如果你想在嵌套的子文件夹中运行一个命令来安装npm包，你可以通过npm和main package运行一个脚本。Json在根目录中。脚本将访问每个子目录并运行npm install。

下面是一个.js脚本，将实现预期的结果:

var fs = require('fs');
var resolve = require('path').resolve;
var join = require('path').join;
var cp = require('child_process');
var os = require('os');
    
// get library path
var lib = resolve(__dirname, '../lib/');
    
fs.readdirSync(lib).forEach(function(mod) {
    var modPath = join(lib, mod);
    
    // ensure path has package.json
    if (!fs.existsSync(join(modPath, 'package.json'))) {
        return;
    }

    // npm binary based on OS
    var npmCmd = os.platform().startsWith('win') ? 'npm.cmd' : 'npm';

    // install folder
    cp.spawn(npmCmd, ['i'], {
        env: process.env,
        cwd: modPath,
        stdio: 'inherit'
    });
})

请注意，这是取自StrongLoop文章的示例，该文章专门处理模块化node.js项目结构(包括嵌套组件和包)。json文件)。

如前所述，您也可以使用bash脚本实现相同的功能。

编辑:使代码在Windows工作

我的解决方案非常相似。 纯粹的node . js

下面的脚本检查所有子文件夹(递归地)，只要它们有包。Json，并在每个文件中运行NPM install。 我们可以为它添加例外:允许没有package.json的文件夹。在下面的例子中，一个这样的文件夹是“packages”。 可以将其作为“预安装”脚本运行。

const path = require('path')
const fs = require('fs')
const child_process = require('child_process')

const root = process.cwd()
npm_install_recursive(root)

// Since this script is intended to be run as a "preinstall" command,
// it will do `npm install` automatically inside the root folder in the end.
console.log('===================================================================')
console.log(`Performing "npm install" inside root folder`)
console.log('===================================================================')

// Recurses into a folder
function npm_install_recursive(folder)
{
    const has_package_json = fs.existsSync(path.join(folder, 'package.json'))

    // Abort if there's no `package.json` in this folder and it's not a "packages" folder
    if (!has_package_json && path.basename(folder) !== 'packages')
    {
        return
    }

    // If there is `package.json` in this folder then perform `npm install`.
    //
    // Since this script is intended to be run as a "preinstall" command,
    // skip the root folder, because it will be `npm install`ed in the end.
    // Hence the `folder !== root` condition.
    //
    if (has_package_json && folder !== root)
    {
        console.log('===================================================================')
        console.log(`Performing "npm install" inside ${folder === root ? 'root folder' : './' + path.relative(root, folder)}`)
        console.log('===================================================================')

        npm_install(folder)
    }

    // Recurse into subfolders
    for (let subfolder of subfolders(folder))
    {
        npm_install_recursive(subfolder)
    }
}

// Performs `npm install`
function npm_install(where)
{
    child_process.execSync('npm install', { cwd: where, env: process.env, stdio: 'inherit' })
}

// Lists subfolders in a folder
function subfolders(folder)
{
    return fs.readdirSync(folder)
        .filter(subfolder => fs.statSync(path.join(folder, subfolder)).isDirectory())
        .filter(subfolder => subfolder !== 'node_modules' && subfolder[0] !== '.')
        .map(subfolder => path.join(folder, subfolder))
}

如果您知道嵌套子目录的名称，我更喜欢使用post-install。在package.json:

"scripts": {
  "postinstall": "cd nested_dir && npm install",
  ...
}

添加Windows支持snozza的答案，以及跳过node_modules文件夹如果存在。

var fs = require('fs')
var resolve = require('path').resolve
var join = require('path').join
var cp = require('child_process')

// get library path
var lib = resolve(__dirname, '../lib/')

fs.readdirSync(lib)
  .forEach(function (mod) {
    var modPath = join(lib, mod)
    // ensure path has package.json
    if (!mod === 'node_modules' && !fs.existsSync(join(modPath, 'package.json'))) return

    // Determine OS and set command accordingly
    const cmd = /^win/.test(process.platform) ? 'npm.cmd' : 'npm';

    // install folder
    cp.spawn(cmd, ['i'], { env: process.env, cwd: modPath, stdio: 'inherit' })
})

