• Overview
• Usage
  • Creating a Package
    • Creating a Library Package
    • Creating an Executable Package
    • Creating a Macro Package
  • Defining Dependencies
  • Publishing a Package
  • Requiring System Libraries
  • Packaging Legacy Code
  • Handling Version-specific Logic
  • Editing a Package
    • Top of Tree Development
  • Resolving Versions (Package.resolved file)
  • Setting the Swift Tools Version
  • Testing
  • Running
  • Setting the Build Configuration
    • Debug
    • Release
    • Additional Flags
  • Depending on Apple Modules
  • Creating C Language Targets
  • Using Shell Completion Scripts
  • Package manifest specification
  • Packages and continuous integration

Simply put: a package is a git repository with semantically versioned tags, that contains Swift sources and a Package.swift manifest file at its root.

A library package contains code which other packages can use and depend on. To get started, create a directory and run swift package init:

$ mkdir MyPackage
$ cd MyPackage
$ swift package init # or swift package init --type library
$ swift build
$ swift test

This will create the directory structure needed for a library package with a target and the corresponding test target to write unit tests. A library package can contain multiple targets as explained in Target Format Reference.

SwiftPM can create native binaries which can be executed from the command line. To get started:

$ mkdir MyExecutable
$ cd MyExecutable
$ swift package init --type executable
$ swift build
$ swift run
Hello, World!

This creates the directory structure needed for executable targets. Any target can be turned into a executable target if there is a main.swift file present in its sources. The complete reference for layout is here.

SwiftPM can generate boilerplate for custom macros:

$ mkdir MyMacro
$ cd MyMacro
$ swift package init --type macro
$ swift build
$ swift run
The value 42 was produced by the code "a + b"

This creates a package with a .macro type target with its required dependencies on swift-syntax, a library .target containing the macro's code, and an .executableTarget and .testTarget for running the macro. The sample macro, StringifyMacro, is documented in the Swift Evolution proposal for Expression Macros and the WWDC Write Swift macros video. See further documentation on macros in The Swift Programming Language book.

To depend on a package, define the dependency and the version in the manifest of your package, and add a product from that package as a dependency, e.g., if you want to use https://github.com/apple/example-package-playingcard as a dependency, add the GitHub URL in the dependencies of Package.swift:

import PackageDescription

let package = Package(
    name: "MyPackage",
    dependencies: [
        .package(url: "https://github.com/apple/example-package-playingcard.git", from: "3.0.4"),
    ],
    targets: [
        .target(
            name: "MyPackage",
            dependencies: ["PlayingCard"]
        ),
        .testTarget(
            name: "MyPackageTests",
            dependencies: ["MyPackage"]
        ),
    ]
)

Now you should be able to import PlayingCard in the MyPackage target.

To publish a package, create and push a semantic version tag:

$ git init
$ git add .
$ git remote add origin [github-URL]
$ git commit -m "Initial Commit"
$ git tag 1.0.0
$ git push origin master --tags

Now other packages can depend on version 1.0.0 of this package using the github url. An example of a published package can be found here: https://github.com/apple/example-package-fisheryates

You can link against system libraries using the package manager. To do so, you'll need to add a special target of type .systemLibrary, and a module.modulemap for each system library you're using.

Let's see an example of adding libgit2 as a dependency to an executable target.

Create a directory called example, and initialize it as a package that builds an executable:

$ mkdir example
$ cd example
example$ swift package init --type executable

Edit the Sources/example/main.swift so it consists of this code:

import Clibgit

let options = git_repository_init_options()
print(options)

To import Clibgit, the package manager requires that the libgit2 library has been installed by a system packager (eg. apt, brew, yum, nuget, etc.). The following files from the libgit2 system-package are of interest:

/usr/local/lib/libgit2.dylib      # .so on Linux
/usr/local/include/git2.h

Note: the system library may be located elsewhere on your system, such as:

• /usr/, or /opt/homebrew/ if you're using Homebrew on an Apple Silicon Mac.

