glob vs micromatch vs minimatch vs picomatch
ファイルパターンマッチングとグロビング戦略の比較
globmicromatchminimatchpicomatch類似パッケージ:

ファイルパターンマッチングとグロビング戦略の比較

globmicromatchminimatchpicomatch は、すべてファイルパスや文字列のパターンマッチングに関わるライブラリですが、役割と動作環境が異なります。glob は実際にファイルシステムを走査して一致するファイルパスを取得するためのツールです。一方、minimatchmicromatchpicomatch は、すでに取得されたパスリストや文字列に対して、グロブパターンが一致するかどうかを判定するマッチングエンジンです。特に micromatchpicomatch は高性能なマッチングを目的として設計されており、ビルドツールやバンドラー内部で広く利用されています。

npmのダウンロードトレンド

3 年

GitHub Starsランキング

統計詳細

パッケージ
ダウンロード数
Stars
サイズ
Issues
公開日時
ライセンス
glob08,7171.61 MB32ヶ月前BlueOak-1.0.0
micromatch03,02656.6 kB442年前MIT
minimatch03,503536 kB71ヶ月前BlueOak-1.0.0
picomatch01,25291.4 kB431ヶ月前MIT

glob vs micromatch vs minimatch vs picomatch: 技術的な深層比較

ファイルパターン matching は、フロントエンド開発においてビルド設定、テスト実行、ファイル監視など、あらゆる場面で必要になります。しかし、globmicromatchminimatchpicomatch は似て非なるツールです。これらを正しく使い分けることは、パフォーマンスと保守性に直結します。

🗂️ 根本的な違い:ファイル走査 vs 文字列マッチング

まず最も重要な点は、glob だけが実際にファイルシステムを走査できるツールだということです。他の 3 つは、あくまで「文字列」と「パターン」を照合するエンジンです。

glob はディスクを走査し、一致するファイルパスの配列を返します。

// glob: ファイルシステムを走査
import { glob } from 'glob';

const files = await glob('**/*.js');
// 結果: ['src/index.js', 'src/utils.js', ...]

minimatchmicromatchpicomatch は、すでに存在するパスリストに対してマッチングを行います。

// minimatch: 単一のパスを判定
import { minimatch } from 'minimatch';

const isMatch = minimatch('src/index.js', '**/*.js');
// 結果: true
// micromatch: リストから一致するものをフィルタ
import micromatch from 'micromatch';

const files = micromatch(['src/index.js', 'README.md'], '**/*.js');
// 結果: ['src/index.js']
// picomatch: マッチャー関数を生成して判定
import picomatch from 'picomatch';

const isMatch = picomatch('**/*.js');
const result = isMatch('src/index.js');
// 結果: true

⚡ パフォーマンスと実行モデル

大量のファイルを処理する場景では、マッチングエンジンの速度がボトルネックになります。minimatch は歴史的に標準でしたが、picomatchmicromatch はより高速に設計されています。

minimatch は毎回パターンを解析するため、ループ処理では遅くなりがちです。

// minimatch: 毎回解析が発生
const files = allFiles.filter(file => minimatch(file, '**/*.test.js'));

picomatch はパターンをコンパイルして関数を返すため、複数回の判定で圧倒的に速いです。

// picomatch: 事前コンパイル済み関数を使用
const matcher = picomatch('**/*.test.js');
const files = allFiles.filter(file => matcher(file));

micromatch も内部で最適化を行っていますが、picomatch を単体で使う方が軽量な場合があります。

// micromatch: 内部最適化済み
const files = micromatch(allFiles, '**/*.test.js');

glob はファイル I/O が含まれるため、単純な速度比較はできませんが、内部で効率的なトラバースを行います。

// glob: I/O コストを含む
const files = await glob('**/*.test.js');

🛠️ 機能とパターンサポート

すべてのライブラリが標準的なグロブパターン(***?)をサポートしていますが、拡張機能に差があります。

minimatch は Node.js の標準挙動に最も近いですが、拡張機能は限定的です。

// minimatch: 基本的な brace expansion
minimatch('a.js', '{a,b}.js'); // true

micromatch は拡張 glob 機能(extglob)を強くサポートしています。

// micromatch: 高度なパターン
micromatch(['a.js', 'b.ts'], '*.!(ts)'); 
// 結果: ['a.js'] (ts 以外にマッチ)

picomatch も同様に高度なパターンをサポートし、オプションで厳密な制御が可能です。

// picomatch: オプション制御
const match = picomatch('*.!(ts)', { strictErrors: false });

glob はこれらのマッチングエンジンを内部で利用しつつ、ファイルシステムの特性(大文字小文字の区別など)を扱います。

// glob: ファイルシステム固有のオプション
const files = await glob('**/*.js', { caseSensitiveMatch: false });

⚠️ 保守性と将来性

ライブラリのメンテナンス状況も選定基準になります。

minimatch は長年維持されていますが、パフォーマンス面で時代遅れになりつつあります。新しいプロジェクトで積極的に選ぶ理由はありません。

glob は最近のバージョン(v9 以降)で ESM 専用となり、CommonJS 環境では注意が必要です。async/await が標準となりました。

micromatchpicomatch は現代のビルドツールエコシステムで事実上の標準であり、活発にメンテナンスされています。

📊 比較サマリー

機能globmicromatchminimatchpicomatch
主な目的ファイル走査文字列マッチング文字列マッチング文字列マッチング
戻り値ファイルパス配列マッチした配列真偽値 (boolean)マッチャー関数
速度I/O に依存高速普通最速
パターン機能標準 + FS オプション拡張 glob 対応標準拡張 glob 対応
モジュールESM (v10)CommonJS/ESMCommonJS/ESMCommonJS/ESM

