内省构建步骤#

当你对特定的Python扩展模块或其他构建目标有疑问时，有多种方法可以弄清楚构建系统到底在做什么。除了查看感兴趣目标的``meson.build``内容外，还包括：

阅读生成在构建目录中的 build.ninja 文件，
使用 meson introspect 来了解更多关于构建选项、依赖关系和用于目标的标志的信息，
阅读 <build-dir>/meson-info/*.json 以获取有关已发现的依赖项、Meson计划安装文件的位置等的详细信息。

这些内容在构建的配置阶段（即 meson setup）运行后都可以获取。通常，查看这些信息比运行构建并阅读完整的构建日志更为有效。

`ninja.build` 文件#

作为一个例子，假设我们对 scipy.linalg._decomp_update 感兴趣。从 scipy/linalg/meson.build 中我们了解到这个扩展是用模板化的 Cython 代码编写的，并且没有使用特殊的编译标志，也没有超出 numpy 的包含目录。所以下一步是查看 build.ninja。在编辑器中打开该文件并搜索 _decomp_update。你将找到这一组通用的和目标特定的规则（注意，此代码块中的注释在 build.ninja 中不存在，仅在此文档部分添加以解释正在发生的事情）：

# These rules are usually not needed to understand the problem, but can be looked up at the top of the file:
rule c_COMPILER
 command = /home/username/anaconda3/envs/scipy-dev/bin/x86_64-conda-linux-gnu-cc $ARGS -MD -MQ $out -MF $DEPFILE -o $out -c $in
 deps = gcc
 depfile = $DEPFILE_UNQUOTED
 description = Compiling C object $out

rule c_LINKER
 command = /home/username/anaconda3/envs/scipy-dev/bin/x86_64-conda-linux-gnu-cc $ARGS -o $out $in $LINK_ARGS
 description = Linking target $out

# step 1: `.pyx.in` to `.pyx` code generation with Tempita
build scipy/linalg/_decomp_update.pyx: CUSTOM_COMMAND ../scipy/linalg/_decomp_update.pyx.in | ../scipy/_build_utils/tempita.py /home/username/anaconda3/envs/scipy-dev/bin/python3.10
 COMMAND = /home/username/anaconda3/envs/scipy-dev/bin/python3.10 ../scipy/_build_utils/tempita.py ../scipy/linalg/_decomp_update.pyx.in -o scipy/linalg
 description = Generating$ scipy/linalg/_decomp_update$ with$ a$ custom$ command

# step 2: `.pyx` to `.c` compilation with Cython
build scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/_decomp_update.c: CUSTOM_COMMAND scipy/linalg/_decomp_update.pyx | /home/username/code/scipy/scipy/_build_utils/cythoner.py scipy/__init__.py scipy/linalg/__init__.py scipy/linalg/cython_blas.pyx
 DESC = Generating$ 'scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/_decomp_update.c'.
 COMMAND = /home/username/anaconda3/envs/scipy-dev/bin/python3.10 /home/username/code/scipy/scipy/_build_utils/cythoner.py scipy/linalg/_decomp_update.pyx scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/_decomp_update.c

# step 3: use C compiler to go from `.c` to object file (`.o`)
build scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/meson-generated__decomp_update.c.o: c_COMPILER scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/_decomp_update.c
 DEPFILE = scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/meson-generated__decomp_update.c.o.d
 DEPFILE_UNQUOTED = scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/meson-generated__decomp_update.c.o.d
 ARGS = -Iscipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p -Iscipy/linalg -I../scipy/linalg -I/home/username/anaconda3/envs/scipy-dev/lib/python3.10/site-packages/numpy/core/include -I/home/username/anaconda3/envs/scipy-dev/include/python3.10 -fvisibility=hidden -fdiagnostics-color=always -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -std=c99 -O2 -g -Wno-unused-but-set-variable -Wno-unused-function -Wno-conversion -Wno-misleading-indentation -fPIC -Wno-cpp

# step 4: generate a symbol file (uses `meson --internal symbolextractor`); you can safely ignore this step
build scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/_decomp_update.cpython-310-x86_64-linux-gnu.so.symbols: SHSYM scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so
 IMPLIB = scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so

# step 5: link the `.o` file to obtain the file extension module (`.so`)
build scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so: c_LINKER scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/meson-generated__decomp_update.c.o | /home/username/anaconda3/envs/scipy-dev/x86_64-conda-linux-gnu/sysroot/lib64/libm-2.12.so /home/username/anaconda3/envs/scipy-dev/x86_64-conda-linux-gnu/sysroot/usr/lib64/libm.a
 LINK_ARGS = -L/home/username/anaconda3/envs/scipy-dev/lib -Wl,--as-needed -Wl,--allow-shlib-undefined -shared -fPIC -Wl,--start-group -lm -Wl,--end-group -Wl,-O2 -Wl,--sort-common -Wl,--as-needed -Wl,-z,relro -Wl,-z,now -Wl,--disable-new-dtags -Wl,--gc-sections -Wl,--allow-shlib-undefined -Wl,-rpath,/home/username/anaconda3/envs/scipy-dev/lib -Wl,-rpath-link,/home/username/anaconda3/envs/scipy-dev/lib