• C:\vcpkg\installed\x64-windows\include on Windows, if you're using vcpkg. On most Unix-like systems, you can use pkg-config to lookup where a library is installed:

  example$ pkg-config --cflags libgit2 -I/usr/local/libgit2/1.6.4/include

First, let's define the target in the package description:

// swift-tools-version: 5.8
// The swift-tools-version declares the minimum version of Swift required to build this package.

import PackageDescription

let package = Package(
    name: "example",
    targets: [
        // systemLibrary is a special type of build target that wraps a system library
        // in a target that other targets can require as their dependency.
        .systemLibrary(
            name: "Clibgit",
            pkgConfig: "libgit2",
            providers: [
                .brew(["libgit2"]),
                .apt(["libgit2-dev"])
            ]
        )
    ]
)

Note: For Windows-only packages pkgConfig should be omitted as pkg-config is not expected to be available. If you don't want to use the pkgConfig parameter you can pass the path of a directory containing the library using the -L flag in the command line when building your package instead.

example$ swift build -Xlinker -L/usr/local/lib/

Next, create a directory Sources/Clibgit in your example project, and add a module.modulemap and the header file to it:

module Clibgit [system] {
  header "git2.h"
  link "git2"
  export *
}

The header file should look like this:

// git2.h
#pragma once
#include <git2.h>

Note: Avoiding specifying an absolute path to git2.h provided by the library in the module.modulemap. Doing so will break compatibility of your project between machines that may use a different file system layout or install libraries to different paths.

The convention we hope the community will adopt is to prefix such modules with C and to camelcase the modules as per Swift module name conventions. Then the community is free to name another module simply libgit which contains more “Swifty” function wrappers around the raw C interface.

The example directory structure should look like this now:

.
├── Package.swift
└── Sources
    ├── Clibgit
    │   ├── git2.h
    │   └── module.modulemap
    └── main.swift

At this point, your system library target is fully defined, and you can now use that target as a dependency in other targets in your Package.swift, like this:

import PackageDescription

let package = Package(
    name: "example",
    targets: [
        .executableTarget(
            name: "example",

            // example executable requires "Clibgit" target as its dependency.
            // It's a systemLibrary target defined below.
            dependencies: ["Clibgit"],
            path: "Sources"
        ),

        // systemLibrary is a special type of build target that wraps a system library
        // in a target that other targets can require as their dependency.
        .systemLibrary(
            name: "Clibgit",
            pkgConfig: "libgit2",
            providers: [
                .brew(["libgit2"]),
                .apt(["libgit2-dev"])
            ]
        )
    ]
)

Now if we type swift build in our example app directory we will create an executable:

example$ swift build
…
example$ .build/debug/example
git_repository_init_options(version: 0, flags: 0, mode: 0, workdir_path: nil, description: nil, template_path: nil, initial_head: nil, origin_url: nil)
example$

Let’s see another example of using IJG’s JPEG library from an executable, which has some caveats.

Create a directory called example, and initialize it as a package that builds an executable:

$ mkdir example
$ cd example
example$ swift package init --type executable

Edit the Sources/main.swift so it consists of this code:

import CJPEG

let jpegData = jpeg_common_struct()
print(jpegData)

Install the JPEG library, on macOS you can use Homebrew package manager: brew install jpeg. jpeg is a keg-only formula, meaning it won't be linked to /usr/local/lib, and you'll have to link it manually at build time.

Just like in the previous example, run mkdir Sources/CJPEG and add the following module.modulemap:

module CJPEG [system] {
    header "shim.h"
    header "/usr/local/opt/jpeg/include/jpeglib.h"
    link "jpeg"
    export *
}

Create a shim.h file in the same directory and add #include <stdio.h> in it.

$ echo '#include <stdio.h>' > shim.h

This is because jpeglib.h is not a correct module, that is, it does not contain the required line #include <stdio.h>. Alternatively, you can add #include <stdio.h> to the top of jpeglib.h to avoid creating the shim.h file.

