@balena/dockerignore

dockerignore

dockerignore is a file filter library fully compatible with Docker's .dockerignore file, exposing the same API as the popular ignore package for the .gitignore format.

dockerignore is also:

• Compatible with Linux, macOS and Windows.
• Compact with 0 external dependencies (production install).
• A pure Javascript port of Docker's Golang implementation of the .dockerignore format.
• Well tested with hundreds of test cases including comparison against the actual behavior of docker build on Windows and Linux.
• Relied on in production environments by balena-cli, now-cli and other prominent projects built around Docker.

What's different from ignore?

The .dockerignore spec has several differences from .gitignore:

• * in .gitignore matches everything, whereas in .dockerignore it only matches files in the current directory (like glob). This difference is important when whitelisting after a * rule.
• abc in .gitignore matches all abc files and directories, however deeply nested, whereas in .dockerignore it matches only at ./abc and not in subdirectories like ./somedir/abc.
• With .gitignore, when a parent directory is ignored, subdirectories cannot be re-added (using !) since git simply avoids walking through the subtree as an optimization. With .dockerignore, a subdirectory can be re-added even if a parent directory has been ignored.
• For a complete list of differences, check out the .gitignore spec and the .dockerignore spec.

What's the same as ignore?

The entire API. dockerignore started as a fork of node-ignore, and even reuses the same index.d.ts file for TypeScript definitions. Under the hood, node-ignore's matching logic was rewritten to closely match Docker's implementation (modeled mainly around dockerignore.go and fileutils.go).

Compatibility

dockerignore works with Node.js version 8 and above, on Linux, macOS and Windows. The code is compiled with Babel.

Install

npm install --save @balena/dockerignore

Usage

const ignore = require('@balena/dockerignore')
const ig = ignore().add(['.abc/*', '!.abc/d/'])

Typescript type definitions are also included:

import ignore from '@balena/dockerignore'
const ig = ignore().add(['.abc/*', '!.abc/d/'])

Filter the given paths

const paths = [
  '.abc/a.js',    // filtered out
  '.abc/d/e.js'   // included
]

ig.filter(paths)        // ['.abc/d/e.js']
ig.ignores('.abc/a.js') // true

As the filter function

paths.filter(ig.createFilter()); // ['.abc/d/e.js']

Windows paths are supported

ig.filter(['.abc\\a.js', '.abc\\d\\e.js'])
// if the code above runs on windows, the result will be
// ['.abc\\d\\e.js']

Backslashes vs. forward slashes

dockerignore behaves just like the Docker CLI ("docker build") in relation to the backslash (\) and forward slash (/) characters:

OS	Location	Slash (/)	Backslash (\)
Linux, macOS	.dockerignore	Path separator	Escape character
Linux, macOS	filter(), ignores()	Path separator	Part of file name
Windows	.dockerignore	Path separator	Path separator
Windows	filter(), ignores()	Path separator	Path separator

This means that forward slashes can be used in the .dockerignore file for cross-platform compatibility. This is consistent with how Windows works generally: both forward slashes and backslashes are accepted as path separators by the Command Prompt (cmd.exe) or PowerShell, and by library functions like the Golang filepath.Clean or the Node.js path.normalize.

The use of the backslash as an escape character (Linux and macOS only) is not documented in the .dockerignore specification. "Reasonable" uses are probably to escape the few characters that have a special meaning in the .dockerignore file, namely "*#!\" (excluding the double quotes), as opposed to characters that have a special meaning in regular expressions generally. The "escaping" behavior for any other characters (e.g. '\\b') is undefined and subject to implementation-specific interpretation that may change at any time.

