如何在Bazel中添加来自Git的Python包作为依赖

你已经通过http_archive拉取了nvdiffrast的源码，接下来需要借助rules_python来基于setup.py构建这个包，具体步骤如下：

1. 引入rules_python规则 #

首先在WORKSPACE文件中添加rules_python依赖，确保能处理Python包和C/CUDA扩展的构建：

load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")

# 引入rules_python
http_archive(
    name = "rules_python",
    sha256 = "f84bf656c22957d0d7728e933302e00a0d3d1578b899d379cfb1f216a4d9c487",
    url = "https://github.com/bazelbuild/rules_python/releases/download/0.25.0/rules_python-0.25.0.tar.gz",
)

# 注册Python工具链
load("@rules_python//python:repositories.bzl", "python_register_toolchains")
python_register_toolchains(
    name = "python3_10",
    python_version = "3.10",
)

# 你的nvdiffrast源码拉取配置
http_archive(
    name = "nvdiffrast",
    url = "https://github.com/NVlabs/nvdiffrast/archive/refs/tags/v0.3.1.tar.gz",
    integrity = "sha256:替换为实际的哈希值",
    strip_prefix = "nvdiffrast-0.3.1",
)

2. 为nvdiffrast添加BUILD配置 #

nvdiffrast包含C/CUDA扩展，需要编写BUILD.bazel定义构建规则。推荐在本地项目中创建配置文件，比如third_party/nvdiffrast/BUILD.bazel，内容如下：

load("@rules_python//python:defs.bzl", "py_library", "py_extension")
load("@cuda//:defs.bzl", "cuda_library") # 需提前引入CUDA规则以支持GPU编译

# 构建CUDA/C++扩展
py_extension(
    name = "_nvdiffrast",
    srcs = glob([
        "nvdiffrast/*.cpp",
        "nvdiffrast/*.cu",
    ]),
    hdrs = glob([
        "nvdiffrast/*.h",
    ]),
    deps = [
        "@cuda//:cuda_headers",
    ],
    copts = [
        "-std=c++11",
        "-DNVDR_TORCH", # 基于PyTorch使用时添加该宏定义
    ],
    linkopts = [
        "-L/usr/local/cuda/lib64",
        "-lcudart",
    ],
)

# 定义Python库，关联编译好的扩展
py_library(
    name = "nvdiffrast",
    srcs = glob([
        "nvdiffrast/*.py",
        "setup.py",
    ]),
    data = [":_nvdiffrast"],
    imports = ["."],
    visibility = ["//visibility:public"],
    deps = [
        # 添加nvdiffrast依赖的其他Python包，比如PyTorch
        "@pytorch//:torch",
    ],
)

然后在WORKSPACE的http_archive中添加build_file参数，指向上述配置：

http_archive(
    name = "nvdiffrast",
    url = "https://github.com/NVlabs/nvdiffrast/archive/refs/tags/v0.3.1.tar.gz",
    integrity = "sha256:替换为实际的哈希值",
    strip_prefix = "nvdiffrast-0.3.1",
    build_file = "//third_party/nvdiffrast:BUILD.bazel",
)

3. 在项目中依赖该包 #

在你的Python目标（如py_binary或py_library）中，直接添加依赖即可：

py_binary(
    name = "my_app",
    srcs = ["main.py"],
    deps = [
        "@nvdiffrast//:nvdiffrast",
    ],
)

注意事项 #

• 若需CUDA支持，要确保Bazel能识别CUDA环境，需额外引入CUDA相关规则并配置路径
• 编译选项需根据本地环境调整，比如CUDA安装路径、C++标准版本等
• 若nvdiffrast依赖其他PyPI包，用rules_python的pip_parse或pip_install统一管理

内容的提问来源于stack exchange，提问作者zlenyk