档)
+            },
+        })
+
+    # VisibleLevel=2 尚未充分测试的函数插件，放在这里
+    if UserVisibleLevel >= 2:
+        function_plugins.update({
+        })
+
+    return function_plugins
+
+

main.py

@@ -1,113 +1,113 @@
-import gradio as gr 
-import os
-import markdown, mdtex2html
+import os; os.environ['no_proxy'] = '*' # 避免代理网络产生意外污染
+import gradio as gr
 from predict import predict
-from show_math import convert as convert_math
+from toolbox import format_io, find_free_port, on_file_uploaded, on_report_generated, get_conf
 
-def find_free_port():
-    import socket
-    from contextlib import closing
-    with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as s:
-        s.bind(('', 0))
-        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
-        return s.getsockname()[1]
-    
-PORT = find_free_port()
+# 建议您复制一个config_private.py放自己的秘密, 如API和代理网址, 避免不小心传github被别人看到
+proxies, WEB_PORT, LLM_MODEL, CONCURRENT_COUNT, AUTHENTICATION = \
+    get_conf('proxies', 'WEB_PORT', 'LLM_MODEL', 'CONCURRENT_COUNT', 'AUTHENTICATION')
+
+
+# 如果WEB_PORT是-1, 则随机选取WEB端口
+PORT = find_free_port() if WEB_PORT <= 0 else WEB_PORT
+AUTHENTICATION = None if AUTHENTICATION == [] else AUTHENTICATION
 
 initial_prompt = "Serve me as a writing and programming assistant."
 title_html = """<h1 align="center">ChatGPT 学术优化</h1>"""
 
+# 问询记录, python 版本建议3.9+（越新越好）
 import logging
 os.makedirs('gpt_log', exist_ok=True)
-logging.basicConfig(filename='gpt_log/predict.log', level=logging.INFO)
-
+try:logging.basicConfig(filename='gpt_log/chat_secrets.log', level=logging.INFO, encoding='utf-8')
+except:logging.basicConfig(filename='gpt_log/chat_secrets.log', level=logging.INFO)
+print('所有问询记录将自动保存在本地目录./gpt_log/chat_secrets.log, 请注意自我隐私保护哦！')
 
+# 一些普通功能模块
 from functional import get_functionals
 functional = get_functionals()
-def reset_textbox(): return gr.update(value='')
 
-def text_divide_paragraph(text):
-    if '```' in text:
-        # careful input
-        return text
-    else:
-        # wtf input
-        lines = text.split("\n")
-        for i, line in enumerate(lines):
-            if i!=0: lines[i] = "<p>"+lines[i].replace(" ", "&nbsp;")+"</p>"
-        text = "".join(lines)
-        return text
+# 对一些丧心病狂的实验性功能模块进行测试
+from functional_crazy import get_crazy_functionals
+crazy_functional = get_crazy_functionals()
 
-def markdown_convertion(txt):
-    if ('$' in txt) and ('```' not in txt):
-        math_config = {'mdx_math': {'enable_dollar_delimiter': True}}
-        return markdown.markdown(txt,extensions=['fenced_code','tables']) + '<br><br>' + \
-            markdown.markdown(convert_math(txt, splitParagraphs=False),extensions=['fenced_code','tables'])
-    else:
-        return markdown.markdown(txt,extensions=['fenced_code','tables'])
-
-# math_config = {'mdx_math': {'enable_dollar_delimiter': True}}
-# markdown.markdown(txt, extensions=['fenced_code', 'tables', 'mdx_math'], extension_configs=math_config)
-
-
-def format_io(self,y):
-    if y is None:
-        return []
-    i_ask, gpt_reply = y[-1]
-    
-    i_ask = text_divide_paragraph(i_ask) # 输入部分太自由，预处理一波
-    
-    y[-1] = (
-        None if i_ask is None else markdown.markdown(i_ask, extensions=['fenced_code','tables']),
-        None if gpt_reply is None else markdown_convertion(gpt_reply)
-    )
-    return y
+# 处理markdown文本格式的转变
 gr.Chatbot.postprocess = format_io
 
-with gr.Blocks() as demo:
+# 做一些外观色彩上的调整
+from theme import adjust_theme
+set_theme = adjust_theme()
+
+cancel_handles = []
+with gr.Blocks(theme=set_theme, analytics_enabled=False) as demo:
     gr.HTML(title_html)
     with gr.Row():
         with gr.Column(scale=2):
             chatbot = gr.Chatbot()
-            chatbot.style(height=700)
+            chatbot.style(height=1150)
             chatbot.style()
             history = gr.State([])
-            TRUE = gr.State(True)
-            FALSE = gr.State(False)
         with gr.Column(scale=1):
             with gr.Row():
-                with gr.Column(scale=12):
-                    txt = gr.Textbox(show_label=False, placeholder="Input question here.").style(container=False)
-                with gr.Column(scale=1):
-                    submitBtn = gr.Button("Ask", variant="primary")
+                txt = gr.Textbox(show_label=False, placeholder="Input question here.").style(container=False)
+            with gr.Row():
+                submitBtn = gr.Button("提交", variant="primary")
+            with gr.Row():
+                resetBtn = gr.Button("重置", variant="secondary"); resetBtn.style(size="sm")
+                stopBtn = gr.Button("停止", variant="secondary"); stopBtn.style(size="sm")
+            with gr.Row():
+                from check_proxy import check_proxy
+                statusDisplay = gr.Markdown(f"Tip: 按Enter提交, 按Shift+Enter换行。当前模型: {LLM_MODEL} \n {check_proxy(proxies)}")
             with gr.Row():
                 for k in functional:
                     variant = functional[k]["Color"] if "Color" in functional[k] else "secondary"
                     functional[k]["Button"] = gr.Button(k, variant=variant)
-
-            statusDisplay = gr.Markdown("status: ready")
-            systemPromptTxt = gr.Textbox(show_label=True, placeholder=f"System Prompt", label="System prompt", value=initial_prompt).style(container=True)
-            #inputs, top_p, temperature, top_k, repetition_penalty
+            with gr.Row():
+                gr.Markdown("注意：以下“红颜色”标识的函数插件需从input区读取路径作为参数.")
+            with gr.Row():
+                for k in crazy_functional:
+                    variant = crazy_functional[k]["Color"] if "Color" in crazy_functional[k] else "secondary"
+                    crazy_functional[k]["Button"] = gr.Button(k, variant=variant)
+            with gr.Row():
+                gr.Markdown("上传本地文件，供上面的函数插件调用.")
+            with gr.Row():
+                file_upload = gr.Files(label='任何文件, 但推荐上传压缩文件(zip, tar)', file_count="multiple")
+            system_prompt = gr.Textbox(show_label=True, placeholder=f"System Prompt", label="System prompt", value=initial_prompt).style(container=True)
             with gr.Accordion("arguments", open=False):
                 top_p = gr.Slider(minimum=-0, maximum=1.0, value=1.0, step=0.01,interactive=True, label="Top-p (nucleus sampling)",)
-                temperature = gr.Slider(minimum=-0, maximum=5.0, value=1.0, step=0.01, interactive=True, label="Temperature",)
+                temperature = gr.Slider(minimum=-0, maximum=2.0, value=1.0, step=0.01, interactive=True, label="Temperature",)
 