只是作为参考，以防人们遇到这个问题。你现在可以:

添加包。Json到子文件夹中 在main package.json中安装这个子文件夹作为reference-link:

NPM安装——保存路径/到/my/子文件夹

用例1:如果你想在每个子目录中运行npm命令(每个包。Json是)，你将需要使用postinstall。

因为我经常使用npm-run-all，我用它来保持它的美观和简短(在postinstall的部分):

{
    "install:demo": "cd projects/demo && npm install",
    "install:design": "cd projects/design && npm install",
    "install:utils": "cd projects/utils && npm install",

    "postinstall": "run-p install:*"
}

这有一个额外的好处，我可以一次性安装，或单独安装。如果你不需要这个或者不希望npm-run-all作为依赖项，请查看demisx的答案(在postinstall中使用subshell)。

用例2:如果你将从根目录运行所有的npm命令(并且，例如，不会在子目录中使用npm脚本)，你可以简单地安装每个子目录，就像你安装任何依赖一样:

npm install path/to/any/directory/with/a/package-json

在后一种情况下，如果您找不到任何node_modules或package-lock，请不要感到惊讶。所有的包都将安装在根目录node_modules中，这就是为什么你不能从你的任何子目录运行你的NPM命令(需要依赖)。

如果您不确定，用例1总是有效的。

受这里提供的脚本的启发，我构建了一个可配置的示例:

can be setup to use yarn or npm can be setup to determine the command to use based on lock files so that if you set it to use yarn but a directory only has a package-lock.json it will use npm for that directory (defaults to true). configure logging runs installations in parallel using cp.spawn can do dry runs to let you see what it would do first can be run as a function or auto run using env vars when run as a function, optionally provide array of directories to check returns a promise that resolves when completed allows setting max depth to look if needed knows to stop recursing if it finds a folder with yarn workspaces (configurable) allows skipping directories using a comma separated env var or by passing the config an array of strings to match against or a function which receives the file name, file path, and the fs.Dirent obj and expects a boolean result.

const path = require('path');
const { promises: fs } = require('fs');
const cp = require('child_process');

// if you want to have it automatically run based upon
// process.cwd()
const AUTO_RUN = Boolean(process.env.RI_AUTO_RUN);

/**
 * Creates a config object from environment variables which can then be
 * overriden if executing via its exported function (config as second arg)
 */
const getConfig = (config = {}) => ({
  // we want to use yarn by default but RI_USE_YARN=false will
  // use npm instead
  useYarn: process.env.RI_USE_YARN !== 'false',
  // should we handle yarn workspaces?  if this is true (default)
  // then we will stop recursing if a package.json has the "workspaces"
  // property and we will allow `yarn` to do its thing.
  yarnWorkspaces: process.env.RI_YARN_WORKSPACES !== 'false',
  // if truthy, will run extra checks to see if there is a package-lock.json
  // or yarn.lock file in a given directory and use that installer if so.
  detectLockFiles: process.env.RI_DETECT_LOCK_FILES !== 'false',
  // what kind of logging should be done on the spawned processes?
  // if this exists and it is not errors it will log everything
  // otherwise it will only log stderr and spawn errors
  log: process.env.RI_LOG || 'errors',
  // max depth to recurse?
  maxDepth: process.env.RI_MAX_DEPTH || Infinity,
  // do not install at the root directory?
  ignoreRoot: Boolean(process.env.RI_IGNORE_ROOT),
  // an array (or comma separated string for env var) of directories
  // to skip while recursing. if array, can pass functions which
  // return a boolean after receiving the dir path and fs.Dirent args
  // @see https://nodejs.org/api/fs.html#fs_class_fs_dirent
  skipDirectories: process.env.RI_SKIP_DIRS
    ? process.env.RI_SKIP_DIRS.split(',').map(str => str.trim())
    : undefined,
  // just run through and log the actions that would be taken?
  dry: Boolean(process.env.RI_DRY_RUN),
  ...config
});

