GCC compilation error and some improvement suggestions

[PlumBlossomMaid]

Bug category

• bug - compilation error

Describe the bug

Two template functions in source/matplot/core/axes_type.h fail to compile with GCC due to incorrect template argument counts.
source/matplot/core/axes_type.h 中两个模板函数因模板参数数量错误，无法在 GCC 下编译。

Bug 1: graph() — Extra template argument (~line 2584)
Bug 1: graph() — 多余的模板参数（~第2584行）

to_vector_1d only accepts one template parameter. The extra std::pair<size_t, size_t> causes GCC to reject the call.
to_vector_1d 只接受一个模板参数。多余的 std::pair<size_t, size_t> 导致 GCC 拒绝编译。

// Current:（当前代码，GCC 拒绝）
return graph(to_vector_1d<T1, std::pair<size_t, size_t>>(edges), ...);
// Fix:（修复）
return graph(to_vector_1d<T1>(edges), ...);

Bug 2: imshow() — Extra template argument (~line 2596)
Bug 2: imshow() — 多余的模板参数（~第2596行）

to_vector_2d only accepts one template parameter. The original code tried to use to_vector_2d<T1, unsigned char> to convert data to unsigned char before calling the non-template imshow overload (which is designed for grayscale images). This intent is correct, but to_vector_2d doesn't support a second template parameter for the target type.
to_vector_2d 只接受一个模板参数。原始代码试图用 to_vector_2d<T1, unsigned char> 将数据转换为 unsigned char 后调用非模板 imshow 重载（该重载专为灰度图设计）。这个意图是正确的，但 to_vector_2d 不支持第二个模板参数来指定目标类型。

// Current:（当前代码，GCC 拒绝）
return imshow(to_vector_2d<T1, unsigned char>(gray_scale_img));

The correct fix depends on the intended API design:
正确的修复取决于 API 设计意图：

Option A — Add a two-parameter to_vector_2d that supports type conversion:
方案 A — 添加支持类型转换的双参数 to_vector_2d：

template <class T, class U>
auto to_vector_2d(const T &v) {
    // Convert each element to type U
    // ...
}

Option B — Explicitly convert to unsigned char in the template:
方案 B — 在模板中显式转换为 unsigned char：

template <class T1>
matrix_handle imshow(const IterableIterables<T1> &gray_scale_img) {
    auto v = to_vector_2d<T1>(gray_scale_img);
    std::vector<std::vector<unsigned char>> uc(v.size());
    for (size_t i = 0; i < v.size(); i++) {
        uc[i].resize(v[i].size());
        for (size_t j = 0; j < v[i].size(); j++) {
            uc[i][j] = static_cast<unsigned char>(std::clamp(v[i][j], 0.0, 255.0));
        }
    }
    return imshow(uc);
}

The same issue exists in the RGB imshow overload (~line 2606) — all four channels use to_vector_2d<T1, unsigned char> with the wrong template argument count, and all use T1 instead of T1, T2, T3, T4.
RGB 版 imshow 重载（~第2606行）也有同样的问题——四个通道都用了 to_vector_2d<T1, unsigned char>，模板参数数量错误，且全部使用 T1 而非 T1, T2, T3, T4。

Steps to Reproduce

git clone https://github.com/alandefreitas/matplotplusplus.git
cd matplotplusplus
git checkout v1.2.1
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release -DMATPLOTPP_BUILD_EXAMPLES=OFF -DMATPLOTPP_BUILD_TESTS=OFF
cmake --build . -j$(nproc)

Then compile the following minimal reproduction file:
然后编译以下最小复现文件：

// test_reproduce.cpp
#include <matplot/matplot.h>

void test_graph() {
    std::vector<std::vector<size_t>> edges = {{0,1},{1,2}};
    matplot::gca()->graph(edges);
}

void test_imshow() {
    std::vector<std::vector<double>> data = {{1,2,3},{4,5,6},{7,8,9}};
    matplot::imshow(data);
}

int main() { return 0; }

g++ -std=c++17 -I source -I build/source/matplot -c test_reproduce.cpp -o /dev/null

Output

Details

test_reproduce.cpp:5:26: error: no matching function for call to 'matplot::axes_type::graph(std::vector<std::vector<long unsigned int> >&)'
source/matplot/core/axes_type.h:2580:24: note: candidate: 'template<class T1, class T2> ... graph(IterableValues<T1>&, IterableValues<T2>&, ...)'
source/matplot/core/axes_type.h:2580:24: note:   template argument deduction/substitution failed:
test_reproduce.cpp:5:26: note:   couldn't deduce template parameter 'T2'

source/matplot/core/axes_type.h:2597:58: error: no matching function for call to 'to_vector_2d<..., unsigned char>(...)'
source/matplot/core/axes_type.h:2597:58: error: wrong number of template arguments (2, should be 1)

Platform

• cross-platform issue - linux

Environment Details:

• OS: Ubuntu
• OS Version: 22.04, 24.04
• Compiler: GCC
• Compiler version: 11.4, 12.3, 13.3

Additional context

We maintain a combined patch for these issues in our project Insight7 (https://github.com/PlumBlossomMaid/Insight7), a C++ scientific computing framework inspired by NumPy/SciPy/matplotlib/PaddlePaddle. Matplot++ is our primary plotting backend.
我们在自己的项目 Insight7（https://github.com/PlumBlossomMaid/Insight7 ，受 NumPy/SciPy/matplotlib/PaddlePaddle 启发的 C++ 科学计算框架）中维护了这些问题的合并补丁。Matplot++ 是我们的主要绘图后端。

The full diff is available at patches/matplotplusplus/gcc-compat.patch.
完整 diff 见 patches/matplotplusplus/gcc-compat.patch。

We plan to contribute to Matplot++ around the end of 2026.
我们计划在 2026 年底左右开始为 Matplot++ 贡献代码。

We also have improvement suggestions:
我们还有一些改进建议：

1. CI: actions/upload-artifact@v3 shut down since April 2026, upgrade to v4.
   CI: actions/upload-artifact@v3 自 2026 年 4 月起已停用，请升级到 v4。

2. Chinese README: A README.zh-CN.md would significantly improve adoption in the large Chinese C++ community.
   中文 README: 添加 README.zh-CN.md 可显著提升项目在中国 C++ 社区的采用率。

3. Alternative backend: Exploring AGG (Anti-Grain Geometry) as a gnuplot replacement would enable zero-dependency deployment.
   替代后端: 探索 AGG（Anti-Grain Geometry）替代 gnuplot，可实现零依赖部署。

4. AI-assisted development: Adding a QWEN.md project configuration file would help AI coding assistants Qwen Code automatically understand the project's architecture, coding conventions, and technical decisions. The QWEN.md file acts as "a project manual for AI" — when an AI assistant starts a new session, it reads this file and immediately understands the project context, reducing repetitive explanations and improving code generation quality.
   AI 辅助开发: 添加 QWEN.md 项目配置文件，让 AI 编程助手 Qwen Code 自动理解项目的架构、编码规范和技术决策。QWEN.md 相当于"给 AI 看的项目说明书"——AI 助手开启新会话时会自动读取该文件，立即理解项目上下文，减少重复解释，提升代码生成质量。

   Additionally, creating .qwen/skills/ directory with reusable workflow patterns (e.g., how to add a new plot function, how to test with gnuplot) would further improve AI-assisted development efficiency.
   此外，创建 .qwen/skills/ 目录存放可复用的工作流模式（例如如何添加新的绘图函数、如何用 gnuplot 测试），可以进一步提升 AI 辅助开发的效率。