-    txt.submit(predict, [txt, top_p, temperature, chatbot, history, systemPromptTxt], [chatbot, history, statusDisplay])
-    submitBtn.click(predict, [txt, top_p, temperature, chatbot, history, systemPromptTxt], [chatbot, history, statusDisplay], show_progress=True)
-    # submitBtn.click(reset_textbox, [], [txt])
+    predict_args = dict(fn=predict, inputs=[txt, top_p, temperature, chatbot, history, system_prompt], outputs=[chatbot, history, statusDisplay], show_progress=True)
+    empty_txt_args = dict(fn=lambda: "", inputs=[], outputs=[txt]) # 用于在提交后清空输入栏
+
+    cancel_handles.append(txt.submit(**predict_args))
+    # txt.submit(**empty_txt_args) 在提交后清空输入栏
+    cancel_handles.append(submitBtn.click(**predict_args))
+    # submitBtn.click(**empty_txt_args) 在提交后清空输入栏
+    resetBtn.click(lambda: ([], [], "已重置"), None, [chatbot, history, statusDisplay])
     for k in functional:
-        functional[k]["Button"].click(predict, 
-            [txt, top_p, temperature, chatbot,history, systemPromptTxt, FALSE, TRUE, gr.State(k)], [chatbot, history, statusDisplay], show_progress=True)
-
-print(f"URL http://localhost:{PORT}")
-demo.title = "ChatGPT 学术优化"
+        click_handle = functional[k]["Button"].click(predict,
+            [txt, top_p, temperature, chatbot, history, system_prompt, gr.State(True), gr.State(k)], [chatbot, history, statusDisplay], show_progress=True)
+        cancel_handles.append(click_handle)
+    file_upload.upload(on_file_uploaded, [file_upload, chatbot, txt], [chatbot, txt])
+    for k in crazy_functional:
+        click_handle = crazy_functional[k]["Button"].click(crazy_functional[k]["Function"],
+            [txt, top_p, temperature, chatbot, history, system_prompt, gr.State(PORT)], [chatbot, history, statusDisplay]
+        )
+        try: click_handle.then(on_report_generated, [file_upload, chatbot], [file_upload, chatbot])
+        except: pass
+        cancel_handles.append(click_handle)
+    stopBtn.click(fn=None, inputs=None, outputs=None, cancels=cancel_handles)
 
+# gradio的inbrowser触发不太稳定，回滚代码到原始的浏览器打开函数
 def auto_opentab_delay():
     import threading, webbrowser, time
-    def open(): time.sleep(2)
-    webbrowser.open_new_tab(f'http://localhost:{PORT}')
+    print(f"URL http://localhost:{PORT}")
+    def open(): 
+        time.sleep(2)
+        webbrowser.open_new_tab(f'http://localhost:{PORT}')
     t = threading.Thread(target=open)
     t.daemon = True; t.start()
 
 auto_opentab_delay()
-demo.queue().launch(server_name="0.0.0.0", share=True, server_port=PORT)
+demo.title = "ChatGPT 学术优化"
+demo.queue(concurrency_count=CONCURRENT_COUNT).launch(server_name="0.0.0.0", share=True, server_port=PORT, auth=AUTHENTICATION)

predict.py

@@ -1,78 +1,230 @@
+# 借鉴了 https://github.com/GaiZhenbiao/ChuanhuChatGPT 项目
+
+"""
+    该文件中主要包含三个函数
+
+    不具备多线程能力的函数：
+    1. predict: 正常对话时使用，具备完备的交互功能，不可多线程
+
+    具备多线程调用能力的函数
+    2. predict_no_ui：高级实验性功能模块调用，不会实时显示在界面上，参数简单，可以多线程并行，方便实现复杂的功能逻辑
+    3. predict_no_ui_long_connection：在实验过程中发现调用predict_no_ui处理长文档时，和openai的连接容易断掉，这个函数用stream的方式解决这个问题，同样支持多线程
+"""
+
 import json
 import gradio as gr
 import logging
 import traceback
 import requests
 import importlib
-import os
 
-if os.path.exists('config_private.py'):
-    # 放自己的秘密如API和代理网址
-    from config_private import proxies, API_URL, API_KEY
-else:
-    from config import proxies, API_URL, API_KEY
+# config_private.py放自己的秘密如API和代理网址
+# 读取时首先看是否存在私密的config_private配置文件（不受git管控），如果有，则覆盖原config文件
+from toolbox import get_conf
+proxies, API_URL, API_KEY, TIMEOUT_SECONDS, MAX_RETRY, LLM_MODEL = \
+    get_conf('proxies', 'API_URL', 'API_KEY', 'TIMEOUT_SECONDS', 'MAX_RETRY', 'LLM_MODEL')
+
+timeout_bot_msg = '[Local Message] Request timeout. Network error. Please check proxy settings in config.py.' + \
+                  '网络错误，检查代理服务器是否可用，以及代理设置的格式是否正确，格式须是[协议]://[地址]:[端口]，缺一不可。'
+
+def get_full_error(chunk, stream_response):
+    """
+        获取完整的从Openai返回的报错
+    """
+    while True:
+        try:
+            chunk += next(stream_response)
+        except:
+            break
+    return chunk
+
+def predict_no_ui(inputs, top_p, temperature, history=[], sys_prompt=""):
+    """
+        发送至chatGPT，等待回复，一次性完成，不显示中间过程。
+        predict函数的简化版。
+        用于payload比较大的情况，或者用于实现多线、带嵌套的复杂功能。
+
+        inputs 是本次问询的输入
+        top_p, temperature是chatGPT的内部调优参数
+        history 是之前的对话列表
+        （注意无论是inputs还是history，内容太长了都会触发token数量溢出的错误，然后raise ConnectionAbortedError）
+    """
+    headers, payload = generate_payload(inputs, top_p, temperature, history, system_prompt=sys_prompt, stream=False)
+
+    retry = 0
+    while True:
+        try:
+            # make a POST request to the API endpoint, stream=False
+            response = requests.post(API_URL, headers=headers, proxies=proxies,
+                                    json=payload, stream=False, timeout=TIMEOUT_SECONDS*2); break
+        except requests.exceptions.ReadTimeout as e:
+            retry += 1
+            traceback.print_exc()
+            if retry > MAX_RETRY: raise TimeoutError
+            if MAX_RETRY!=0: print(f'请求超时，正在重试 ({retry}/{MAX_RETRY}) ……')
+
+    try:
+        result = json.loads(response.text)["choices"][0]["message"]["content"]
+        return result
+    except Exception as e:
+        if "choices" not in response.text: print(response.text)
+        raise ConnectionAbortedError("Json解析不合常规，可能是文本过长" + response.text)
 