💡 実務での推奨事項

ファイルを探す必要がある場合glob 一択です。ただし、バージョン 10 は ESM 専用なので、プロジェクトの設定を確認してください。

ビルドツールや設定ファイルでパターンを判定する場合micromatch が最もバランスが良いです。Webpack や Rollup などのエコシステムとの親和性が高く、予期せぬ挙動が少ないです。

独自のツールや CLI で大量のファイルをフィルタする場合picomatch を使用してマッチャー関数を生成し、ループ内で使い回すのが最も効率的です。

既存のレガシーコードを維持する場合minimatch が使われている場合は、すぐに書き換える必要はありません。ただし、新規追加部分にはより高速な alternatif を検討してください。

🏁 結論

これら 4 つは競合関係というより、役割が補完し合っています。glob でファイルを集め、picomatchmicromatch で詳細なフィルタリングを行うのが、現代のフロントエンドアーキテクチャにおける標準的な組み合わせです。パフォーマンスが重要な場景では、minimatch から picomatch への移行を検討する価値があります。

選び方: glob vs micromatch vs minimatch vs picomatch

  • glob:

    実際にディスク上のファイルを走査してパスリストを取得する必要がある場合は glob を選択します。ファイルシステムの非同期処理やストリーミング対応が必要な場合に最適です。ただし、最近のバージョンは ESM 専用かつ Promise ベースであるため、環境との互換性を確認する必要があります。

  • micromatch:

    複雑なパターン matching や高いパフォーマンスが必要な文字列マッチングには micromatch を選択します。Webpack や Vite などのビルドツールで標準的に採用されており、信頼性が高いです。内部で picomatch を使用しているため、機能と速度のバランスが取れています。

  • minimatch:

    既存のプロジェクトで標準的な挙動が必要だったり、他のライブラリとの依存関係で既に導入されている場合は minimatch を選択します。Node.js の標準挙動に近く、安定していますが、大量のパスを処理する場合は速度面で劣る可能性があります。

  • picomatch:

    究極の速度と軽量化を追求する場合は picomatch を選択します。マッチャー関数を生成して使い回すスタイルのため、同じパターンで多数のファイルを判定する场景に最も適しています。micromatch の内部エンジンとしても使われており、単体でも非常に高速です。

glob のREADME

Glob

Match files using the patterns the shell uses.

The most correct and second fastest glob implementation in JavaScript. (See Comparison to Other JavaScript Glob Implementations at the bottom of this readme.)

a fun cartoon logo made of glob
characters

Usage

Install with npm

npm i glob

[!NOTE] The npm package name is not node-glob that's a different thing that was abandoned years ago. Just glob.

// load using import
import { glob, globSync, globStream, globStreamSync, Glob } from 'glob'
// or using commonjs, that's fine, too
const {
  glob,
  globSync,
  globStream,
  globStreamSync,
  Glob,
} = require('glob')

// the main glob() and globSync() resolve/return array of filenames

// all js files, but don't look in node_modules
const jsfiles = await glob('**/*.js', { ignore: 'node_modules/**' })

// pass in a signal to cancel the glob walk
const stopAfter100ms = await glob('**/*.css', {
  signal: AbortSignal.timeout(100),
})

// multiple patterns supported as well
const images = await glob(['css/*.{png,jpeg}', 'public/*.{png,jpeg}'])

// but of course you can do that with the glob pattern also
// the sync function is the same, just returns a string[] instead
// of Promise<string[]>
const imagesAlt = globSync('{css,public}/*.{png,jpeg}')

// you can also stream them, this is a Minipass stream
const filesStream = globStream(['**/*.dat', 'logs/**/*.log'])

// construct a Glob object if you wanna do it that way, which
// allows for much faster walks if you have to look in the same
// folder multiple times.
const g = new Glob('**/foo', {})
// glob objects are async iterators, can also do globIterate() or
// g.iterate(), same deal
for await (const file of g) {
  console.log('found a foo file:', file)
}
// pass a glob as the glob options to reuse its settings and caches
const g2 = new Glob('**/bar', g)
// sync iteration works as well
for (const file of g2) {
  console.log('found a bar file:', file)
}

// you can also pass withFileTypes: true to get Path objects
// these are like a fs.Dirent, but with some more added powers
// check out https://isaacs.github.io/path-scurry/classes/PathBase.html
// for more info on their API
const g3 = new Glob('**/baz/**', { withFileTypes: true })
g3.stream().on('data', path => {
  console.log(
    'got a path object',
    path.fullpath(),
    path.isDirectory(),
    path.readdirSync().map(e => e.name),
  )
})

// if you use stat:true and withFileTypes, you can sort results
// by things like modified time, filter by permission mode, etc.
// All Stats fields will be available in that case. Slightly
// slower, though.
// For example:
const results = await glob('**', { stat: true, withFileTypes: true })

const timeSortedFiles = results
  .sort((a, b) => a.mtimeMs - b.mtimeMs)
  .map(path => path.fullpath())

const groupReadableFiles = results
  .filter(path => path.mode & 0o040)
  .map(path => path.fullpath())

// custom ignores can be done like this, for example by saying
// you'll ignore all markdown files, and all folders named 'docs'
const customIgnoreResults = await glob('**', {
  ignore: {
    ignored: p => /\.md$/.test(p.name),
    childrenIgnored: p => p.isNamed('docs'),
  },
})