Now to use the CJPEG package we must declare our dependency in our example app’s Package.swift:

import PackageDescription

let package = Package(
    name: "example",
    targets: [
        .executableTarget(
            name: "example",
            dependencies: ["CJPEG"],
            path: "Sources"
            ),
        .systemLibrary(
            name: "CJPEG",
            providers: [
                .brew(["jpeg"])
            ])
    ]
)

Now if we type swift build in our example app directory we will create an executable:

example$ swift build -Xlinker -L/usr/local/jpeg/lib
…
example$ .build/debug/example
jpeg_common_struct(err: nil, mem: nil, progress: nil, client_data: nil, is_decompressor: 0, global_state: 0)
example$

We have to specify the path where the libjpeg is present using -Xlinker because there is no pkg-config file for it. We plan to provide a solution to avoid passing the flag in the command line.

Some system packages provide multiple libraries (.so and .dylib files). In such cases you should add all the libraries to that Swift modulemap package’s .modulemap file:

module CFoo [system] {
    header "/usr/local/include/foo/foo.h"
    link "foo"
    export *
}

module CFooBar [system] {
    header "/usr/include/foo/bar.h"
    link "foobar"
    export *
}

module CFooBaz [system] {
    header "/usr/include/foo/baz.h"
    link "foobaz"
    export *
}

foobar and foobaz link to foo; we don’t need to specify this information in the module-map because the headers foo/bar.h and foo/baz.h both include foo/foo.h. It is very important however that those headers do include their dependent headers, otherwise when the modules are imported into Swift the dependent modules will not get imported automatically and link errors will happen. If these link errors occur for consumers of a package that consumes your package, the link errors can be especially difficult to debug.

Module maps must contain absolute paths, thus they are not cross-platform. We intend to provide a solution for this in the package manager. In the long term, we hope that system libraries and system packagers will provide module maps and thus this component of the package manager will become redundant.

Notably the above steps will not work if you installed JPEG and JasPer with Homebrew since the files will be installed to /usr/local on Intel Macs, or /opt/homebrew on Apple silicon Macs. For now adapt the paths, but as said, we plan to support basic relocations like these.

Version the module maps semantically. The meaning of semantic version is less clear here, so use your best judgement. Do not follow the version of the system library the module map represents; version the module map(s) independently.

Follow the conventions of system packagers; for example, the debian package for python3 is called python3, as there is not a single package for python and python is designed to be installed side-by-side. Were you to make a module map for python3 you should name it CPython3.

At this time you will need to make another module map package to represent system packages that are built with optional dependencies.

For example, libarchive optionally depends on xz, which means it can be compiled with xz support, but it is not required. To provide a package that uses libarchive with xz you must make a CArchive+CXz package that depends on CXz and provides CArchive.

You may be working with code that builds both as a package and not. For example, you may be packaging a project that also builds with Xcode.

In these cases, you can use the preprocessor definition SWIFT_PACKAGE to conditionally compile code for Swift packages.

In your source file:

#if SWIFT_PACKAGE
import Foundation
#endif

The package manager is designed to support packages which work with a variety of Swift project versions, including both the language and the package manager version.

In most cases, if you want to support multiple Swift versions in a package you should do so by using the language-specific version checks available in the source code itself. However, in some circumstances this may become unmanageable, specifically, when the package manifest itself cannot be written to be Swift version agnostic (for example, because it optionally adopts new package manager features not present in older versions).

The package manager has support for a mechanism to allow Swift version-specific customizations for the both package manifest and the package versions which will be considered.

The tags which define the versions of the package available for clients to use can optionally be suffixed with a marker in the form of @swift-3. When the package manager is determining the available tags for a repository, if a version-specific marker is available which matches the current tool version, then it will only consider the versions which have the version-specific marker. Conversely, version-specific tags will be ignored by any non-matching tool version.