 
+def predict_no_ui_long_connection(inputs, top_p, temperature, history=[], sys_prompt=""):
+    """
+        发送至chatGPT，等待回复，一次性完成，不显示中间过程。但内部用stream的方法避免有人中途掐网线。
+    """
+    headers, payload = generate_payload(inputs, top_p, temperature, history, system_prompt=sys_prompt, stream=True)
 
-def compose_system(system_prompt):
-    return {"role": "system", "content": system_prompt}
+    retry = 0
+    while True:
+        try:
+            # make a POST request to the API endpoint, stream=False
+            response = requests.post(API_URL, headers=headers, proxies=proxies,
+                                    json=payload, stream=True, timeout=TIMEOUT_SECONDS); break
+        except requests.exceptions.ReadTimeout as e:
+            retry += 1
+            traceback.print_exc()
+            if retry > MAX_RETRY: raise TimeoutError
+            if MAX_RETRY!=0: print(f'请求超时，正在重试 ({retry}/{MAX_RETRY}) ……')
+
+    stream_response =  response.iter_lines()
+    result = ''
+    while True:
+        try: chunk = next(stream_response).decode()
+        except StopIteration: break
+        if len(chunk)==0: continue
+        if not chunk.startswith('data:'): 
+            chunk = get_full_error(chunk.encode('utf8'), stream_response)
+            raise ConnectionAbortedError("OpenAI拒绝了请求:" + chunk.decode())
+        delta = json.loads(chunk.lstrip('data:'))['choices'][0]["delta"]
+        if len(delta) == 0: break
+        if "role" in delta: continue
+        if "content" in delta: result += delta["content"]; print(delta["content"], end='')
+        else: raise RuntimeError("意外Json结构："+delta)
+    return result
 
 
-def compose_user(user_input):
-    return {"role": "user", "content": user_input}
-
-
-def predict(inputs, top_p, temperature, chatbot=[], history=[], system_prompt='', retry=False, 
+def predict(inputs, top_p, temperature, chatbot=[], history=[], system_prompt='', 
             stream = True, additional_fn=None):
-
+    """
+        发送至chatGPT，流式获取输出。
+        用于基础的对话功能。
+        inputs 是本次问询的输入
+        top_p, temperature是chatGPT的内部调优参数
+        history 是之前的对话列表（注意无论是inputs还是history，内容太长了都会触发token数量溢出的错误）
+        chatbot 为WebUI中显示的对话列表，修改它，然后yeild出去，可以直接修改对话界面内容
+        additional_fn代表点击的哪个按钮，按钮见functional.py
+    """
     if additional_fn is not None:
         import functional
-        importlib.reload(functional)
+        importlib.reload(functional)    # 热更新prompt
         functional = functional.get_functionals()
+        if "PreProcess" in functional[additional_fn]: inputs = functional[additional_fn]["PreProcess"](inputs)  # 获取预处理函数（如果有的话）
         inputs = functional[additional_fn]["Prefix"] + inputs + functional[additional_fn]["Suffix"]
 
     if stream:
         raw_input = inputs
         logging.info(f'[raw_input] {raw_input}')
         chatbot.append((inputs, ""))
-        yield chatbot, history, "Waiting"
+        yield chatbot, history, "等待响应"
 
+    headers, payload = generate_payload(inputs, top_p, temperature, history, system_prompt, stream)
+    history.append(inputs); history.append(" ")
+
+    retry = 0
+    while True:
+        try:
+            # make a POST request to the API endpoint, stream=True
+            response = requests.post(API_URL, headers=headers, proxies=proxies,
+                                    json=payload, stream=True, timeout=TIMEOUT_SECONDS);break
+        except:
+            retry += 1
+            chatbot[-1] = ((chatbot[-1][0], timeout_bot_msg))
+            retry_msg = f"，正在重试 ({retry}/{MAX_RETRY}) ……" if MAX_RETRY > 0 else ""
+            yield chatbot, history, "请求超时"+retry_msg
+            if retry > MAX_RETRY: raise TimeoutError
+
+    gpt_replying_buffer = ""
+    
+    is_head_of_the_stream = True
+    if stream:
+        stream_response =  response.iter_lines()
+        while True:
+            chunk = next(stream_response)
+            # print(chunk.decode()[6:])
+            if is_head_of_the_stream:
+                # 数据流的第一帧不携带content
+                is_head_of_the_stream = False; continue
+            
+            if chunk:
+                try:
+                    if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
+                        # 判定为数据流的结束，gpt_replying_buffer也写完了
+                        logging.info(f'[response] {gpt_replying_buffer}')
+                        break
+                    # 处理数据流的主体
+                    chunkjson = json.loads(chunk.decode()[6:])
+                    status_text = f"finish_reason: {chunkjson['choices'][0]['finish_reason']}"
+                    # 如果这里抛出异常，一般是文本过长，详情见get_full_error的输出
+                    gpt_replying_buffer = gpt_replying_buffer + json.loads(chunk.decode()[6:])['choices'][0]["delta"]["content"]
+                    history[-1] = gpt_replying_buffer
+                    chatbot[-1] = (history[-2], history[-1])
+                    yield chatbot, history, status_text
+
+                except Exception as e:
+                    traceback.print_exc()
+                    yield chatbot, history, "Json解析不合常规"
+                    chunk = get_full_error(chunk, stream_response)
+                    error_msg = chunk.decode()
+                    if "reduce the length" in error_msg:
+                        chatbot[-1] = (chatbot[-1][0], "[Local Message] Input (or history) is too long, please reduce input or clear history by refreshing this page.")
+                        history = []
+                    elif "Incorrect API key" in error_msg:
+                        chatbot[-1] = (chatbot[-1][0], "[Local Message] Incorrect API key provided.")
+                    else:
+                        from toolbox import regular_txt_to_markdown
+                        tb_str = regular_txt_to_markdown(traceback.format_exc())
+                        chatbot[-1] = (chatbot[-1][0], f"[Local Message] Json Error \n\n {tb_str} \n\n {regular_txt_to_markdown(chunk.decode()[4:])}")
+                    yield chatbot, history, "Json解析不合常规" + error_msg
+                    return
+
+def generate_payload(inputs, top_p, temperature, history, system_prompt, stream):
+    """
+        整合所有信息，选择LLM模型，生成http请求，为发送请求做准备
+    """
     headers = {
         "Content-Type": "application/json",
         "Authorization": f"Bearer {API_KEY}"
     }
 