// another fun use case, only return files with the same name as
// their parent folder, plus either `.ts` or `.js`
const folderNamedModules = await glob('**/*.{ts,js}', {
  ignore: {
    ignored: p => {
      const pp = p.parent
      return !(p.isNamed(pp.name + '.ts') || p.isNamed(pp.name + '.js'))
    },
  },
})

// find all files edited in the last hour, to do this, we ignore
// all of them that are more than an hour old
const newFiles = await glob('**', {
  // need stat so we have mtime
  stat: true,
  // only want the files, not the dirs
  nodir: true,
  ignore: {
    ignored: p => {
      return new Date() - p.mtime > 60 * 60 * 1000
    },
    // could add similar childrenIgnored here as well, but
    // directory mtime is inconsistent across platforms, so
    // probably better not to, unless you know the system
    // tracks this reliably.
  },
})

[!NOTE] Glob patterns should always use / as a path separator, even on Windows systems, as \ is used to escape glob characters. If you wish to use \ as a path separator instead of using it as an escape character on Windows platforms, you may set windowsPathsNoEscape:true in the options. In this mode, special glob characters cannot be escaped, making it impossible to match a literal * ? and so on in filenames.

Command Line Interface

The glob CLI has been moved to the glob-bin package, and must be installed separately, as of version 13.

npm install glob-bin

glob(pattern: string | string[], options?: GlobOptions) => Promise<string[] | Path[]>

Perform an asynchronous glob search for the pattern(s) specified. Returns Path objects if the withFileTypes option is set to true. See below for full options field desciptions.

globSync(pattern: string | string[], options?: GlobOptions) => string[] | Path[]

Synchronous form of glob().

Alias: glob.sync()

globIterate(pattern: string | string[], options?: GlobOptions) => AsyncGenerator<string>

Return an async iterator for walking glob pattern matches.

Alias: glob.iterate()

globIterateSync(pattern: string | string[], options?: GlobOptions) => Generator<string>

Return a sync iterator for walking glob pattern matches.

Alias: glob.iterate.sync(), glob.sync.iterate()

globStream(pattern: string | string[], options?: GlobOptions) => Minipass<string | Path>

Return a stream that emits all the strings or Path objects and then emits end when completed.

Alias: glob.stream()

globStreamSync(pattern: string | string[], options?: GlobOptions) => Minipass<string | Path>

Syncronous form of globStream(). Will read all the matches as fast as you consume them, even all in a single tick if you consume them immediately, but will still respond to backpressure if they're not consumed immediately.

Alias: glob.stream.sync(), glob.sync.stream()

hasMagic(pattern: string | string[], options?: GlobOptions) => boolean

Returns true if the provided pattern contains any "magic" glob characters, given the options provided.

Brace expansion is not considered "magic" unless the magicalBraces option is set, as brace expansion just turns one string into an array of strings. So a pattern like 'x{a,b}y' would return false, because 'xay' and 'xby' both do not contain any magic glob characters, and it's treated the same as if you had called it on ['xay', 'xby']. When magicalBraces:true is in the options, brace expansion is treated as a pattern having magic.

escape(pattern: string, options?: GlobOptions) => string

Escape all magic characters in a glob pattern, so that it will only ever match literal strings

If the windowsPathsNoEscape option is used, then characters are escaped by wrapping in [], because a magic character wrapped in a character class can only be satisfied by that exact character.

Slashes (and backslashes in windowsPathsNoEscape mode) cannot be escaped or unescaped.

unescape(pattern: string, options?: GlobOptions) => string

Un-escape a glob string that may contain some escaped characters.

If the windowsPathsNoEscape option is used, then square-brace escapes are removed, but not backslash escapes. For example, it will turn the string '[*]' into *, but it will not turn '\\*' into '*', because \ is a path separator in windowsPathsNoEscape mode.

When windowsPathsNoEscape is not set, then both brace escapes and backslash escapes are removed.

Slashes (and backslashes in windowsPathsNoEscape mode) cannot be escaped or unescaped.

Class Glob

An object that can perform glob pattern traversals.

const g = new Glob(pattern: string | string[], options: GlobOptions)

Options object is required.

See full options descriptions below.

[!NOTE] A previous Glob object can be passed as the GlobOptions to another Glob instantiation to re-use settings and caches with a new pattern.

Traversal functions can be called multiple times to run the walk again.

g.stream()

Stream results asynchronously.

g.streamSync()

Stream results synchronously.

g.iterate()

Default async iteration function. Returns an AsyncGenerator that iterates over the results.

g.iterateSync()

Default sync iteration function. Returns a Generator that iterates over the results.

g.walk()

Returns a Promise that resolves to the results array.

g.walkSync()

Returns a results array.

Properties

All options are stored as properties on the Glob object.

  • opts The options provided to the constructor.
  • patterns An array of parsed immutable Pattern objects.

Options

Exported as GlobOptions TypeScript interface. A GlobOptions object may be provided to any of the exported methods, and must be provided to the Glob constructor.

All options are optional, boolean, and false by default, unless otherwise noted.

All resolved options are added to the Glob object as properties.

If you are running many glob operations, you can pass a Glob object as the options argument to a subsequent operation to share the previously loaded cache.

  • cwd String path or file:// string or URL object. The current working directory in which to search. Defaults to process.cwd(). See also: "Windows, CWDs, Drive Letters, and UNC Paths", below.

    This option may be either a string path or a file:// URL object or string.

