use：初步能用。

.vscode/settings.json

@@ -18,6 +18,7 @@
             }
         ],
         "args": [
+            "/home/yun/Code/filecomplete", "cpp,h", "/home/yun/Code/question.txt"
         ]
     },
     "cmake.configureSettings": {
@@ -91,6 +92,7 @@
         "stdexcept": "cpp",
         "streambuf": "cpp",
         "typeinfo": "cpp",
-        "valarray": "cpp"
+        "valarray": "cpp",
+        "fstream": "cpp"
     }
 }

README.md

@@ -1,7 +1,7 @@
 # DeepSeek访问
 
-尝试实现一个可以提交大量附件，并让`DeepSeek`帮分析提取数据等的一个工具。
+尝试实现一个可以提交大量附件（按照官方要求是最多50个文件，单个不超过500MB，且仅支持文本文件），并让`DeepSeek`帮分析提取数据等的一个工具。
 
-# 说明
+# 编译
 
-如果是`windows`环境，必须使用`windows terminal`（或者其他支持读入`utf-8`字符串的终端）以支持从`cmd`命令行读入`utf-8`字符串。
+需要`cpp17`。

jsondata.cxx

@@ -1,7 +1,8 @@
 #include "jsondata.h"
 #include <iostream>
 
-CJsonOper::CJsonOper()
+CJsonOper::CJsonOper(const std::string& user_name, const std::string& model, const std::string& assistant_name)
+    : user_(user_name), model_(model), assistant_(assistant_name)
 {
 }
 
@@ -11,15 +12,53 @@ CJsonOper::~CJsonOper()
 
 std::string CJsonOper::format_request(const std::string& content)
 {
-    // 定义变量
+    std::string model = model_;
-    std::string model = "deepseek-r1";
+    std::string role = user_;
-    std::string role = "user";
-
-    // 构造 JSON 对象
     nlohmann::json json_data = {{"model", model}, {"messages", {{{"role", role}, {"content", content}}}}};
     return json_data.dump();
 }
 
+std::vector<std::string> CJsonOper::split(const std::string& input, const std::string& delimiter)
+{
+    std::vector<std::string> result;
+    size_t pos = 0, prev = 0;
+    while ((pos = input.find(delimiter, prev)) != std::string::npos) {
+        result.push_back(input.substr(prev, pos - prev));
+        prev = pos + delimiter.size();
+    }
+    result.push_back(input.substr(prev));
+    return result;
+}
+
+std::string CJsonOper::multi_format_reuqest(const std::string& content, size_t per_sec_size)
+{
+    std::string model = model_;
+    std::string role = user_;
+    nlohmann::json json_data;
+    json_data["model"] = model;
+
+    std::vector<nlohmann::json> messages;
+    size_t s = 0;
+    while (s < content.size()) {
+        size_t i = 0;
+        size_t t = 0;
+        while (i < per_sec_size && s + i < content.size()) {
+            t = get_u8_len(content[s + i]);
+            if (t == 0) {
+                std::cerr << "invalid codec!!!" << std::endl;
+                exit(1);
+            }
+            i += t;
+        }
+        std::string part = content.substr(s, i);
+        messages.push_back({{"role", role}, {"content", "\n附加数据：\n" + part}});
+        s += i;
+    }
+
+    json_data["messages"] = messages;
+    return json_data.dump();
+}
+
 Message CJsonOper::parse(const std::string& data)
 {
     Message re;
@@ -62,3 +101,85 @@ bool CJsonOper::save_md(const std::string& data, const std::string& id)
     of.close();
     return true;
 }
+
+bool CJsonOper::read_txt(const std::string& path, std::string& out)
+{
+    std::ifstream file(path);
+    if (!file.is_open()) {
+        std::cout << "open failed: " << path << std::endl;
+        return false;
+    }
+    std::istreambuf_iterator<char> iterf(file);
+    std::istreambuf_iterator<char> iter;
+    std::string content(iterf, iter);
+    out = content;
+    return true;
+}
+
+size_t CJsonOper::get_u8_len(unsigned char ch)
+{
+    if (ch <= 0x7F) {
+        return 1;
+    } else if ((ch & 0xE0) == 0xC0) {
+        return 2;
+    } else if ((ch & 0xF0) == 0xE0) {
+        return 3;
+    } else if ((ch & 0xF8) == 0xF0) {
+        return 4;
+    } else if ((ch & 0xFC) == 0xF8) {
+        return 5;
+    } else if ((ch & 0xFE) == 0xFC) {
+        return 6;
+    } else {
+        std::cerr << "invalid u8 first ch." << std::endl;
+        exit(1);
+    }
+    return 0;
+}
+
+std::string CJsonOper::trim(const std::string& input)
+{
+    size_t start = input.find_first_not_of(" \t\n\r\f\v");
+    if (start == std::string::npos) {
+        return "";
+    }
+    size_t end = input.find_last_not_of(" \t\n\r\f\v");
+    return input.substr(start, end - start + 1);
+}
+
+std::string CJsonOper::get_all_dir_content(const std::string& dir, const std::string& types)
+{
+    auto vec = split(types, ",");
+    std::vector<std::string> t;
+    for (const auto& item : vec) {
+        auto c = trim(item);
+        if (c.empty()) {
+            continue;
+        }
+        t.push_back("." + item);
+        std::cout << "use type:" << item << std::endl;
+    }
+    std::vector<std::string> task;
+    for (const auto& entry : fs::directory_iterator(dir)) {
+        if (!fs::is_regular_file(entry)) {
+            continue;
+        }
+        auto exten = entry.path().filename().extension().string();
+        if (std::find(t.begin(), t.end(), exten) != t.end()) {
+            std::cout << "Parse:" << entry.path().string() << std::endl;
+            task.push_back(entry.path().string());
+        }
+    }
+    // 提取内容
+    std::string content;
+    for (const auto& item : task) {
+        std::string one;
+        if (read_txt(item, one)) {
+            content.append("\n\n" + one);
+        } else {
+            std::cerr << "Can't read file: " << item << std::endl;
+            exit(1);
+        }
+    }
+    return content;
+}