-    chat_counter = len(history) // 2
+    conversation_cnt = len(history) // 2
 
-    print(f"chat_counter - {chat_counter}")
-
-    messages = [compose_system(system_prompt)]
-    if chat_counter:
-        for index in range(0, 2*chat_counter, 2):
-            d1 = {}
-            d1["role"] = "user"
-            d1["content"] = history[index]
-            d2 = {}
-            d2["role"] = "assistant"
-            d2["content"] = history[index+1]
-            if d1["content"] != "":
-                if d2["content"] != "" or retry:
-                    messages.append(d1)
-                    messages.append(d2)
+    messages = [{"role": "system", "content": system_prompt}]
+    if conversation_cnt:
+        for index in range(0, 2*conversation_cnt, 2):
+            what_i_have_asked = {}
+            what_i_have_asked["role"] = "user"
+            what_i_have_asked["content"] = history[index]
+            what_gpt_answer = {}
+            what_gpt_answer["role"] = "assistant"
+            what_gpt_answer["content"] = history[index+1]
+            if what_i_have_asked["content"] != "":
+                if what_gpt_answer["content"] == "": continue
+                if what_gpt_answer["content"] == timeout_bot_msg: continue
+                messages.append(what_i_have_asked)
+                messages.append(what_gpt_answer)
             else:
-                messages[-1]['content'] = d2['content']
-    if retry and chat_counter:
-        messages.pop()
-    else:
-        temp3 = {}
-        temp3["role"] = "user"
-        temp3["content"] = inputs
-        messages.append(temp3)
-        chat_counter += 1
-    # messages
+                messages[-1]['content'] = what_gpt_answer['content']
+
+    what_i_ask_now = {}
+    what_i_ask_now["role"] = "user"
+    what_i_ask_now["content"] = inputs
+    messages.append(what_i_ask_now)
+
     payload = {
-        "model": "gpt-3.5-turbo",
-        # "model": "gpt-4",
+        "model": LLM_MODEL,
         "messages": messages, 
         "temperature": temperature,  # 1.0,
         "top_p": top_p,  # 1.0,
@@ -81,54 +233,8 @@ def predict(inputs, top_p, temperature, chatbot=[], history=[], system_prompt=''
         "presence_penalty": 0,
         "frequency_penalty": 0,
     }
+    
+    print(f" {LLM_MODEL} : {conversation_cnt} : {inputs}")
+    return headers,payload
 
-    history.append(inputs)
 
-    try:
-        # make a POST request to the API endpoint using the requests.post method, passing in stream=True
-        response = requests.post(API_URL, headers=headers, proxies=proxies,
-                                json=payload, stream=True, timeout=15)
-    except:
-        chatbot[-1] = ((chatbot[-1][0], 'Requests Timeout, Network Error.'))
-        yield chatbot, history, "Requests Timeout"
-        raise TimeoutError
-
-    token_counter = 0
-    partial_words = ""
-
-    counter = 0
-    if stream:
-        stream_response =  response.iter_lines()
-        while True:
-            chunk = next(stream_response)
-            # print(chunk)
-            
-            if chunk == b'data: [DONE]':
-                break
-
-            if counter == 0:
-                counter += 1
-                continue
-            counter += 1
-            # check whether each line is non-empty
-            if chunk:
-                # decode each line as response data is in bytes
-                try:
-                    if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
-                        logging.info(f'[response] {chatbot[-1][-1]}')
-                        break
-                except Exception as e:
-                    traceback.print_exc()
-
-                chunkjson = json.loads(chunk.decode()[6:])
-                status_text = f"id: {chunkjson['id']}, finish_reason: {chunkjson['choices'][0]['finish_reason']}"
-                partial_words = partial_words + \
-                    json.loads(chunk.decode()[6:])[
-                        'choices'][0]["delta"]["content"]
-                if token_counter == 0:
-                    history.append(" " + partial_words)
-                else:
-                    history[-1] = partial_words
-                chatbot[-1] = (history[-2], history[-1])
-                token_counter += 1
-                yield chatbot, history, status_text