  • root A string path resolved against the cwd option, which is used as the starting point for absolute patterns that start with /, (but not drive letters or UNC paths on Windows).

    To start absolute and non-absolute patterns in the same path, you can use {root:''}. However, be aware that on Windows systems, a pattern like x:/* or //host/share/* will always start in the x:/ or //host/share directory, regardless of the root setting.

[!NOTE] This doesn't necessarily limit the walk to the root directory, and doesn't affect the cwd starting point for non-absolute patterns. A pattern containing .. will still be able to traverse out of the root directory, if it is not an actual root directory on the filesystem, and any non-absolute patterns will be matched in the cwd. For example, the pattern /../* with {root:'/some/path'} will return all files in /some, not all files in /some/path. The pattern * with {root:'/some/path'} will return all the entries in the cwd, not the entries in /some/path.

  • windowsPathsNoEscape Use \\ as a path separator only, and never as an escape character. If set, all \\ characters are replaced with / in the pattern.

[!NOTE] This makes it impossible to match against paths containing literal glob pattern characters, but allows matching with patterns constructed using path.join() and path.resolve() on Windows platforms, mimicking the (buggy!) behavior of Glob v7 and before on Windows. Please use with caution, and be mindful of the caveat below about Windows paths. (For legacy reasons, this is also set if allowWindowsEscape is set to the exact value false.)

  • dot Include .dot files in normal matches and globstar matches. Note that an explicit dot in a portion of the pattern will always match dot files.

  • magicalBraces Treat brace expansion like {a,b} as a "magic" pattern. Has no effect if {@link nobrace} is set.

    Only has effect on the {@link hasMagic} function, no effect on glob pattern matching itself.

  • dotRelative Prepend all relative path strings with ./ (or .\ on Windows).

    Without this option, returned relative paths are "bare", so instead of returning './foo/bar', they are returned as 'foo/bar'.

    Relative patterns starting with '../' are not prepended with ./, even if this option is set.

  • mark Add a / character to directory matches. Note that this requires additional stat calls.

  • nobrace Do not expand {a,b} and {1..3} brace sets.

  • noglobstar Do not match ** against multiple filenames. (Ie, treat it as a normal * instead.)

  • noext Do not match "extglob" patterns such as +(a|b).

  • nocase Perform a case-insensitive match. This defaults to true on macOS and Windows systems, and false on all others.

[!NOTE] nocase should only be explicitly set when it is known that the filesystem's case sensitivity differs from the platform default. If set true on case-sensitive file systems, or false on case-insensitive file systems, then the walk may return more or less results than expected.

As a shortcut to avoid excessive RegExp creations, Glob will use string portions as-is to readdir() calls while doing its traversal. If you are setting a nocase: true match on a file system that is in fact case sensitive, then this will result in matches not being found that you might expect, because for example the pattern Foo/* will fail to read the FOO/ or foo/ directories.

On the other hand, if you set nocase: false on a case-insensitive system, then the opposite problem occurs: Foo/* will match foo/bar, but because we only detect the existence of the foo/ folder by successfully performing a readdir, there's no way to know what the "real" case is, and the match will be reported as Foo/bar, using the case of the string portion of the glob pattern.

The default is usually correct, however it is possible to mount file systems with a different case-sensitivity from the host system. If you know this is the case, set this flag appropriately to the file system you are searching.

  • maxDepth Specify a number to limit the depth of the directory traversal to this many levels below the cwd.

  • matchBase Perform a basename-only match if the pattern does not contain any slash characters. That is, *.js would be treated as equivalent to **/*.js, matching all js files in all directories.

  • nodir Do not match directories, only files. (Note: to match only directories, put a / at the end of the pattern.)

[!NOTE] When follow and nodir are both set, then symbolic links to directories are also omitted.

  • stat Call lstat() on all entries, whether required or not to determine whether it's a valid match. When used with withFileTypes, this means that matches will include data such as modified time, permissions, and so on. Note that this will incur a performance cost due to the added system calls.

  • ignore string or string[], or an object with ignored and childrenIgnored methods.

    If a string or string[] is provided, then this is treated as a glob pattern or array of glob patterns to exclude from matches. To ignore all children within a directory, as well as the entry itself, append '/**' to the ignore pattern.

    If an object is provided that has ignored(path) and/or childrenIgnored(path) methods, then these methods will be called to determine whether any Path is a match or if its children should be traversed, respectively.

    The path argument to the methods will be a path-scurry Path object, which extends fs.Dirent with additional useful methods like .fullpath(), .relative(), and more.

[!NOTE] ignore patterns are always in dot:true mode, regardless of any other settings.

  • follow Follow symlinked directories when expanding ** patterns. This can result in a lot of duplicate references in the presence of cyclic links, and make performance quite bad.

    By default, a ** in a pattern will follow 1 symbolic link if it is not the first item in the pattern, or none if it is the first item in the pattern, following the same behavior as Bash.

[!NOTE] When follow and nodir are both set, then symbolic links to directories are also omitted.

  • realpath Set to true to call fs.realpath on all of the results. In the case of an entry that cannot be resolved, the entry is omitted. This incurs a slight performance penalty, of course, because of the added system calls.

  • absolute Set to true to always receive absolute paths for matched files. Set to false to always receive relative paths for matched files.

    By default, when this option is not set, absolute paths are returned for patterns that are absolute, and otherwise paths are returned that are relative to the cwd setting.

    This does not make an extra system call to get the realpath, it only does string path resolution.

    absolute may not be used along with withFileTypes.