For example, suppose the package Foo has the tags [1.0.0, 1.2.0@swift-3, 1.3.0]. If version 3.0 of the package manager is evaluating the available versions for this repository, it will only ever consider version 1.2.0. However, version 4.0 would consider only 1.0.0 and 1.3.0.

This feature is intended for use in the following scenarios:

1. A package wishes to maintain support for Swift 3.0 in older versions, but newer versions of the package require Swift 4.0 for the manifest to be readable. Since Swift 3.0 will not know to ignore those versions, it would fail when performing dependency resolution on the package if no action is taken. In this case, the author can re-tag the last versions which supported Swift 3.0 appropriately.

2. A package wishes to maintain dual support for Swift 3.0 and Swift 4.0 at the same version numbers, but this requires substantial differences in the code. In this case, the author can maintain parallel tag sets for both versions.

It is not expected that the packages would ever use this feature unless absolutely necessary to support existing clients. Specifically, packages should not adopt this syntax for tagging versions supporting the latest GM Swift version.

The package manager supports looking for any of the following marked tags, in order of preference:

1. MAJOR.MINOR.PATCH (e.g., [email protected])
2. MAJOR.MINOR (e.g., [email protected])
3. MAJOR (e.g., 1.2.0@swift-3)

The package manager will additionally look for a version-specific marked manifest version when loading the particular version of a package, by searching for a manifest in the form of [email protected]. The set of markers looked for is the same as for version-specific tag selection.

This feature is intended for use in cases where a package wishes to maintain compatibility with multiple Swift project versions, but requires a substantively different manifest file for this to be viable (e.g., due to changes in the manifest API).

It is not expected the packages would ever use this feature unless absolutely necessary to support existing clients. Specifically, packages should not adopt this syntax for tagging versions supporting the latest GM Swift version.

In case the current Swift version doesn't match any version-specific manifest, the package manager will pick the manifest with the most compatible tools version. For example, if there are three manifests:

Package.swift (tools version 3.0) [email protected] (tools version 4.0) [email protected] (tools version 4.2)

The package manager will pick Package.swift on Swift 3, [email protected] on Swift 4, and [email protected] on Swift 4.2 and above because its tools version will be most compatible with future version of the package manager.

Swift package manager supports editing dependencies, when your work requires making a change to one of your dependencies (for example, to fix a bug, or add a new API). The package manager moves the dependency into a location under the Packages/ directory where it can be edited.

For the packages which are in the editable state, swift build will always use the exact sources in this directory to build, regardless of their state, Git repository status, tags, or the tag desired by dependency resolution. In other words, this will just build against the sources that are present. When an editable package is present, it will be used to satisfy all instances of that package in the dependency graph. It is possible to edit all, some, or none of the packages in a dependency graph, without restriction.

Editable packages are best used to do experimentation with dependency code, or to create and submit a patch in the dependency owner's repository (upstream). There are two ways to put a package in editable state:

$ swift package edit Foo --branch bugFix

This will create a branch called bugFix from the currently resolved version and put the dependency Foo in the Packages/ directory.

$ swift package edit Foo --revision 969c6a9

This is similar to the previous version, except that the Package Manager will leave the dependency at a detached HEAD on the specified revision.

Note: If the branch or revision option is not provided, the Package Manager will checkout the currently resolved version on a detached HEAD.

Once a package is in an editable state, you can navigate to the directory Packages/Foo to make changes, build and then push the changes or open a pull request to the upstream repository.

You can end editing a package using unedit command:

$ swift package unedit Foo

This will remove the edited dependency from Packages/ and put the originally resolved version back.

This command fails if there are uncommitted changes or changes which are not pushed to the remote repository. If you want to discard these changes and unedit, you can use the --force option:

$ swift package unedit Foo --force

This feature allows overriding a dependency with a local checkout on the filesystem. This checkout is completely unmanaged by the package manager and will be used as-is. The only requirement is that the package name in the overridden checkout should not change. This is extremely useful when developing multiple packages in tandem or when working on packages alongside an application.