project_self_analysis.md

@@ -0,0 +1,122 @@
+# chatgpt-academic项目自译解报告
+（Author补充：以下分析均由本项目调用ChatGPT一键生成，如果有不准确的地方，全怪GPT😄）
+
+## [0/10] 程序摘要: check_proxy.py
+
+这个程序是一个用来检查代理服务器是否有效的 Python 程序代码。程序文件名为 check_proxy.py。其中定义了一个函数 check_proxy，该函数接收一个代理配置信息 proxies，使用 requests 库向一个代理服务器发送请求，获取该代理的所在地信息并返回。如果请求超时或者异常，该函数将返回一个代理无效的结果。
+
+程序代码分为两个部分，首先是 check_proxy 函数的定义部分，其次是程序文件的入口部分，在该部分代码中，程序从 config_private.py 文件或者 config.py 文件中加载代理配置信息，然后调用 check_proxy 函数来检测代理服务器是否有效。如果配置文件 config_private.py 存在，则会加载其中的代理配置信息，否则会从 config.py 文件中读取。
+
+## [1/10] 程序摘要: config.py
+
+本程序文件名为config.py，主要功能是存储应用所需的常量和配置信息。
+
+其中，包含了应用所需的OpenAI API密钥、API接口地址、网络代理设置、超时设置、网络端口和OpenAI模型选择等信息，在运行应用前需要进行相应的配置。在未配置网络代理时，程序给出了相应的警告提示。
+
+此外，还包含了一个检查函数，用于检查是否忘记修改API密钥。
+
+总之，config.py文件是应用中的一个重要配置文件，用来存储应用所需的常量和配置信息，需要在应用运行前进行相应的配置。
+
+## [2/10] 程序摘要: config_private.py
+
+该文件是一个配置文件，命名为config_private.py。它是一个Python脚本，用于配置OpenAI的API密钥、模型和其它相关设置。该配置文件还可以设置是否使用代理。如果使用代理，需要设置代理协议、地址和端口。在设置代理之后，该文件还包括一些用于测试代理是否正常工作的代码。该文件还包括超时时间、随机端口、重试次数等设置。在文件末尾，还有一个检查代码，如果没有更改API密钥，则抛出异常。
+
+## [3/10] 程序摘要: functional.py
+
+该程序文件名为 functional.py，其中包含一个名为 get_functionals 的函数，该函数返回一个字典，该字典包含了各种翻译、校对等功能的名称、前缀、后缀以及默认按钮颜色等信息。具体功能包括：英语学术润色、中文学术润色、查找语法错误、中英互译、中译英、学术中译英、英译中、解释代码等。该程序的作用为提供各种翻译、校对等功能的模板，以便后续程序可以直接调用。
+
+（Author补充：这个文件汇总了模块化的Prompt调用，如果发现了新的好用Prompt，别藏着哦^_^速速PR）
+
+
+## [4/10] 程序摘要: functional_crazy.py
+
+这个程序文件 functional_crazy.py 导入了一些 python 模块，并提供了一个函数 get_crazy_functionals()，该函数返回不同实验功能的描述和函数。其中，使用的的模块包括：
+
+- crazy_functions.读文章写摘要 中的 读文章写摘要
+- crazy_functions.生成函数注释 中的 批量生成函数注释
+- crazy_functions.解析项目源代码 中的 解析项目本身、解析一个Python项目、解析一个C项目的头文件、解析一个C项目
+- crazy_functions.高级功能函数模板 中的 高阶功能模板函数
+
+返回的实验功能函数包括：
+
+- "[实验] 请解析并解构此项目本身"，包含函数：解析项目本身
+- "[实验] 解析整个py项目（配合input输入框）"，包含函数：解析一个Python项目
+- "[实验] 解析整个C++项目头文件（配合input输入框）"，包含函数：解析一个C项目的头文件
+- "[实验] 解析整个C++项目（配合input输入框）"，包含函数：解析一个C项目
+- "[实验] 读tex论文写摘要（配合input输入框）"，包含函数：读文章写摘要
+- "[实验] 批量生成函数注释（配合input输入框）"，包含函数：批量生成函数注释
+- "[实验] 实验功能函数模板"，包含函数：高阶功能模板函数
+
+这些函数用于系统开发和测试，方便开发者进行特定程序语言后台功能开发的测试和实验，增加系统可靠稳定性和用户友好性。
+
+（Author补充：这个文件汇总了模块化的函数，如此设计以方便任何新功能的加入）
+
+## [5/10] 程序摘要: main.py
+
+该程序是一个基于Gradio框架的聊天机器人应用程序。用户可以通过输入问题来获取答案，并与聊天机器人进行对话。该应用程序还集成了一些实验性功能模块，用户可以通过上传本地文件或点击相关按钮来使用这些模块。程序还可以生成对话日志，并且具有一些外观上的调整。在运行时，它会自动打开一个网页并在本地启动服务器。
+
+
+## [6/10] 程序摘要: predict.py
+
+该程序文件名为predict.py，主要是针对一个基于ChatGPT的聊天机器人进行交互和预测。
+
+第一部分是导入所需的库和配置文件。
+
+第二部分是一个用于获取Openai返回的完整错误信息的函数。
+
+第三部分是用于一次性完成向ChatGPT发送请求和等待回复的函数。
+
+第四部分是用于基础的对话功能的函数，通过stream参数可以选择是否显示中间的过程。
+
+第五部分是用于整合所需信息和选择LLM模型生成的HTTP请求。
+
+（Author补充：主要是predict_no_ui和predict两个函数。前者不用stream，方便、高效、易用。后者用stream，展现效果好。）
+
+## [7/10] 程序摘要: show_math.py
+
+这是一个名为show_math.py的Python程序文件，主要用于将Markdown-LaTeX混合文本转换为HTML格式，并包括MathML数学公式。程序使用latex2mathml.converter库将LaTeX公式转换为MathML格式，并使用正则表达式递归地翻译输入的Markdown-LaTeX混合文本。程序包括转换成双美元符号($$)形式、转换成单美元符号($)形式、转换成\[\]形式以及转换成\(\)形式的LaTeX数学公式。如果转换中出现错误，程序将返回相应的错误消息。
+
+## [8/10] 程序摘要: theme.py
+
+这是一个名为theme.py的程序文件，用于设置Gradio界面的颜色和字体主题。该文件中定义了一个名为adjust_theme()的函数，其作用是返回一个Gradio theme对象，设置了Gradio界面的颜色和字体主题。在该函数里面，使用了Graido可用的颜色列表，主要参数包括primary_hue、neutral_hue、font和font_mono等，用于设置Gradio界面的主题色调、字体等。另外，该函数还实现了一些参数的自定义，如input_background_fill_dark、button_transition、button_shadow_hover等，用于设置Gradio界面的渐变、阴影等特效。如果Gradio版本过于陈旧，该函数会抛出异常并返回None。
+
+## [9/10] 程序摘要: toolbox.py
+
+该文件为Python程序文件，文件名为toolbox.py。主要功能包括：
+
+1. 导入markdown、mdtex2html、threading、functools等模块。
+2. 定义函数predict_no_ui_but_counting_down，用于生成对话。
+3. 定义函数write_results_to_file，用于将对话记录生成Markdown文件。
+4. 定义函数regular_txt_to_markdown，将普通文本转换为Markdown格式的文本。
+5. 定义装饰器函数CatchException，用于捕获函数执行异常并返回生成器。
+6. 定义函数report_execption，用于向chatbot中添加错误信息。
+7. 定义函数text_divide_paragraph，用于将文本按照段落分隔符分割开，生成带有段落标签的HTML代码。
+8. 定义函数markdown_convertion，用于将Markdown格式的文本转换为HTML格式。
+9. 定义函数format_io，用于将输入和输出解析为HTML格式。
+10. 定义函数find_free_port，用于返回当前系统中可用的未使用端口。
+11. 定义函数extract_archive，用于解压归档文件。
+12. 定义函数find_recent_files，用于查找最近创建的文件。
+13. 定义函数on_file_uploaded，用于处理上传文件的操作。
+14. 定义函数on_report_generated，用于处理生成报告文件的操作。
+
+## 程序的整体功能和构架做出概括。然后用一张markdown表格整理每个文件的功能。
+
+这是一个基于Gradio框架的聊天机器人应用，支持通过文本聊天来获取答案，并可以使用一系列实验性功能模块，例如生成函数注释、解析项目源代码、读取Latex论文写摘要等。 程序架构分为前端和后端两个部分。前端使用Gradio实现，包括用户输入区域、应答区域、按钮、调用方式等。后端使用Python实现，包括聊天机器人模型、实验性功能模块、模板模块、管理模块、主程序模块等。
+
+每个程序文件的功能如下：
+
+| 文件名 | 功能描述 |
+|:----:|:----:|
+| check_proxy.py | 检查代理服务器是否有效 |
+| config.py | 存储应用所需的常量和配置信息 |
+| config_private.py | 存储Openai的API密钥、模型和其他相关设置 |
+| functional.py | 提供各种翻译、校对等实用模板 |
+| functional_crazy.py | 提供一些实验性质的高级功能 |
+| main.py | 基于Gradio框架的聊天机器人应用程序的主程序 |
+| predict.py | 用于chatbot预测方案创建，向ChatGPT发送请求和获取回复 |
+| show_math.py | 将Markdown-LaTeX混合文本转换为HTML格式，并包括MathML数学公式 |
+| theme.py | 设置Gradio界面的颜色和字体主题 |
+| toolbox.py | 定义一系列工具函数，用于对输入输出进行格式转换、文件操作、异常捕捉和处理等 |
+
+这些程序文件共同组成了一个聊天机器人应用程序的前端和后端实现，使用户可以方便地进行聊天，并可以使用相应的实验功能模块。
+