Absolute paths

Leading and trailing slashes (or backslashes on Windows) are removed from .dockerignore patterns, so '/a', 'a/' and '/a/' are all equivalent to 'a' in a .dockerignore file, and they all anchor to the "leftmost" directory when matching against relative paths. For example, pattern 'a' is compared with 'x' for a given path 'x/y/z'. This follows Docker's Golang implementation for compatibility. Conversely, a given absolute path will not match a non-wildcard pattern. More examples:

  ignore().add('a').ignores('a')    // true
  ignore().add('/a').ignores('a')   // true
  ignore().add('/a/').ignores('a')  // true
  ignore().add('a').ignores('/a')   // false
  ignore().add('/a').ignores('/a')  // false
  ignore().add('/a/').ignores('/a') // false

Considering pattern slash removal, the cases above may be reduced to:

  ignore().add('a').ignores('a')  // true
  ignore().add('a').ignores('/a') // false

The 'false' outcome for these examples may appear to mismatch the behavior of "docker build", when the source argument for the Dockerfile ADD or COPY instructions is an absolute path (starting with a slash). The explanation is that docker converts absolute source paths to relative paths (relative to the "build context") prior to pattern matching: https://github.com/moby/moby/blob/v19.03.8/pkg/archive/archive.go#L806 https://github.com/moby/moby/blob/v19.03.8/pkg/archive/archive.go#L825

... while dockerignore mirrors the implementation of the pattern matcher itself. The advice is for your application to do the same as the docker CLI: use relative paths for pattern matching. This is also generally more portable across different environments: development machine, CI pipelines, servers or end user devices.

Comments

A pattern starting with '#' (hash) is ignored as a comment. The hash can be prefixed with a slash or backslash in order to match a file name that also starts with a hash:

ignore().add('#abc').ignores('#abc')   // false
ignore().add('/#abc').ignores('#abc')  // true
ignore().add('\\#abc').ignores('#abc') // true

This works because of the leading slash removal from patterns described in Absolute Paths.

Exclusion patterns

Patterns starting with '!' (exclamation mark) define matching exclusions (exceptions) as documented in the .dockerignore specification. For compatibility with Docker's implementation, patterns starting with '/!' or '!/' (but not '/!/') will also be considered exclusion patterns, in addition to slash removal described in Absolute Paths. Backslash escaping as '\\!' may be used in order to match a file or directory name that starts with the exclamation mark, but this is only possible on Linux and macOS, not on Windows. Again, it only behaves this way for compatibility with Docker's implementation.

Options

Matching is case-insensitive by default, following the ignore API (ignorecase). Note however that Docker performs case-sensitive matching. Use the ignorecase: false option to align with Docker's behavior:

const ig = ignore({ ignorecase: false }) // for case-sensitive matching

Methods

.add(pattern)

.add(patterns)

• pattern String|Ignore An ignore pattern string, or the Ignore instance
• patterns Array.<pattern> Array of ignore patterns.

Adds a rule or several rules to the current manager.

Returns this

pattern could either be a line of ignore pattern or a string of multiple ignore patterns, which means we could just ignore().add() the content of a ignore file:

ignore()
.add(fs.readFileSync(filenameOfGitignore).toString())
.filter(filenames)

pattern could also be an ignore instance, so that we could easily inherit the rules of another Ignore instance.

.ignores(pathname)

Returns Boolean whether pathname should be ignored.

ig.ignores('.abc/a.js')    // true

.filter(paths)

Filters the given array of pathnames, and returns the filtered array.

• paths Array.<path> The array of pathnames to be filtered.

.createFilter()

Creates a filter function which could filter an array of paths with Array.prototype.filter.

Returns function(path) the filter function.

Contributing

Contributions are always welcome!

1. Fork this repository to your own GitHub account and then clone it to your local device.
2. Install the dependencies: npm install
3. Add a test case (if applicable) and ensure it currently fails
4. Add code to pass the test
5. Make a pull request (additional tests will run on CI to ensure that your test case agrees with an actual docker build)

Acknowledgements

The initial work on this project was done by Pranay Prakash (@pranaygp) / ▲ZEIT, Kael Zhang (@kaelzhang) and the node-ignore contributors.

Paulo Castro (@pdcastro) / balena.io forked the repository in year 2020 (encouraged by Zeit) and put in a substantial effort on Windows support, cross-platform compatibility and testing, leading to release 1.0.0.