  • posix Set to true to use / as the path separator in returned results. On POSIX systems, this has no effect. On Windows systems, this will return / delimited path results, and absolute paths will be returned in their fully resolved UNC path form, e.g. instead of 'C:\\foo\\bar', it will return //?/C:/foo/bar.

  • platform Defaults to the value of process.platform if available, or 'linux' if not. Setting platform:'win32' on non-Windows systems may cause strange behavior.

  • withFileTypes Return path-scurry Path objects instead of strings. These are similar to a NodeJS fs.Dirent object, but with additional methods and properties.

    withFileTypes may not be used along with absolute.

  • signal An AbortSignal which will cancel the Glob walk when triggered.

  • fs An override object to pass in custom filesystem methods. See path-scurry docs for what can be overridden.

  • scurry A PathScurry object used to traverse the file system. If the nocase option is set explicitly, then any provided scurry object must match this setting.

  • includeChildMatches boolean, default true. Do not match any children of any matches. For example, the pattern **\/foo would match a/foo, but not a/foo/b/foo in this mode.

    This is especially useful for cases like "find all node_modules folders, but not the ones in node_modules".

    In order to support this, the Ignore implementation must support an add(pattern: string) method. If using the default Ignore class, then this is fine, but if this is set to false, and a custom Ignore is provided that does not have an add() method, then it will throw an error.

    For example:

    const results = await glob(
      [
        // likely to match first, since it's just a stat
        'a/b/c/d/e/f',
    
        // this pattern is more complicated! It must to various readdir()
        // calls and test the results against a regular expression, and that
        // is certainly going to take a little bit longer.
        //
        // So, later on, it encounters a match at 'a/b/c/d/e', but it's too
        // late to ignore a/b/c/d/e/f, because it's already been emitted.
        'a/[bdf]/?/[a-z]/*',
      ],
      { includeChildMatches: false },
    )
    

    It's best to only set this to false if you can be reasonably sure that no components of the pattern will potentially match one another's file system descendants, or if the occasional included child entry will not cause problems.

[!NOTE] It only ignores matches that would be a descendant of a previous match, and only if that descendant is matched after the ancestor is encountered. Since the file system walk happens in indeterminate order, it's possible that a match will already be added before its ancestor, if multiple or braced patterns are used.

  • braceExpandMax number, defaults to 10_000. This is the maximum number of {x,y,...} patterns to expand. It is very unlikely that you'll need more than this, and setting it higher exposes the system to out-of-memory errors.

Glob Primer

Much more information about glob pattern expansion can be found by running man bash and searching for Pattern Matching.

"Globs" are the patterns you type when you do stuff like ls *.js on the command line, or put build/* in a .gitignore file.

Before parsing the path part patterns, braced sections are expanded into a set. Braced sections start with { and end with }, with 2 or more comma-delimited sections within. Braced sections may contain slash characters, so a{/b/c,bcd} would expand into a/b/c and abcd.

The following characters have special magic meaning when used in a path portion. With the exception of **, none of these match path separators (ie, / on all platforms, and \ on Windows).

  • * Matches 0 or more characters in a single path portion. When alone in a path portion, it must match at least 1 character. If dot:true is not specified, then * will not match against a . character at the start of a path portion.
  • ? Matches 1 character. If dot:true is not specified, then ? will not match against a . character at the start of a path portion.
  • [...] Matches a range of characters, similar to a RegExp range. If the first character of the range is ! or ^ then it matches any character not in the range. If the first character is ], then it will be considered the same as \], rather than the end of the character class.
  • !(pattern|pattern|pattern) Matches anything that does not match any of the patterns provided. May not contain / characters. Similar to *, if alone in a path portion, then the path portion must have at least one character.
  • ?(pattern|pattern|pattern) Matches zero or one occurrence of the patterns provided. May not contain / characters.
  • +(pattern|pattern|pattern) Matches one or more occurrences of the patterns provided. May not contain / characters.
  • *(a|b|c) Matches zero or more occurrences of the patterns provided. May not contain / characters.
  • @(pattern|pat*|pat?erN) Matches exactly one of the patterns provided. May not contain / characters.
  • ** If a "globstar" is alone in a path portion, then it matches zero or more directories and subdirectories searching for matches. It does not crawl symlinked directories, unless {follow:true} is passed in the options object. A pattern like a/b/** will only match a/b if it is a directory. Follows 1 symbolic link if not the first item in the pattern, or 0 if it is the first item, unless follow:true is set, in which case it follows all symbolic links.

[:class:] patterns are supported by this implementation, but [=c=] and [.symbol.] style class patterns are not.

Dots

If a file or directory path portion has a . as the first character, then it will not match any glob pattern unless that pattern's corresponding path part also has a . as its first character.

For example, the pattern a/.*/c would match the file at a/.b/c. However the pattern a/*/c would not, because * does not start with a dot character.

You can make glob treat dots as normal characters by setting dot:true in the options.

Basename Matching

If you set matchBase:true in the options, and the pattern has no slashes in it, then it will seek for any file anywhere in the tree with a matching basename. For example, *.js would match test/simple/basic.js.

Empty Sets

If no matching files are found, then an empty array is returned. This differs from the shell, where the pattern itself is returned. For example:

$ echo a*s*d*f
a*s*d*f

Comparisons to other fnmatch/glob implementations

While strict compliance with the existing standards is a worthwhile goal, some discrepancies exist between node-glob and other implementations, and are intentional.