requirements.txt

@@ -1,3 +1,3 @@
-gradio
+gradio>=3.23
 requests[socks]
 mdtex2html

show_math.py

@@ -1,3 +1,5 @@
+# This program is written by: https://github.com/polarwinkel/mdtex2html
+
 from latex2mathml.converter import convert as tex2mathml
 import re
 
@@ -75,4 +77,4 @@ def convert(mdtex, extensions=[], splitParagraphs=True):
             result = convert(parts[0]+mathml+incomplete, extensions, splitParagraphs=False)
     if not found:
         result = mdtex
-    return result
+    return result

theme.py

@@ -0,0 +1,82 @@
+import gradio as gr 
+
+# gradio可用颜色列表
+# gr.themes.utils.colors.slate (石板色)
+# gr.themes.utils.colors.gray (灰色)
+# gr.themes.utils.colors.zinc (锌色)
+# gr.themes.utils.colors.neutral (中性色)
+# gr.themes.utils.colors.stone (石头色)
+# gr.themes.utils.colors.red (红色)
+# gr.themes.utils.colors.orange (橙色)
+# gr.themes.utils.colors.amber (琥珀色)
+# gr.themes.utils.colors.yellow (黄色)
+# gr.themes.utils.colors.lime (酸橙色)
+# gr.themes.utils.colors.green (绿色)
+# gr.themes.utils.colors.emerald (祖母绿)
+# gr.themes.utils.colors.teal (青蓝色)
+# gr.themes.utils.colors.cyan (青色)
+# gr.themes.utils.colors.sky (天蓝色)
+# gr.themes.utils.colors.blue (蓝色)
+# gr.themes.utils.colors.indigo (靛蓝色)
+# gr.themes.utils.colors.violet (紫罗兰色)
+# gr.themes.utils.colors.purple (紫色)
+# gr.themes.utils.colors.fuchsia (洋红色)
+# gr.themes.utils.colors.pink (粉红色)
+# gr.themes.utils.colors.rose (玫瑰色)
+
+def adjust_theme():
+    try: 
+        color_er = gr.themes.utils.colors.pink
+        set_theme = gr.themes.Default( 
+                        primary_hue=gr.themes.utils.colors.orange,
+                        neutral_hue=gr.themes.utils.colors.gray,
+                        font=["sans-serif", "Microsoft YaHei", "ui-sans-serif", "system-ui", "sans-serif", gr.themes.utils.fonts.GoogleFont("Source Sans Pro")], 
+                        font_mono=["ui-monospace", "Consolas", "monospace", gr.themes.utils.fonts.GoogleFont("IBM Plex Mono")])
+        set_theme.set(
+            # Colors
+            input_background_fill_dark="*neutral_800",
+            # Transition
+            button_transition="none",
+            # Shadows
+            button_shadow="*shadow_drop",
+            button_shadow_hover="*shadow_drop_lg",
+            button_shadow_active="*shadow_inset",
+            input_shadow="0 0 0 *shadow_spread transparent, *shadow_inset",
+            input_shadow_focus="0 0 0 *shadow_spread *secondary_50, *shadow_inset",
+            input_shadow_focus_dark="0 0 0 *shadow_spread *neutral_700, *shadow_inset",
+            checkbox_label_shadow="*shadow_drop",
+            block_shadow="*shadow_drop",
+            form_gap_width="1px",
+            # Button borders
+            input_border_width="1px",
+            input_background_fill="white",
+            # Gradients
+            stat_background_fill="linear-gradient(to right, *primary_400, *primary_200)",
+            stat_background_fill_dark="linear-gradient(to right, *primary_400, *primary_600)",
+            error_background_fill=f"linear-gradient(to right, {color_er.c100}, *background_fill_secondary)",
+            error_background_fill_dark="*background_fill_primary",
+            checkbox_label_background_fill="linear-gradient(to top, *neutral_50, white)",
+            checkbox_label_background_fill_dark="linear-gradient(to top, *neutral_900, *neutral_800)",
+            checkbox_label_background_fill_hover="linear-gradient(to top, *neutral_100, white)",
+            checkbox_label_background_fill_hover_dark="linear-gradient(to top, *neutral_900, *neutral_800)",
+            button_primary_background_fill="linear-gradient(to bottom right, *primary_100, *primary_300)",
+            button_primary_background_fill_dark="linear-gradient(to bottom right, *primary_500, *primary_600)",
+            button_primary_background_fill_hover="linear-gradient(to bottom right, *primary_100, *primary_200)",
+            button_primary_background_fill_hover_dark="linear-gradient(to bottom right, *primary_500, *primary_500)",
+            button_primary_border_color_dark="*primary_500",
+            button_secondary_background_fill="linear-gradient(to bottom right, *neutral_100, *neutral_200)",
+            button_secondary_background_fill_dark="linear-gradient(to bottom right, *neutral_600, *neutral_700)",
+            button_secondary_background_fill_hover="linear-gradient(to bottom right, *neutral_100, *neutral_100)",
+            button_secondary_background_fill_hover_dark="linear-gradient(to bottom right, *neutral_600, *neutral_600)",
+            button_cancel_background_fill=f"linear-gradient(to bottom right, {color_er.c100}, {color_er.c200})",
+            button_cancel_background_fill_dark=f"linear-gradient(to bottom right, {color_er.c600}, {color_er.c700})",
+            button_cancel_background_fill_hover=f"linear-gradient(to bottom right, {color_er.c100}, {color_er.c100})",
+            button_cancel_background_fill_hover_dark=f"linear-gradient(to bottom right, {color_er.c600}, {color_er.c600})",
+            button_cancel_border_color=color_er.c200,
+            button_cancel_border_color_dark=color_er.c600,
+            button_cancel_text_color=color_er.c600,
+            button_cancel_text_color_dark="white",
+        )
+    except: 
+        set_theme = None; print('gradio版本较旧, 不能自定义字体和颜色')
+    return set_theme