function handleSpawnedProcess(dir, log, proc) {
  return new Promise((resolve, reject) => {
    proc.on('error', error => {
      console.log(`
----------------
  [RI] | [ERROR] | Failed to Spawn Process
  - Path:   ${dir}
  - Reason: ${error.message}
----------------
  `);
      reject(error);
    });

    if (log) {
      proc.stderr.on('data', data => {
        console.error(`[RI] | [${dir}] | ${data}`);
      });
    }

    if (log && log !== 'errors') {
      proc.stdout.on('data', data => {
        console.log(`[RI] | [${dir}] | ${data}`);
      });
    }

    proc.on('close', code => {
      if (log && log !== 'errors') {
        console.log(`
----------------
  [RI] | [COMPLETE] | Spawned Process Closed
  - Path: ${dir}
  - Code: ${code}
----------------
        `);
      }
      if (code === 0) {
        resolve();
      } else {
        reject(
          new Error(
            `[RI] | [ERROR] | [${dir}] | failed to install with exit code ${code}`
          )
        );
      }
    });
  });
}

async function recurseDirectory(rootDir, config) {
  const {
    useYarn,
    yarnWorkspaces,
    detectLockFiles,
    log,
    maxDepth,
    ignoreRoot,
    skipDirectories,
    dry
  } = config;

  const installPromises = [];

  function install(cmd, folder, relativeDir) {
    const proc = cp.spawn(cmd, ['install'], {
      cwd: folder,
      env: process.env
    });
    installPromises.push(handleSpawnedProcess(relativeDir, log, proc));
  }

  function shouldSkipFile(filePath, file) {
    if (!file.isDirectory() || file.name === 'node_modules') {
      return true;
    }
    if (!skipDirectories) {
      return false;
    }
    return skipDirectories.some(check =>
      typeof check === 'function' ? check(filePath, file) : check === file.name
    );
  }

  async function getInstallCommand(folder) {
    let cmd = useYarn ? 'yarn' : 'npm';
    if (detectLockFiles) {
      const [hasYarnLock, hasPackageLock] = await Promise.all([
        fs
          .readFile(path.join(folder, 'yarn.lock'))
          .then(() => true)
          .catch(() => false),
        fs
          .readFile(path.join(folder, 'package-lock.json'))
          .then(() => true)
          .catch(() => false)
      ]);
      if (cmd === 'yarn' && !hasYarnLock && hasPackageLock) {
        cmd = 'npm';
      } else if (cmd === 'npm' && !hasPackageLock && hasYarnLock) {
        cmd = 'yarn';
      }
    }
    return cmd;
  }

  async function installRecursively(folder, depth = 0) {
    if (dry || (log && log !== 'errors')) {
      console.log('[RI] | Check Directory --> ', folder);
    }

    let pkg;

    if (folder !== rootDir || !ignoreRoot) {
      try {
        // Check if package.json exists, if it doesnt this will error and move on
        pkg = JSON.parse(await fs.readFile(path.join(folder, 'package.json')));
        // get the command that we should use.  if lock checking is enabled it will
        // also determine what installer to use based on the available lock files
        const cmd = await getInstallCommand(folder);
        const relativeDir = `${path.basename(rootDir)} -> ./${path.relative(
          rootDir,
          folder
        )}`;
        if (dry || (log && log !== 'errors')) {
          console.log(
            `[RI] | Performing (${cmd} install) at path "${relativeDir}"`
          );
        }
        if (!dry) {
          install(cmd, folder, relativeDir);
        }
      } catch {
        // do nothing when error caught as it simply indicates package.json likely doesnt
        // exist.
      }
    }

    if (
      depth >= maxDepth ||
      (pkg && useYarn && yarnWorkspaces && pkg.workspaces)
    ) {
      // if we have reached maxDepth or if our package.json in the current directory
      // contains yarn workspaces then we use yarn for installing then this is the last
      // directory we will attempt to install.
      return;
    }

    const files = await fs.readdir(folder, { withFileTypes: true });

    return Promise.all(
      files.map(file => {
        const filePath = path.join(folder, file.name);
        return shouldSkipFile(filePath, file)
          ? undefined
          : installRecursively(filePath, depth + 1);
      })
    );
  }

  await installRecursively(rootDir);
  await Promise.all(installPromises);
}

async function startRecursiveInstall(directories, _config) {
  const config = getConfig(_config);
  const promise = Array.isArray(directories)
    ? Promise.all(directories.map(rootDir => recurseDirectory(rootDir, config)))
    : recurseDirectory(directories, config);
  await promise;
}

if (AUTO_RUN) {
  startRecursiveInstall(process.cwd());
}

module.exports = startRecursiveInstall;