The double-star character ** is supported by default, unless the noglobstar flag is set. This is supported in the manner of bsdglob and bash 5, where ** only has special significance if it is the only thing in a path part. That is, a/**/b will match a/x/y/b, but a/**b will not.

[!NOTE] Symlinked directories are not traversed as part of a **, though their contents may match against subsequent portions of the pattern. This prevents infinite loops and duplicates and the like. You can force glob to traverse symlinks with ** by setting {follow:true} in the options.

There is no equivalent of the nonull option. A pattern that does not find any matches simply resolves to nothing. (An empty array, immediately ended stream, etc.)

If brace expansion is not disabled, then it is performed before any other interpretation of the glob pattern. Thus, a pattern like +(a|{b),c)}, which would not be valid in bash or zsh, is expanded first into the set of +(a|b) and +(a|c), and those patterns are checked for validity. Since those two are valid, matching proceeds.

The character class patterns [:class:] (POSIX standard named classes) style class patterns are supported and Unicode-aware, but [=c=] (locale-specific character collation weight), and [.symbol.] (collating symbol), are not.

Repeated Slashes

Unlike Bash and zsh, repeated / are always coalesced into a single path separator.

Comments and Negation

Previously, this module let you mark a pattern as a "comment" if it started with a # character, or a "negated" pattern if it started with a ! character.

These options were deprecated in version 5, and removed in version 6.

To specify things that should not match, use the ignore option.

Windows

Please only use forward-slashes in glob expressions.

Though Windows uses either / or \ as its path separator, only / characters are used by this glob implementation. You must use forward-slashes only in glob expressions. Back-slashes will always be interpreted as escape characters, not path separators.

Results from absolute patterns such as /foo/* are mounted onto the root setting using path.join. On Windows, this will by default result in /foo/* matching C:\foo\bar.txt.

To automatically coerce all \ characters to / in pattern strings, thus making it impossible to escape literal glob characters, you may set the windowsPathsNoEscape option to true.

Windows, CWDs, Drive Letters, and UNC Paths

On POSIX systems, when a pattern starts with /, any cwd option is ignored, and the traversal starts at /, plus any non-magic path portions specified in the pattern.

On Windows systems, the behavior is similar, but the concept of an "absolute path" is somewhat more involved.

UNC Paths

A UNC path may be used as the start of a pattern on Windows platforms. For example, a pattern like: //?/x:/* will return all file entries in the root of the x: drive. A pattern like //ComputerName/Share/* will return all files in the associated share.

UNC path roots are always compared case insensitively.

Drive Letters

A pattern starting with a drive letter, like c:/*, will search in that drive, regardless of any cwd option provided.

If the pattern starts with /, and is not a UNC path, and there is an explicit cwd option set with a drive letter, then the drive letter in the cwd is used as the root of the directory traversal.

For example, glob('/tmp', { cwd: 'c:/any/thing' }) will return ['c:/tmp'] as the result.

If an explicit cwd option is not provided, and the pattern starts with /, then the traversal will run on the root of the drive provided as the cwd option. (That is, it is the result of path.resolve('/').)

Race Conditions

Glob searching, by its very nature, is susceptible to race conditions, since it relies on directory walking.

As a result, it is possible that a file that exists when glob looks for it may have been deleted or modified by the time it returns the result.

By design, this implementation caches all readdir calls that it makes, in order to cut down on system overhead. However, this also makes it even more susceptible to races, especially if the cache object is reused between glob calls.

Users are thus advised not to use a glob result as a guarantee of filesystem state in the face of rapid changes. For the vast majority of operations, this is never a problem.

See Also:

Glob Logo

Glob's logo was created by Tanya Brassie. Logo files can be found here.

The logo is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

Contributing

Any change to behavior (including bugfixes) must come with a test.

Patches that fail tests or reduce performance will be rejected.

# to run tests
npm test

# to re-generate test fixtures
npm run test-regen

# run the benchmarks
npm run bench

# to profile javascript
npm run prof

Comparison to Other JavaScript Glob Implementations

tl;dr

  • If you want glob matching that is as faithful as possible to Bash pattern expansion semantics, and as fast as possible within that constraint, use this module.
  • If you are reasonably sure that the patterns you will encounter are relatively simple, and want the absolutely fastest glob matcher out there, use fast-glob.
  • If you are reasonably sure that the patterns you will encounter are relatively simple, and want the convenience of automatically respecting .gitignore files, use globby.

There are some other glob matcher libraries on npm, but these three are (in my opinion, as of 2023) the best.


full explanation

Every library reflects a set of opinions and priorities in the trade-offs it makes. Other than this library, I can personally recommend both globby and fast-glob, though they differ in their benefits and drawbacks.

Both have very nice APIs and are reasonably fast.

fast-glob is, as far as I am aware, the fastest glob implementation in JavaScript today. However, there are many cases where the choices that fast-glob makes in pursuit of speed mean that its results differ from the results returned by Bash and other sh-like shells, which may be surprising.

In my testing, fast-glob is around 10-20% faster than this module when walking over 200k files nested 4 directories deep1. However, there are some inconsistencies with Bash matching behavior that this module does not suffer from:

  • ** only matches files, not directories
  • .. path portions are not handled unless they appear at the start of the pattern
  • ./!(<pattern>) will not match any files that start with <pattern>, even if they do not match <pattern>. For example, !(9).txt will not match 9999.txt.
  • Some brace patterns in the middle of a pattern will result in failing to find certain matches.
  • Extglob patterns are allowed to contain / characters.