toolbox.py

@@ -0,0 +1,249 @@
+import markdown, mdtex2html, threading, importlib, traceback, importlib, inspect
+from show_math import convert as convert_math
+from functools import wraps
+
+def predict_no_ui_but_counting_down(i_say, i_say_show_user, chatbot, top_p, temperature, history=[], sys_prompt=''):
+    """
+        调用简单的predict_no_ui接口，但是依然保留了些许界面心跳功能，当对话太长时，会自动采用二分法截断
+    """
+    import time
+    from predict import predict_no_ui
+    from toolbox import get_conf
+    TIMEOUT_SECONDS, MAX_RETRY = get_conf('TIMEOUT_SECONDS', 'MAX_RETRY')
+    # 多线程的时候，需要一个mutable结构在不同线程之间传递信息
+    # list就是最简单的mutable结构，我们第一个位置放gpt输出，第二个位置传递报错信息
+    mutable = [None, '']
+    # multi-threading worker
+    def mt(i_say, history):
+        while True:
+            try:
+                mutable[0] = predict_no_ui(inputs=i_say, top_p=top_p, temperature=temperature, history=history, sys_prompt=sys_prompt)
+                break
+            except ConnectionAbortedError as e:
+                if len(history) > 0:
+                    history = [his[len(his)//2:] for his in history if his is not None]
+                    mutable[1] = 'Warning! History conversation is too long, cut into half. '
+                else:
+                    i_say = i_say[:len(i_say)//2]
+                    mutable[1] = 'Warning! Input file is too long, cut into half. '
+            except TimeoutError as e:
+                mutable[0] = '[Local Message] Failed with timeout.'
+                raise TimeoutError
+    # 创建新线程发出http请求
+    thread_name = threading.Thread(target=mt, args=(i_say, history)); thread_name.start()
+    # 原来的线程则负责持续更新UI，实现一个超时倒计时，并等待新线程的任务完成
+    cnt = 0
+    while thread_name.is_alive():
+        cnt += 1
+        chatbot[-1] = (i_say_show_user, f"[Local Message] {mutable[1]}waiting gpt response {cnt}/{TIMEOUT_SECONDS*2*(MAX_RETRY+1)}"+''.join(['.']*(cnt%4)))
+        yield chatbot, history, '正常'
+        time.sleep(1)
+    # 把gpt的输出从mutable中取出来
+    gpt_say = mutable[0]
+    if gpt_say=='[Local Message] Failed with timeout.': raise TimeoutError
+    return gpt_say
+
+def write_results_to_file(history, file_name=None):
+    """
+        将对话记录history以Markdown格式写入文件中。如果没有指定文件名，则使用当前时间生成文件名。
+    """
+    import os, time
+    if file_name is None:
+        # file_name = time.strftime("chatGPT分析报告%Y-%m-%d-%H-%M-%S", time.localtime()) + '.md'
+        file_name = 'chatGPT分析报告' + time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime()) + '.md'
+    os.makedirs('./gpt_log/', exist_ok=True)
+    with open(f'./gpt_log/{file_name}', 'w', encoding = 'utf8') as f:
+        f.write('# chatGPT 分析报告\n')
+        for i, content in enumerate(history):
+            if i%2==0: f.write('## ')
+            f.write(content)
+            f.write('\n\n')
+    res = '以上材料已经被写入' + os.path.abspath(f'./gpt_log/{file_name}')
+    print(res)
+    return res
+
+def regular_txt_to_markdown(text):
+    """
+        将普通文本转换为Markdown格式的文本。
+    """
+    text = text.replace('\n', '\n\n')
+    text = text.replace('\n\n\n', '\n\n')
+    text = text.replace('\n\n\n', '\n\n')
+    return text
+
+def CatchException(f):
+    """
+        装饰器函数，捕捉函数f中的异常并封装到一个生成器中返回，并显示到聊天当中。
+    """
+    @wraps(f)
+    def decorated(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT):
+        try:
+            yield from f(txt, top_p, temperature, chatbot, history, systemPromptTxt, WEB_PORT)
+        except Exception as e:
+            from check_proxy import check_proxy
+            from toolbox import get_conf
+            proxies, = get_conf('proxies')
+            tb_str = regular_txt_to_markdown(traceback.format_exc())
+            chatbot[-1] = (chatbot[-1][0], f"[Local Message] 实验性函数调用出错: \n\n {tb_str} \n\n 当前代理可用性: \n\n {check_proxy(proxies)}")
+            yield chatbot, history, f'异常 {e}'
+    return decorated
+
+def HotReload(f):
+    """
+        装饰器函数，实现函数插件热更新
+    """
+    @wraps(f)
+    def decorated(*args, **kwargs):
+        fn_name = f.__name__
+        f_hot_reload = getattr(importlib.reload(inspect.getmodule(f)), fn_name)
+        yield from f_hot_reload(*args, **kwargs)
+    return decorated
+
+def report_execption(chatbot, history, a, b):
+    """
+        向chatbot中添加错误信息
+    """
+    chatbot.append((a, b))
+    history.append(a); history.append(b)
+
+def text_divide_paragraph(text):
+    """
+        将文本按照段落分隔符分割开，生成带有段落标签的HTML代码。
+    """
+    if '```' in text:
+        # careful input
+        return text
+    else:
+        # wtf input
+        lines = text.split("\n")
+        for i, line in enumerate(lines):
+            lines[i] = lines[i].replace(" ", " ")
+        text = "</br>".join(lines)
+        return text
+
+def markdown_convertion(txt):