使用 `meson introspect`#

如果我们想从另一个角度来看 _decomp_update ，我们可以使用（例如） meson introspect --targets -i <build-dir> > targets.json 来生成可读的 JSON。在生成的文件中搜索我们感兴趣的目标显示：

{
    "name": "_decomp_update",
    "id": "b4ac6f0@@_decomp_update@cus",
    "type": "custom",
    "defined_in": "/home/username/code/scipy/scipy/linalg/meson.build",
    "filename": [
        "/home/username/code/scipy/build/scipy/linalg/_decomp_update.pyx"
    ],
    "build_by_default": false,
    "target_sources": [
        {
            "language": "unknown",
            "compiler": [
                "/home/username/anaconda3/envs/scipy-dev/bin/python3.10",
                "/home/username/code/scipy/scipy/_build_utils/tempita.py",
                "@INPUT@",
                "-o",
                "@OUTDIR@"
            ],
            "parameters": [],
            "sources": [
                "/home/username/code/scipy/scipy/linalg/_decomp_update.pyx.in"
            ],
            "generated_sources": []
        }
    ],
    "extra_files": [],
    "subproject": null,
    "installed": false
},
{
    "name": "_decomp_update.cpython-310-x86_64-linux-gnu",
    "id": "b4ac6f0@@_decomp_update.cpython-310-x86_64-linux-gnu@sha",
    "type": "shared module",
    "defined_in": "/home/username/code/scipy/scipy/linalg/meson.build",
    "filename": [
        "/home/username/code/scipy/build/scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so"
    ],
    "build_by_default": true,
    "target_sources": [
        {
            "language": "c",
            "compiler": [
                "/home/username/anaconda3/envs/scipy-dev/bin/x86_64-conda-linux-gnu-cc"
            ],
            "parameters": [
                "-I/home/username/code/scipy/build/scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p",
                "-I/home/username/code/scipy/build/scipy/linalg",
                "-I/home/username/code/scipy/scipy/linalg",
                "-I/home/username/anaconda3/envs/scipy-dev/lib/python3.10/site-packages/numpy/core/include",
                "-I/home/username/anaconda3/envs/scipy-dev/include/python3.10",
                "-fvisibility=hidden",
                "-fdiagnostics-color=always",
                "-D_FILE_OFFSET_BITS=64",
                "-Wall",
                "-Winvalid-pch",
                "-std=c99",
                "-O2",
                "-g",
                "-Wno-unused-but-set-variable",
                "-Wno-unused-function",
                "-Wno-conversion",
                "-Wno-misleading-indentation",
                "-fPIC",
                "-Wno-cpp"
            ],
            "sources": [],
            "generated_sources": [
                "/home/username/code/scipy/build/scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so.p/_decomp_update.c"
            ]
        }
    ],
    "extra_files": [],
    "subproject": null,
    "installed": true,
    "install_filename": [
        "/home/username/code/scipy/build-install/lib/python3.10/site-packages/scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so"
    ]
},

这告诉我们很多信息，比如将使用哪些包含目录，在哪里可以找到Cython生成的C代码，以及使用了哪些编译标志。meson introspect --help 提供了关于全部功能及其使用方法的良好文档。

`meson-info` JSON 文件#

在 <build-dir>/meson-info/ 中有许多不同的 JSON 文件。这些文件的名称具有描述性，暗示了它们的内容。例如，最终的 _decomp_update 扩展安装位置在 intro-install_plan.json 中描述（注意，这些文件不是预先格式化的，通过 JSON 格式化工具运行它们会有所帮助）：

"/home/username/code/scipy/build/scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so":{
   "destination":"{py_platlib}/scipy/linalg/_decomp_update.cpython-310-x86_64-linux-gnu.so",
   "tag":"runtime"
},

我们可能还对了解构建的配置阶段检测到了哪些依赖项感兴趣。因此，我们查看 intro-dependencies.json：

[
   {
      "name":"python",
      "version":"3.10",
      "compile_args":[
         "-I/home/username/anaconda3/envs/scipy-dev/include/python3.10"
      ],
      "link_args":[

      ]
   },
   {
      "name":"openblas",
      "version":"0.3.20",
      "compile_args":[
         "-I/home/username/anaconda3/envs/scipy-dev/include"
      ],
      "link_args":[
         "/home/username/anaconda3/envs/scipy-dev/lib/libopenblas.so"
      ]
   },
   {
      "name":"threads",
      "version":"unknown",
      "compile_args":[
         "-pthread"
      ],
      "link_args":[
         "-pthread"
      ]
   }
]

这告诉我们找到了三个依赖项。注意：numpy 和一些其他构建时依赖项在这里缺失，因为我们没有（还）使用内置的 dependency() Meson 命令搜索这些依赖项。

内省构建步骤#

ninja.build 文件#

使用 meson introspect#

meson-info JSON 文件#

`ninja.build` 文件#

使用 `meson introspect`#

`meson-info` JSON 文件#