Glob

Match files using the patterns the shell uses.

The safest (no ReDoS), most correct, and second fastest glob implementation in JavaScript. (See Comparison to Other JavaScript Glob Implementations at the bottom of this readme.)

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).

• normalize Normalize unicode in the pattern, and in the file names that it is matched against, with this form: 'NFC', 'NFD', 'NFKC', or 'NFKD'. true means 'NFC'. A char such as é can be one code point, or e followed by a combining accent, and some file systems (such as on macOS) can give names in either form. This is passed to minimatch, so it applies to the portions of the pattern with magic, and to the ignore patterns. The portions of the pattern without magic are looked up with path-scurry, which already compares the names of the entries that it knows in their 'NFKD' form.

• 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, Glob will use string portions as-is to readdir() calls while doing its traversal, in place of making a case-insensitive matcher for each of them. 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. Ignore patterns that start with / start at the same place as other patterns that start with / (see the root option).

  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 the pattern, 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 only match a portion of the pattern if a literal . in the pattern matches that 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.

Each portion of a pattern is matched by minimatch, with Bash 5 as the reference, and patterns are never compiled to regular expressions. These are the known differences from Bash:

• If dot is not set, a portion of a pattern can match a file name that starts with . when extglobs before a literal . match the empty string. For example, @(|x).* matches .a. Bash does not match these dotfiles.
• If an extglob such as ?(a is not closed, Bash matches the whole pattern as a literal string. In this module, only the ?( is literal, and the rest of the pattern is still matched as a glob.
• A pattern is split into portions on every /, including a / inside an extglob, so @(a|b/c)/* has an extglob that is not closed. Bash keeps the extglob whole, and drops each alternative with a / in it, as no file name can contain a /. So in Bash, @(a|b/c)/* is the same as @(a)/*.
• With nocase, case is folded with Unicode simple case folding, one code point at a time. Bash uses the case functions of the C library for the current locale.

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('/').)

ReDoS Protection

Many glob implementations make a regular expression from each pattern. A pattern such as a*a*a*a*a*a*a*a*a*a*a, tested against a name such as aaaaaaaaaaaaaaaaaaaaaaaaaaaaab that does not match, can make a backtracking regular expression engine try each way that the * characters can split the name. The time that this takes grows exponentially with the length of the name, so a pattern from an untrusted source can stop a process for minutes or more. This is a Regular Expression Denial of Service (ReDoS).

This module never makes a regular expression from a pattern. Each portion of a pattern between / characters that has magic is matched by a Matcher from @isaacs/nonrecursive-glob-match. A Matcher is not a RegExp, and it does not backtrack. It tests all of the ways that the pattern can match at the same time, in one pass over the name, so no pattern can make it try the ways one after another. The time to match a name grows no faster than its length times the size of the portion of the pattern, for each kind of pattern, including nested and negated extglobs such as +(a|*(b|!(c))). The pattern above matches or fails in about a millisecond.

The only regular expressions that a Matcher uses are for [...] classes and case folding, and each of them only tests one character at a time, so they can not backtrack over a name.

These limits also apply:

• The states that a Matcher reaches are cached, so that names are matched faster, but the cache has a fixed size. When it is full, it is cleared.
• Brace expansion makes no more than braceExpandMax patterns (10,000 by default), so a pattern such as {1..1000000} can not use an unbounded amount of memory.
• A pattern can have at most 65,536 characters. A longer pattern throws an error.
• An ignore pattern with many ** portions that are not next to each other is matched with a limit on how deeply minimatch recurses (200 by default). Past that limit, a path is not treated as a match, so that path is not ignored.

The time that a walk takes still depends on the file system. A pattern such as ** walks every directory under the cwd, so if the patterns come from an untrusted source, set the cwd to the folder that they can search, and use maxDepth and an abort signal to limit the walk. Symbolic links to directories are not followed by ** unless the follow option is set.

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:

• man sh
• man bash Pattern Matching
• man 3 fnmatch
• man 5 gitignore
• minimatch documentation

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

# run the benchmarks in a Docker container that is limited to 4
# CPUs and 8GB of memory, a few times, and print the medians
npm run bench:docker

# 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 fastest sync 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 sync 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 the benchmarks below, which walk over 200k files nested 4 directories deep, the sync fast-glob takes about 25% less time than this module, and the async fast-glob takes about 5% more, for the patterns where they return the same results1. 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). In the benchmarks below, the sync globby takes about 10% less time than this module, and the async globby takes about 15% more, for the patterns where they return the same results. It takes about 10-20% more time 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.