Globby exhibits all of the same pattern semantics as fast-glob, (as it is a wrapper around fast-glob) and is slightly slower than node-glob (by about 10-20% in the benchmark test set, or in other words, anywhere from 20-50% slower than fast-glob). However, it adds some API conveniences that may be worth the costs.

  • Support for .gitignore and other ignore files.
  • Support for negated globs (ie, patterns starting with ! rather than using a separate ignore option).

The priority of this module is "correctness" in the sense of performing a glob pattern expansion as faithfully as possible to the behavior of Bash and other sh-like shells, with as much speed as possible.

[!NOTE] Prior versions of node-glob are not on this list. Former versions of this module are far too slow for any cases where performance matters at all, and were designed with APIs that are extremely dated by current JavaScript standards.


[1]: In the cases where this module returns results and fast-glob doesn't, it's even faster, of course.

lumpy space princess saying 'oh my GLOB'

Benchmark Results

The first number is time, smaller is better.

The second number is the count of results returned.

--- pattern: '**' ---
~~ sync ~~
node fast-glob sync             0m0.598s  200364
node globby sync                0m0.765s  200364
node current globSync mjs       0m0.683s  222656
node current glob syncStream    0m0.649s  222656
~~ async ~~
node fast-glob async            0m0.350s  200364
node globby async               0m0.509s  200364
node current glob async mjs     0m0.463s  222656
node current glob stream        0m0.411s  222656

--- pattern: '**/..' ---
~~ sync ~~
node fast-glob sync             0m0.486s  0
node globby sync                0m0.769s  200364
node current globSync mjs       0m0.564s  2242
node current glob syncStream    0m0.583s  2242
~~ async ~~
node fast-glob async            0m0.283s  0
node globby async               0m0.512s  200364
node current glob async mjs     0m0.299s  2242
node current glob stream        0m0.312s  2242

--- pattern: './**/0/**/0/**/0/**/0/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.490s  10
node globby sync                0m0.517s  10
node current globSync mjs       0m0.540s  10
node current glob syncStream    0m0.550s  10
~~ async ~~
node fast-glob async            0m0.290s  10
node globby async               0m0.296s  10
node current glob async mjs     0m0.278s  10
node current glob stream        0m0.302s  10

--- pattern: './**/[01]/**/[12]/**/[23]/**/[45]/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.500s  160
node globby sync                0m0.528s  160
node current globSync mjs       0m0.556s  160
node current glob syncStream    0m0.573s  160
~~ async ~~
node fast-glob async            0m0.283s  160
node globby async               0m0.301s  160
node current glob async mjs     0m0.306s  160
node current glob stream        0m0.322s  160

--- pattern: './**/0/**/0/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.502s  5230
node globby sync                0m0.527s  5230
node current globSync mjs       0m0.544s  5230
node current glob syncStream    0m0.557s  5230
~~ async ~~
node fast-glob async            0m0.285s  5230
node globby async               0m0.305s  5230
node current glob async mjs     0m0.304s  5230
node current glob stream        0m0.310s  5230

--- pattern: '**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.580s  200023
node globby sync                0m0.771s  200023
node current globSync mjs       0m0.685s  200023
node current glob syncStream    0m0.649s  200023
~~ async ~~
node fast-glob async            0m0.349s  200023
node globby async               0m0.509s  200023
node current glob async mjs     0m0.427s  200023
node current glob stream        0m0.388s  200023

--- pattern: '{**/*.txt,**/?/**/*.txt,**/?/**/?/**/*.txt,**/?/**/?/**/?/**/*.txt,**/?/**/?/**/?/**/?/**/*.txt}' ---
~~ sync ~~
node fast-glob sync             0m0.589s  200023
node globby sync                0m0.771s  200023
node current globSync mjs       0m0.716s  200023
node current glob syncStream    0m0.684s  200023
~~ async ~~
node fast-glob async            0m0.351s  200023
node globby async               0m0.518s  200023
node current glob async mjs     0m0.462s  200023
node current glob stream        0m0.468s  200023

--- pattern: '**/5555/0000/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.496s  1000
node globby sync                0m0.519s  1000
node current globSync mjs       0m0.539s  1000
node current glob syncStream    0m0.567s  1000
~~ async ~~
node fast-glob async            0m0.285s  1000
node globby async               0m0.299s  1000
node current glob async mjs     0m0.305s  1000
node current glob stream        0m0.301s  1000

--- pattern: './**/0/**/../[01]/**/0/../**/0/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.484s  0
node globby sync                0m0.507s  0
node current globSync mjs       0m0.577s  4880
node current glob syncStream    0m0.586s  4880
~~ async ~~
node fast-glob async            0m0.280s  0
node globby async               0m0.298s  0
node current glob async mjs     0m0.327s  4880
node current glob stream        0m0.324s  4880

--- pattern: '**/????/????/????/????/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.547s  100000
node globby sync                0m0.673s  100000
node current globSync mjs       0m0.626s  100000
node current glob syncStream    0m0.618s  100000
~~ async ~~
node fast-glob async            0m0.315s  100000
node globby async               0m0.414s  100000
node current glob async mjs     0m0.366s  100000
node current glob stream        0m0.345s  100000

--- pattern: './{**/?{/**/?{/**/?{/**/?,,,,},,,,},,,,},,,}/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.588s  100000
node globby sync                0m0.670s  100000
node current globSync mjs       0m0.717s  200023
node current glob syncStream    0m0.687s  200023
~~ async ~~
node fast-glob async            0m0.343s  100000
node globby async               0m0.418s  100000
node current glob async mjs     0m0.519s  200023
node current glob stream        0m0.451s  200023