+    """
+        将Markdown格式的文本转换为HTML格式。如果包含数学公式，则先将公式转换为HTML格式。
+    """
+    if ('$' in txt) and ('```' not in txt):
+        return markdown.markdown(txt,extensions=['fenced_code','tables']) + '<br><br>' + \
+            markdown.markdown(convert_math(txt, splitParagraphs=False),extensions=['fenced_code','tables'])
+    else:
+        return markdown.markdown(txt,extensions=['fenced_code','tables'])
+
+
+def format_io(self, y):
+    """
+        将输入和输出解析为HTML格式。将y中最后一项的输入部分段落化，并将输出部分的Markdown和数学公式转换为HTML格式。
+    """
+    if y is None or y == []: return []
+    i_ask, gpt_reply = y[-1]
+    i_ask = text_divide_paragraph(i_ask) # 输入部分太自由，预处理一波
+    y[-1] = (
+        None if i_ask is None else markdown.markdown(i_ask, extensions=['fenced_code','tables']),
+        None if gpt_reply is None else markdown_convertion(gpt_reply)
+    )
+    return y
+
+
+def find_free_port():
+    """
+        返回当前系统中可用的未使用端口。
+    """
+    import socket
+    from contextlib import closing
+    with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as s:
+        s.bind(('', 0))
+        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+        return s.getsockname()[1]
+
+
+def extract_archive(file_path, dest_dir):
+    import zipfile
+    import tarfile
+    import os
+    # Get the file extension of the input file
+    file_extension = os.path.splitext(file_path)[1]
+
+    # Extract the archive based on its extension
+    if file_extension == '.zip':
+        with zipfile.ZipFile(file_path, 'r') as zipobj:
+            zipobj.extractall(path=dest_dir)
+            print("Successfully extracted zip archive to {}".format(dest_dir))
+
+    elif file_extension in ['.tar', '.gz', '.bz2']:
+        with tarfile.open(file_path, 'r:*') as tarobj:
+            tarobj.extractall(path=dest_dir)
+            print("Successfully extracted tar archive to {}".format(dest_dir))
+    else:
+        return
+
+def find_recent_files(directory):
+    """
+        me: find files that is created with in one minutes under a directory with python, write a function
+        gpt: here it is!
+    """
+    import os
+    import time
+    current_time = time.time()
+    one_minute_ago = current_time - 60
+    recent_files = []
+
+    for filename in os.listdir(directory):
+        file_path = os.path.join(directory, filename)
+        if file_path.endswith('.log'): continue
+        created_time = os.path.getctime(file_path)
+        if created_time >= one_minute_ago:
+            if os.path.isdir(file_path): continue
+            recent_files.append(file_path)
+
+    return recent_files
+
+
+def on_file_uploaded(files, chatbot, txt):
+    if len(files) == 0: return chatbot, txt
+    import shutil, os, time, glob
+    from toolbox import extract_archive
+    try: shutil.rmtree('./private_upload/')
+    except: pass
+    time_tag = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime())
+    os.makedirs(f'private_upload/{time_tag}', exist_ok=True)
+    for file in files:
+        file_origin_name = os.path.basename(file.orig_name)
+        shutil.copy(file.name, f'private_upload/{time_tag}/{file_origin_name}')
+        extract_archive(f'private_upload/{time_tag}/{file_origin_name}', 
+                        dest_dir=f'private_upload/{time_tag}/{file_origin_name}.extract')
+    moved_files = [fp for fp in glob.glob('private_upload/**/*', recursive=True)]
+    txt = f'private_upload/{time_tag}'
+    moved_files_str = '\t\n\n'.join(moved_files)
+    chatbot.append(['我上传了文件，请查收', 
+                    f'[Local Message] 收到以下文件: \n\n{moved_files_str}\n\n调用路径参数已自动修正到: \n\n{txt}\n\n现在您点击任意实验功能时，以上文件将被作为输入参数'])
+    return chatbot, txt
+
+
+def on_report_generated(files, chatbot):
+    from toolbox import find_recent_files
+    report_files = find_recent_files('gpt_log')
+    if len(report_files) == 0: return report_files, chatbot
+    # files.extend(report_files)
+    chatbot.append(['汇总报告如何远程获取？', '汇总报告已经添加到右侧文件上传区，请查收。'])
+    return report_files, chatbot
+
+def get_conf(*args):
+    # 建议您复制一个config_private.py放自己的秘密, 如API和代理网址, 避免不小心传github被别人看到
+    res = []
+    for arg in args:
+        try: r = getattr(importlib.import_module('config_private'), arg)
+        except: r = getattr(importlib.import_module('config'), arg)
+        res.append(r)
+        # 在读取API_KEY时，检查一下是不是忘了改config
+        if arg=='API_KEY' and len(r) != 51:
+            assert False, "正确的API_KEY密钥是51位，请在config文件中修改API密钥, 添加海外代理之后再运行。" + \
+                        "（如果您刚更新过代码，请确保旧版config_private文件中没有遗留任何新增键值）"
+    return res
+
+def clear_line_break(txt):
+    txt = txt.replace('\n', ' ')
+    txt = txt.replace('  ', ' ')
+    txt = txt.replace('  ', ' ')
+    return txt