随着它的使用:

const installRecursively = require('./recursive-install');

installRecursively(process.cwd(), { dry: true })

根据@Scott的回答，只要知道子目录名，安装|postinstall脚本是最简单的方法。这就是我如何对多个子dirs运行它。例如，假设我们在monorepo根目录下有api/、web/和shared/子项目:

// In monorepo root package.json
{
...
 "scripts": {
    "postinstall": "(cd api && npm install); (cd web && npm install); (cd shared && npm install)"
  },
}

在Windows上，替换;在圆括号&&之间。

// In monorepo root package.json
{
...
 "scripts": {
    "postinstall": "(cd api && npm install) && (cd web && npm install) && (cd shared && npm install)"
  },
}

如果你在你的系统上找到了实用程序，你可以尝试在你的应用程序根目录下运行以下命令: 找到。！ -path "*/node_modules/*" -name "包。-execdir NPM install \;

基本上，找到所有的包。Json文件，并在该目录下运行NPM install，跳过所有node_modules目录。

find . -maxdepth 1 -type d \( ! -name . \) -exec bash -c "cd '{}' && npm install" \;

正如fgblomqvist在评论中提到的，npm现在也支持工作区了。

有些答案相当古老。我认为现在我们有一些新的选择来建立单回购。

我建议使用纱线工作区:

工作区是一种设置包架构的新方法，从Yarn 1.0开始默认提供。它允许你设置多个包，这样你只需要运行一次yarn install就可以一次安装所有的包。

如果你更喜欢或不得不使用npm，我建议你看看lerna:

Lerna是一个工具，它优化了使用git和npm管理多包存储库的工作流。

Lerna也可以完美地使用纱线工作区。我刚刚完成了一个monorepo项目的设置-示例。

下面是一个配置为使用npm + lerna - MDC Web的多包项目的示例:他们使用package运行lerna引导。json是postinstall。

[macOS、Linux用户]:

我创建了一个bash文件来安装项目和嵌套文件夹中的所有依赖项。

find . -name node_modules -prune -o -name package.json -execdir npm install \;

解释:在根目录中，排除node_modules文件夹(即使在嵌套文件夹中)，找到包含包的目录。Json文件，然后运行NPM install命令。

如果你只是想找到指定的文件夹(例如:abc123, def456文件夹)，运行如下:

find ./abc123/* ./def456/* -name node_modules -prune -o -name package.json -execdir npm install \;

接受的答案是有效的，但是你可以使用——prefix在选定的位置运行npm命令。

"postinstall": "npm --prefix ./nested_dir install"

——prefix适用于任何npm命令，而不仅仅是install。

还可以查看当前前缀with

npm prefix

并将全局安装(-g)文件夹设置为

npm config set prefix "folder_path"

也许是TMI，但你懂的…

要在每个子目录上运行npm install，你可以这样做:

"scripts": {
  ...
  "install:all": "for D in */; do npm install --cwd \"${D}\"; done"
}

在哪里

安装:所有只是脚本的名称，你可以随意命名

当前迭代的目录名

*/指定要查找子目录的位置。目录/*/将列出目录/内的所有目录，目录/*/*/将列出两层中的所有目录。

在指定文件夹中安装所有依赖项

你也可以运行一些命令，例如:

D在*/;&& npm install——cwd \"${D}\";完成

每次迭代都会打印“Installing stuff on your_subfolder/”。

这也适用于纱线

任何可以获取目录列表并运行shell命令的语言都可以为您完成此任务。

我知道这不是OP想要的答案，但这是一个永远有效的答案。你需要创建一个子目录名数组，然后循环遍历它们并运行npm i，或者任何你需要运行的命令。

作为参考，我尝试了npm I **/，它只是安装了父目录中所有子目录中的模块。这很不直观，但不用说，这不是你需要的解决方案。