The command to attach (or create) a local checkout is:

$ swift package edit <package name> --path <path/to/dependency>

For example, if Foo depends on Bar and you have a checkout of Bar at /workspace/bar:

foo$ swift package edit Bar --path /workspace/bar

A checkout of Bar will be created if it doesn't exist at the given path. If a checkout exists, package manager will validate the package name at the given path and attach to it.

The package manager will also create a symlink in the Packages/ directory to the checkout path.

Use unedit command to stop using the local checkout:

$ swift package unedit <package name>
# Example:
$ swift package unedit Bar

The package manager records the result of dependency resolution in a Package.resolved file in the top-level of the package, and when this file is already present in the top-level, it is used when performing dependency resolution, rather than the package manager finding the latest eligible version of each package. Running swift package update updates all dependencies to the latest eligible versions and updates the Package.resolved file accordingly.

Resolved versions will always be recorded by the package manager. Some users may choose to add the Package.resolved file to their package's .gitignore file. When this file is checked in, it allows a team to coordinate on what versions of the dependencies they should use. If this file is gitignored, each user will separately choose when to get new versions based on when they run the swift package update command, and new users will start with the latest eligible version of each dependency. Either way, for a package which is a dependency of other packages (e.g., a library package), that package's Package.resolved file will not have any effect on its client packages.

The swift package resolve command resolves the dependencies, taking into account the current version restrictions in the Package.swift manifest and Package.resolved resolved versions file, and issuing an error if the graph cannot be resolved. For packages which have previously resolved versions recorded in the Package.resolved file, the resolve command will resolve to those versions as long as they are still eligible. If the resolved version's file changes (e.g., because a teammate pushed a new version of the file) the next resolve command will update packages to match that file. After a successful resolve command, the checked out versions of all dependencies and the versions recorded in the resolved versions file will match. In most cases the resolve command will perform no changes unless the Package.swift manifest or Package.resolved file have changed.

Most SwiftPM commands will implicitly invoke the swift package resolve functionality before running, and will cancel with an error if dependencies cannot be resolved.

The tools version declares the minimum version of the Swift tools required to use the package, determines what version of the PackageDescription API should be used in the Package.swift manifest, and determines which Swift language compatibility version should be used to parse the Package.swift manifest.

When resolving package dependencies, if the version of a dependency that would normally be chosen specifies a Swift tools version which is greater than the version in use, that version of the dependency will be considered ineligible and dependency resolution will continue with evaluating the next-best version. If no version of a dependency (which otherwise meets the version requirements from the package dependency graph) supports the version of the Swift tools in use, a dependency resolution error will result.

The Swift tools version is specified by a special comment in the first line of the Package.swift manifest. To specify a tools version, a Package.swift file must begin with the string // swift-tools-version:, followed by a version number specifier.

The version number specifier follows the syntax defined by semantic versioning 2.0, with an amendment that the patch version component is optional and considered to be 0 if not specified. The semver syntax allows for an optional pre-release version component or build version component; those components will be completely ignored by the package manager currently. After the version number specifier, an optional ; character may be present; it, and anything else after it until the end of the first line, will be ignored by this version of the package manager, but is reserved for the use of future versions of the package manager.

Some Examples:

// swift-tools-version:3.1
// swift-tools-version:3.0.2
// swift-tools-version:4.0

The following Swift tools version commands are supported:

• Report tools version of the package:

    $ swift package tools-version
  

• Set the package's tools version to the version of the tools currently in use:

    $ swift package tools-version --set-current
  

• Set the tools version to a given value:

    $ swift package tools-version --set <value>
  

Use the swift test tool to run the tests of a Swift package. For more information on the test tool, run swift test --help.

Use the swift run [executable [arguments...]] tool to run an executable product of a Swift package. The executable's name is optional when running without arguments and when there is only one executable product. For more information on the run tool, run swift run --help.

SwiftPM allows two build configurations: Debug (default) and Release.