--- pattern: '**/!(0|9).txt' ---
~~ sync ~~
node fast-glob sync             0m0.573s  160023
node globby sync                0m0.731s  160023
node current globSync mjs       0m0.680s  180023
node current glob syncStream    0m0.659s  180023
~~ async ~~
node fast-glob async            0m0.345s  160023
node globby async               0m0.476s  160023
node current glob async mjs     0m0.427s  180023
node current glob stream        0m0.388s  180023

--- pattern: './{*/**/../{*/**/../{*/**/../{*/**/../{*/**,,,,},,,,},,,,},,,,},,,,}/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.483s  0
node globby sync                0m0.512s  0
node current globSync mjs       0m0.811s  200023
node current glob syncStream    0m0.773s  200023
~~ async ~~
node fast-glob async            0m0.280s  0
node globby async               0m0.299s  0
node current glob async mjs     0m0.617s  200023
node current glob stream        0m0.568s  200023

--- pattern: './*/**/../*/**/../*/**/../*/**/../*/**/../*/**/../*/**/../*/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.485s  0
node globby sync                0m0.507s  0
node current globSync mjs       0m0.759s  200023
node current glob syncStream    0m0.740s  200023
~~ async ~~
node fast-glob async            0m0.281s  0
node globby async               0m0.297s  0
node current glob async mjs     0m0.544s  200023
node current glob stream        0m0.464s  200023

--- pattern: './*/**/../*/**/../*/**/../*/**/../*/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.486s  0
node globby sync                0m0.513s  0
node current globSync mjs       0m0.734s  200023
node current glob syncStream    0m0.696s  200023
~~ async ~~
node fast-glob async            0m0.286s  0
node globby async               0m0.296s  0
node current glob async mjs     0m0.506s  200023
node current glob stream        0m0.483s  200023

--- pattern: './0/**/../1/**/../2/**/../3/**/../4/**/../5/**/../6/**/../7/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.060s  0
node globby sync                0m0.074s  0
node current globSync mjs       0m0.067s  0
node current glob syncStream    0m0.066s  0
~~ async ~~
node fast-glob async            0m0.060s  0
node globby async               0m0.075s  0
node current glob async mjs     0m0.066s  0
node current glob stream        0m0.067s  0

--- pattern: './**/?/**/?/**/?/**/?/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.568s  100000
node globby sync                0m0.651s  100000
node current globSync mjs       0m0.619s  100000
node current glob syncStream    0m0.617s  100000
~~ async ~~
node fast-glob async            0m0.332s  100000
node globby async               0m0.409s  100000
node current glob async mjs     0m0.372s  100000
node current glob stream        0m0.351s  100000

--- pattern: '**/*/**/*/**/*/**/*/**' ---
~~ sync ~~
node fast-glob sync             0m0.603s  200113
node globby sync                0m0.798s  200113
node current globSync mjs       0m0.730s  222137
node current glob syncStream    0m0.693s  222137
~~ async ~~
node fast-glob async            0m0.356s  200113
node globby async               0m0.525s  200113
node current glob async mjs     0m0.508s  222137
node current glob stream        0m0.455s  222137

--- pattern: './**/*/**/*/**/*/**/*/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.622s  200000
node globby sync                0m0.792s  200000
node current globSync mjs       0m0.722s  200000
node current glob syncStream    0m0.695s  200000
~~ async ~~
node fast-glob async            0m0.369s  200000
node globby async               0m0.527s  200000
node current glob async mjs     0m0.502s  200000
node current glob stream        0m0.481s  200000

--- pattern: '**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.588s  200023
node globby sync                0m0.771s  200023
node current globSync mjs       0m0.684s  200023
node current glob syncStream    0m0.658s  200023
~~ async ~~
node fast-glob async            0m0.352s  200023
node globby async               0m0.516s  200023
node current glob async mjs     0m0.432s  200023
node current glob stream        0m0.384s  200023

--- pattern: './**/**/**/**/**/**/**/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.589s  200023
node globby sync                0m0.766s  200023
node current globSync mjs       0m0.682s  200023
node current glob syncStream    0m0.652s  200023
~~ async ~~
node fast-glob async            0m0.352s  200023
node globby async               0m0.523s  200023
node current glob async mjs     0m0.436s  200023
node current glob stream        0m0.380s  200023

--- pattern: '**/*/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.592s  200023
node globby sync                0m0.776s  200023
node current globSync mjs       0m0.691s  200023
node current glob syncStream    0m0.659s  200023
~~ async ~~
node fast-glob async            0m0.357s  200023
node globby async               0m0.513s  200023
node current glob async mjs     0m0.471s  200023
node current glob stream        0m0.424s  200023

--- pattern: '**/*/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.585s  200023
node globby sync                0m0.766s  200023
node current globSync mjs       0m0.694s  200023
node current glob syncStream    0m0.664s  200023
~~ async ~~
node fast-glob async            0m0.350s  200023
node globby async               0m0.514s  200023
node current glob async mjs     0m0.472s  200023
node current glob stream        0m0.424s  200023

--- pattern: '**/[0-9]/**/*.txt' ---
~~ sync ~~
node fast-glob sync             0m0.544s  100000
node globby sync                0m0.636s  100000
node current globSync mjs       0m0.626s  100000
node current glob syncStream    0m0.621s  100000
~~ async ~~
node fast-glob async            0m0.322s  100000
node globby async               0m0.404s  100000
node current glob async mjs     0m0.360s  100000
node current glob stream        0m0.352s  100000