Benchmark Results

These are from npm run bench:docker, which runs the benchmarks in a Linux Docker container that is limited to 4 CPUs and 8GB of memory, with the test files in memory. Each time is the median of 3 runs, and includes starting node.

The first number is time, smaller is better.

The second number is the count of results returned. The totals at the end are over all of the patterns, including the ones where the implementations return different results.

--- pattern: '{0000,0,1111,1}/{0000,0,1111,1}/{0000,0,1111,1}/**' ---
~~ sync ~~
fast-glob sync           	0m0.046s  1600
globby sync              	0m0.050s  1600
fs.globSync              	0m0.040s  1776
current globSync mjs     	0m0.048s  1776
current glob syncStream  	0m0.046s  1776
~~ async ~~
fast-glob async          	0m0.047s  1600
globby async             	0m0.053s  1600
fs.glob                  	0m0.054s  1776
current glob async mjs   	0m0.050s  1776
current glob stream      	0m0.051s  1776

--- pattern: '**' ---
~~ sync ~~
fast-glob sync           	0m0.340s  200002
globby sync              	0m0.423s  200002
fs.globSync              	0m0.692s  222223
current globSync mjs     	0m0.437s  222223
current glob syncStream  	0m0.408s  222223
~~ async ~~
fast-glob async          	0m0.491s  200002
globby async             	0m0.569s  200002
fs.glob                  	0m1.889s  222223
current glob async mjs   	0m0.464s  222223
current glob stream      	0m0.442s  222223

--- pattern: '**/..' ---
~~ sync ~~
fast-glob sync           	0m0.260s  0
globby sync              	0m0.432s  200002
fs.globSync              	0m0.531s  2222
current globSync mjs     	0m0.351s  2222
current glob syncStream  	0m0.347s  2222
~~ async ~~
fast-glob async          	0m0.381s  0
globby async             	0m0.569s  200002
fs.glob                  	0m1.632s  2222
current glob async mjs   	0m0.373s  2222
current glob stream      	0m0.366s  2222

--- pattern: './**/0/**/0/**/0/**/0/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.263s  10
globby sync              	0m0.267s  10
fs.globSync              	0m0.547s  10
current globSync mjs     	0m0.335s  10
current glob syncStream  	0m0.334s  10
~~ async ~~
fast-glob async          	0m0.374s  10
globby async             	0m0.388s  10
fs.glob                  	0m1.676s  10
current glob async mjs   	0m0.342s  10
current glob stream      	0m0.340s  10

--- pattern: './**/[01]/**/[12]/**/[23]/**/[45]/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.270s  160
globby sync              	0m0.269s  160
fs.globSync              	0m0.588s  160
current globSync mjs     	0m0.344s  160
current glob syncStream  	0m0.348s  160
~~ async ~~
fast-glob async          	0m0.380s  160
globby async             	0m0.401s  160
fs.glob                  	0m1.715s  160
current glob async mjs   	0m0.346s  160
current glob stream      	0m0.345s  160

--- pattern: './**/0/**/0/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.269s  5230
globby sync              	0m0.276s  5230
fs.globSync              	0m0.555s  5230
current globSync mjs     	0m0.343s  5230
current glob syncStream  	0m0.340s  5230
~~ async ~~
fast-glob async          	0m0.381s  5230
globby async             	0m0.399s  5230
fs.glob                  	0m1.662s  5230
current glob async mjs   	0m0.345s  5230
current glob stream      	0m0.344s  5230

--- pattern: '**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.336s  200000
globby sync              	0m0.425s  200000
fs.globSync              	0m0.697s  200000
current globSync mjs     	0m0.438s  200000
current glob syncStream  	0m0.411s  200000
~~ async ~~
fast-glob async          	0m0.489s  200000
globby async             	0m0.563s  200000
fs.glob                  	0m1.866s  200000
current glob async mjs   	0m0.467s  200000
current glob stream      	0m0.429s  200000

--- pattern: '{**/*.txt,**/?/**/*.txt,**/?/**/?/**/*.txt,**/?/**/?/**/?/**/*.txt,**/?/**/?/**/?/**/?/**/*.txt}' ---
~~ sync ~~
fast-glob sync           	0m0.339s  200000
globby sync              	0m0.428s  200000
fs.globSync              	0m1.093s  200000
current globSync mjs     	0m0.463s  200000
current glob syncStream  	0m0.436s  200000
~~ async ~~
fast-glob async          	0m0.485s  200000
globby async             	0m0.570s  200000
fs.glob                  	0m2.303s  200000
current glob async mjs   	0m0.499s  200000
current glob stream      	0m0.480s  200000