jsondata.h

@@ -1,11 +1,14 @@
 #ifndef JSON_DATA
 #define JSON_DATA
 
+#include <fstream>
 #include <nlohmann/json.hpp>
 #include <optional>
 #include <string>
-#include <fstream>
+#include <vector>
+#include <filesystem>
 
+namespace fs = std::filesystem;
 struct Message {
     std::string message_content;
     std::string reasoning_content;
@@ -19,13 +22,24 @@ using json = nlohmann::json;
 class CJsonOper
 {
 public:
-    CJsonOper();
+    CJsonOper(const std::string& user_name, const std::string& model, const std::string& assistant_name);
     ~CJsonOper();
 
 public:
     std::string format_request(const std::string& content);
+    std::string multi_format_reuqest(const std::string& content, size_t per_sec_size);
     Message parse(const std::string& data);
     static bool save_md(const std::string& data, const std::string& id);
+    static bool read_txt(const std::string& path, std::string& out);
+    static std::vector<std::string> split(const std::string& input, const std::string& delimiter);
+    static size_t get_u8_len(unsigned char ch);
+    static std::string get_all_dir_content(const std::string& dir, const std::string& types);
+    static std::string trim(const std::string& input);
+
+private:
+    std::string user_{};
+    std::string model_{};
+    std::string assistant_{};
 };
 
 #endif

main.cxx

@@ -1,40 +1,69 @@
 
+#include "jsondata.h"
 #include "zapi.h"
 #include <iostream>
 #include <memory>
 #include <string>
-#include "jsondata.h"
+
+constexpr auto API_ENV_KEY = "DASHSCOPE_API_KEY";
+constexpr auto BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions";
+constexpr auto USER_NAME = "user";
+constexpr auto MODEL_NAME = "deepseek-r1";
+constexpr auto ASSISTANT_NAME = "assistant";
 
 std::string get_key()
 {
-    char* v = getenv("DASHSCOPE_API_KEY");
+    char* v = getenv(API_ENV_KEY);
     if (v) {
         return std::string(v);
     }
     return "";
 }
 
-int main()
+int main(int argc, char* argv[])
 {
+    if (argc < 4) {
+        std::cout << "note: you need set env[" << API_ENV_KEY << "] before you start." << std::endl;
+        std::cout << "argument: text type files dir, types(split with ,), question file." << std::endl;
+        std::cout << "example: deepseek-use /home/zhang/cpps cpp,xml,h /home/zhang/question.txt" << std::endl;
+        return 0;
+    }
+
     std::string api_key = get_key();
     if (api_key.empty()) {
         std::cerr << "api key not found." << std::endl;
         return -1;
     }
 
+    std::string question{};
+    if (!CJsonOper::read_txt(argv[3], question)) {
+        return -1;
+    }
+
+    std::string all_content = CJsonOper::get_all_dir_content(argv[1], argv[2]);
+    if (all_content.empty()) {
+        std::cerr << "content is empty." << std::endl;
+        return -1;
+    }
+
+    std::string req_str = question + "\n\n请查看附加数据：\n" + all_content; 
     auto api = std::make_shared<COpenAI>();
-    auto json_oper = std::make_shared<CJsonOper>();
+    auto json_oper = std::make_shared<CJsonOper>(USER_NAME, MODEL_NAME, ASSISTANT_NAME);
-    api->set_base("https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions", api_key);
+    api->set_base(BASE_URL, api_key);
-
+    std::string q = json_oper->multi_format_reuqest(req_str, 1024 * 1);
-    // 请求的 JSON 数据
-    std::string question("对于DeepSeek API，既然对话交互使用的是json格式，那我是否可以把所有的文本内容合并到json中post给deepseek，deepseek有没有说明json格式的数据大小上限值？");
-    std::string q = json_oper->format_request(question);
 
+    //std::cout << q << std::endl;
     std::string recv;
     if (api->post(q, recv)) {
         auto re = json_oper->parse(recv);
-        CJsonOper::save_md("**最终结果：**\n\n" + re.message_content + "\n\n **思考过程：** \n" + re.reasoning_content, re.id);
+        std::string use = "本次`tokens`消耗：" + std::to_string(re.prompt_tokens) + "+" + std::to_string(re.completion_tokens) +
+                          "=" + std::to_string(re.total_tokens);
+        CJsonOper::save_md(
+            use + "\n\n**最终结果：**\n\n" + re.message_content + "\n\n **思考过程：** \n\n" + re.reasoning_content, re.id);
         std::cout << "success." << std::endl;
     }
+    else {
+        std::cout << "request failed." << std::endl;
+    }
     return 0;
 }