By default, running swift build will build in its debug configuration. Alternatively, you can also use swift build -c debug. The build artifacts are located in a directory called debug under the build folder. A Swift target is built with the following flags in debug mode:

• -Onone: Compile without any optimization.
• -g: Generate debug information.
• -enable-testing: Enable the Swift compiler's testability feature.

A C language target is built with the following flags in debug mode:

• -O0: Compile without any optimization.
• -g: Generate debug information.

To build in release mode, type swift build -c release. The build artifacts are located in directory named release under the build folder. A Swift target is built with following flags in release mode:

• -O: Compile with optimizations.
• -whole-module-optimization: Optimize input files (per module) together instead of individually.

A C language target is built with following flags in release mode:

• -O2: Compile with optimizations.

You can pass more flags to the C, C++, or Swift compilers in three different ways:

• Command-line flags passed to these tools: flags like -Xcc or -Xswiftc are used to pass C or Swift flags to all targets, as shown with -Xlinker above.
• Target-specific flags in the manifest: options like cSettings or swiftSettings are used for fine-grained control of compilation flags for particular targets.
• A destination JSON file: once you have a set of working command-line flags that you want applied to all targets, you can collect them in a JSON file and pass them in through extra-cc-flags and extra-swiftc-flags with --destination example.json. Take a look at Utilities/build_ubuntu_cross_compilation_toolchain for an example.

One difference is that C flags passed in the -Xcc command-line or manifest's cSettings are supplied to the Swift compiler too for convenience, but extra-cc-flags aren't.

Swift Package Manager includes a build system that can build for macOS and Linux. Xcode 11 integrates with libSwiftPM to provide support for iOS, watchOS, and tvOS platforms. To build your package with Xcode from command line you can use xcodebuild. An example invocation would be:

xcodebuild -scheme Foo -destination 'generic/platform=iOS'

where Foo would be the name of the library product you're trying to build. You can get the full list of available schemes for you SwiftPM package with xcodebuild -list. You can get the list of available destinations for a given scheme with this invocation:

xcodebuild -showdestinations -scheme Foo

C language targets are similar to Swift targets, except that the C language libraries should contain a directory named include to hold the public headers.

To allow a Swift target to import a C language target, add a target in the manifest file. Swift Package Manager will automatically generate a modulemap for each C language library target for these 3 cases:

• If include/Foo/Foo.h exists and Foo is the only directory under the include directory, and the include directory contains no header files, then include/Foo/Foo.h becomes the umbrella header.

• If include/Foo.h exists and include contains no other subdirectory, then include/Foo.h becomes the umbrella header.

• Otherwise, the include directory becomes an umbrella directory, which means that all headers under it will be included in the module.

In case of complicated include layouts or headers that are not compatible with modules, a custom module.modulemap can be provided in the include directory.

For executable targets, only one valid C language main file is allowed, e.g., it is invalid to have main.c and main.cpp in the same target.

SwiftPM ships with completion scripts for both Bash and ZSH. These files should be generated in order to use them.

Use the following commands to install the Bash completions to ~/.swift-package-complete.bash and automatically load them using your ~/.bash_profile file.

swift package completion-tool generate-bash-script > ~/.swift-package-complete.bash
echo -e "source ~/.swift-package-complete.bash\n" >> ~/.bash_profile
source ~/.swift-package-complete.bash

Alternatively, add the following commands to your ~/.bash_profile file to directly load completions:

# Source Swift completion
if [ -n "`which swift`" ]; then
    eval "`swift package completion-tool generate-bash-script`"
fi

Use the following commands to install the ZSH completions to ~/.zsh/_swift. You can chose a different folder, but the filename should be _swift. This will also add ~/.zsh to your $fpath using your ~/.zshrc file.

mkdir ~/.zsh
swift package completion-tool generate-zsh-script > ~/.zsh/_swift
echo -e "fpath=(~/.zsh \$fpath)\n" >> ~/.zshrc
compinit