--- pattern: '**/5555/0000/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.265s  1000
globby sync              	0m0.269s  1000
fs.globSync              	0m0.525s  1000
current globSync mjs     	0m0.356s  1000
current glob syncStream  	0m0.360s  1000
~~ async ~~
fast-glob async          	0m0.375s  1000
globby async             	0m0.383s  1000
fs.glob                  	0m1.592s  1000
current glob async mjs   	0m0.361s  1000
current glob stream      	0m0.366s  1000

--- pattern: './**/0/**/../[01]/**/0/../**/0/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.261s  0
globby sync              	0m0.263s  0
fs.globSync              	0m0.754s  4880
current globSync mjs     	0m0.351s  4880
current glob syncStream  	0m0.350s  4880
~~ async ~~
fast-glob async          	0m0.363s  0
globby async             	0m0.382s  0
fs.glob                  	0m1.869s  4880
current glob async mjs   	0m0.370s  4880
current glob stream      	0m0.367s  4880

--- pattern: '**/????/????/????/????/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.297s  100000
globby sync              	0m0.349s  100000
fs.globSync              	0m0.681s  100000
current globSync mjs     	0m0.404s  100000
current glob syncStream  	0m0.399s  100000
~~ async ~~
fast-glob async          	0m0.440s  100000
globby async             	0m0.484s  100000
fs.glob                  	0m1.805s  100000
current glob async mjs   	0m0.418s  100000
current glob stream      	0m0.413s  100000

--- pattern: './{**/?{/**/?{/**/?{/**/?,,,,},,,,},,,,},,,}/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.337s  200000
globby sync              	0m0.427s  200000
fs.globSync              	0m1.264s  200000
current globSync mjs     	0m0.465s  200000
current glob syncStream  	0m0.446s  200000
~~ async ~~
fast-glob async          	0m0.491s  200000
globby async             	0m0.562s  200000
fs.glob                  	0m2.508s  200000
current glob async mjs   	0m0.494s  200000
current glob stream      	0m0.473s  200000

--- pattern: '**/!(0|9).txt' ---
~~ sync ~~
fast-glob sync           	0m0.327s  160000
globby sync              	0m0.404s  160000
fs.globSync              	0m0.687s  180000
current globSync mjs     	0m0.462s  180000
current glob syncStream  	0m0.418s  180000
~~ async ~~
fast-glob async          	0m0.474s  160000
globby async             	0m0.552s  160000
fs.glob                  	0m1.876s  180000
current glob async mjs   	0m0.465s  180000
current glob stream      	0m0.436s  180000

--- pattern: './{*/**/../{*/**/../{*/**/../{*/**/../{*/**,,,,},,,,},,,,},,,,},,,,}/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.262s  0
globby sync              	0m0.264s  0
fs.globSync              	0m0.933s  200000
current globSync mjs     	0m0.557s  200000
current glob syncStream  	0m0.541s  200000
~~ async ~~
fast-glob async          	0m0.377s  0
globby async             	0m0.414s  0
fs.glob                  	0m2.135s  200000
current glob async mjs   	0m0.662s  200000
current glob stream      	0m0.639s  200000

--- pattern: './*/**/../*/**/../*/**/../*/**/../*/**/../*/**/../*/**/../*/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.256s  0
globby sync              	0m0.262s  0
fs.globSync              	0m0.716s  200000
current globSync mjs     	0m0.519s  200000
current glob syncStream  	0m0.494s  200000
~~ async ~~
fast-glob async          	0m0.363s  0
globby async             	0m0.374s  0
fs.glob                  	0m1.870s  200000
current glob async mjs   	0m0.565s  200000
current glob stream      	0m0.542s  200000

--- pattern: './*/**/../*/**/../*/**/../*/**/../*/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.259s  0
globby sync              	0m0.262s  0
fs.globSync              	0m0.725s  200000
current globSync mjs     	0m0.488s  200000
current glob syncStream  	0m0.459s  200000
~~ async ~~
fast-glob async          	0m0.366s  0
globby async             	0m0.379s  0
fs.glob                  	0m1.878s  200000
current glob async mjs   	0m0.522s  200000
current glob stream      	0m0.495s  200000

--- pattern: './0/**/../1/**/../2/**/../3/**/../4/**/../5/**/../6/**/../7/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.050s  0
globby sync              	0m0.055s  0
fs.globSync              	0m1.134s  28970
current globSync mjs     	0m0.290s  28970
current glob syncStream  	0m0.287s  28970
~~ async ~~
fast-glob async          	0m0.055s  0
globby async             	0m0.064s  0
fs.glob                  	0m1.745s  28970
current glob async mjs   	0m0.307s  28970
current glob stream      	0m0.301s  28970

--- pattern: './**/?/**/?/**/?/**/?/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.305s  100000
globby sync              	0m0.353s  100000
fs.globSync              	0m0.750s  100000
current globSync mjs     	0m0.400s  100000
current glob syncStream  	0m0.403s  100000
~~ async ~~
fast-glob async          	0m0.466s  100000
globby async             	0m0.513s  100000
fs.glob                  	0m1.900s  100000
current glob async mjs   	0m0.415s  100000
current glob stream      	0m0.412s  100000

--- pattern: '**/*/**/*/**/*/**/*/**' ---
~~ sync ~~
fast-glob sync           	0m0.371s  200000
globby sync              	0m0.470s  200000
fs.globSync              	0m0.907s  220000
current globSync mjs     	0m0.476s  220000
current glob syncStream  	0m0.452s  220000
~~ async ~~
fast-glob async          	0m0.526s  200000
globby async             	0m0.611s  200000
fs.glob                  	0m2.094s  220000
current glob async mjs   	0m0.521s  220000
current glob stream      	0m0.498s  220000

--- pattern: './**/*/**/*/**/*/**/*/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.367s  200000
globby sync              	0m0.459s  200000
fs.globSync              	0m0.911s  200000
current globSync mjs     	0m0.481s  200000
current glob syncStream  	0m0.454s  200000
~~ async ~~
fast-glob async          	0m0.519s  200000
globby async             	0m0.596s  200000
fs.glob                  	0m2.140s  200000
current glob async mjs   	0m0.508s  200000
current glob stream      	0m0.483s  200000

--- pattern: '**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.336s  200000
globby sync              	0m0.425s  200000
fs.globSync              	0m0.698s  200000
current globSync mjs     	0m0.437s  200000
current glob syncStream  	0m0.430s  200000
~~ async ~~
fast-glob async          	0m0.483s  200000
globby async             	0m0.565s  200000
fs.glob                  	0m1.874s  200000
current glob async mjs   	0m0.462s  200000
current glob stream      	0m0.468s  200000

--- pattern: './**/**/**/**/**/**/**/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.346s  200000
globby sync              	0m0.442s  200000
fs.globSync              	0m0.721s  200000
current globSync mjs     	0m0.437s  200000
current glob syncStream  	0m0.417s  200000
~~ async ~~
fast-glob async          	0m0.522s  200000
globby async             	0m0.595s  200000
fs.glob                  	0m1.892s  200000
current glob async mjs   	0m0.486s  200000
current glob stream      	0m0.468s  200000

--- pattern: '**/*/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.343s  200000
globby sync              	0m0.434s  200000
fs.globSync              	0m0.753s  200000
current globSync mjs     	0m0.448s  200000
current glob syncStream  	0m0.446s  200000
~~ async ~~
fast-glob async          	0m0.516s  200000
globby async             	0m0.576s  200000
fs.glob                  	0m1.936s  200000
current glob async mjs   	0m0.492s  200000
current glob stream      	0m0.465s  200000

--- pattern: '**/*/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.341s  200000
globby sync              	0m0.429s  200000
fs.globSync              	0m0.772s  200000
current globSync mjs     	0m0.451s  200000
current glob syncStream  	0m0.434s  200000
~~ async ~~
fast-glob async          	0m0.492s  200000
globby async             	0m0.565s  200000
fs.glob                  	0m1.927s  200000
current glob async mjs   	0m0.480s  200000
current glob stream      	0m0.447s  200000

--- pattern: '**/[0-9]/**/*.txt' ---
~~ sync ~~
fast-glob sync           	0m0.307s  100000
globby sync              	0m0.347s  100000
fs.globSync              	0m0.667s  100000
current globSync mjs     	0m0.393s  100000
current glob syncStream  	0m0.396s  100000
~~ async ~~
fast-glob async          	0m0.458s  100000
globby async             	0m0.490s  100000
fs.glob                  	0m1.894s  100000
current glob async mjs   	0m0.433s  100000
current glob stream      	0m0.426s  100000

--- total of the medians ---
fast-glob sync (sync)               	7.153s
globby sync (sync)                  	8.484s
fs.globSync (sync)                  	18.341s
current globSync mjs (sync)         	10.174s
current glob syncStream (sync)      	9.856s
fast-glob async (async)             	10.314s
globby async (async)                	11.617s
fs.glob (async)                     	45.732s
current glob async mjs (async)      	10.847s
current glob stream (async)         	10.496s