7.在 iOS 上使用 Python

作者:

Russell Keith-Magee (2024-03)

Python on iOS is unlike Python on desktop platforms. On a desktop platform,Python is generally installed as a system resource that can be used by any userof that computer. Users then interact with Python by running apythonexecutable and entering commands at an interactive prompt, or by running aPython script.

On iOS, there is no concept of installing as a system resource. The only unitof software distribution is an "app". There is also no console where you couldrun apython executable, or interact with a Python REPL.

As a result, the only way you can use Python on iOS is in embedded mode - thatis, by writing a native iOS application, and embedding a Python interpreterusinglibPython, and invoking Python code using thePython embeddingAPI. The full Python interpreter, the standard library, and allyour Python code is then packaged as a standalone bundle that can bedistributed via the iOS App Store.

If you're looking to experiment for the first time with writing an iOS app inPython, projects such asBeeWare andKivy will provide a much more approachable user experience.These projects manage the complexities associated with getting an iOS projectrunning, so you only need to deal with the Python code itself.

7.1.Python at runtime on iOS

7.1.1.iOS version compatibility

The minimum supported iOS version is specified at compile time, using the--host option toconfigure. By default, when compiled for iOS,Python will be compiled with a minimum supported iOS version of 13.0. To use adifferent minimum iOS version, provide the version number as part of the--host argument - for example,--host=arm64-apple-ios15.4-simulator would compile an ARM64 simulator buildwith a deployment target of 15.4.

7.1.2.Platform identification

When executing on iOS,sys.platform will report asios. This value willbe returned on an iPhone or iPad, regardless of whether the app is running onthe simulator or a physical device.

Information about the specific runtime environment, including the iOS version,device model, and whether the device is a simulator, can be obtained usingplatform.ios_ver().platform.system() will reportiOS oriPadOS, depending on the device.

os.uname() reports kernel-level details; it will report a name ofDarwin.

7.1.3.Standard library availability

The Python standard library has some notable omissions and restrictions oniOS. See theAPI availability guide for iOS fordetails.

7.1.4.Binary extension modules

One notable difference about iOS as a platform is that App Store distributionimposes hard requirements on the packaging of an application. One of theserequirements governs how binary extension modules are distributed.

The iOS App Store requires thatall binary modules in an iOS app must bedynamic libraries, contained in a framework with appropriate metadata, storedin theFrameworks folder of the packaged app. There can be only a singlebinary per framework, and there can be no executable binary material outsidetheFrameworks folder.

This conflicts with the usual Python approach for distributing binaries, whichallows a binary extension module to be loaded from any location onsys.path. To ensure compliance with App Store policies, an iOS project mustpost-process any Python packages, converting.so binary modules intoindividual standalone frameworks with appropriate metadata and signing. Fordetails on how to perform this post-processing, see the guide foraddingPython to your project.

To help Python discover binaries in their new location, the original.sofile onsys.path is replaced with a.fwork file. This file is a textfile containing the location of the framework binary, relative to the appbundle. To allow the framework to resolve back to the original location, theframework must contain a.origin file that contains the location of the.fwork file, relative to the app bundle.

For example, consider the case of an importfromfoo.barimport_whiz,where_whiz is implemented with the binary modulesources/foo/bar/_whiz.abi3.so, withsources being the locationregistered onsys.path, relative to the application bundle. This modulemust be distributed asFrameworks/foo.bar._whiz.framework/foo.bar._whiz(creating the framework name from the full import path of the module), with anInfo.plist file in the.framework directory identifying the binary as aframework. Thefoo.bar._whiz module would be represented in the originallocation with asources/foo/bar/_whiz.abi3.fwork marker file, containingthe pathFrameworks/foo.bar._whiz/foo.bar._whiz. The framework would alsocontainFrameworks/foo.bar._whiz.framework/foo.bar._whiz.origin, containingthe path to the.fwork file.

When running on iOS, the Python interpreter will install anAppleFrameworkLoader that is able to read andimport.fwork files. Once imported, the__file__ attribute of thebinary module will report as the location of the.fwork file. However, theModuleSpec for the loaded module will report theorigin as the location of the binary in the framework folder.

7.1.5.Compiler stub binaries

Xcode doesn't expose explicit compilers for iOS; instead, it uses anxcrunscript that resolves to a full compiler path (e.g.,xcrun--sdkiphoneosclang to get theclang for an iPhone device). However, using this scriptposes two problems:

  • The output ofxcrun includes paths that are machine specific, resultingin a sysconfig module that cannot be shared between users; and

  • It results inCC/CPP/LD/AR definitions that include spaces.There is a lot of C ecosystem tooling that assumes that you can split acommand line at the first space to get the path to the compiler executable;this isn't the case when usingxcrun.

To avoid these problems, Python provided stubs for these tools. These stubs areshell script wrappers around the underinglyxcrun tools, distributed in abin folder distributed alongside the compiled iOS framework. These scriptsare relocatable, and will always resolve to the appropriate local system paths.By including these scripts in the bin folder that accompanies a framework, thecontents of thesysconfig module becomes useful for end-users to compiletheir own modules. When compiling third-party Python modules for iOS, youshould ensure these stub binaries are on your path.

7.2.Installing Python on iOS

7.2.1.Tools for building iOS apps

Building for iOS requires the use of Apple's Xcode tooling. It is stronglyrecommended that you use the most recent stable release of Xcode. This willrequire the use of the most (or second-most) recently released macOS version,as Apple does not maintain Xcode for older macOS versions. The Xcode CommandLine Tools are not sufficient for iOS development; you need afull Xcodeinstall.

If you want to run your code on the iOS simulator, you'll also need to installan iOS Simulator Platform. You should be prompted to select an iOS SimulatorPlatform when you first run Xcode. Alternatively, you can add an iOS SimulatorPlatform by selecting from the Platforms tab of the Xcode Settings panel.

7.2.2.Adding Python to an iOS project

Python can be added to any iOS project, using either Swift or Objective C. Thefollowing examples will use Objective C; if you are using Swift, you may find alibrary likePythonKit to behelpful.

To add Python to an iOS Xcode project:

  1. Build or obtain a PythonXCFramework. See the instructions inApple/iOS/README.md (in the CPython source distribution) for details onhow to build a PythonXCFramework. At a minimum, you will need a buildthat supportsarm64-apple-ios, plus one of eitherarm64-apple-ios-simulator orx86_64-apple-ios-simulator.

  2. Drag theXCframework into your iOS project. In the followinginstructions, we'll assume you've dropped theXCframework into the rootof your project; however, you can use any other location that you want byadjusting paths as needed.

  3. Add your application code as a folder in your Xcode project. In thefollowing instructions, we'll assume that your user code is in a foldernamedapp in the root of your project; you can use any other location byadjusting paths as needed. Ensure that this folder is associated with yourapp target.

  4. Select the app target by selecting the root node of your Xcode project, thenthe target name in the sidebar that appears.

  5. In the "General" settings, under "Frameworks, Libraries and EmbeddedContent", addPython.xcframework, with "Embed & Sign" selected.

  6. In the "Build Settings" tab, modify the following:

    • 建置選項

      • User Script Sandboxing: No

      • Enable Testability: Yes

    • Search Paths

      • Framework Search Paths:$(PROJECT_DIR)

      • Header Search Paths:"$(BUILT_PRODUCTS_DIR)/Python.framework/Headers"

    • Apple Clang - Warnings - All languages

      • Quoted Include In Framework Header: No

  7. Add a build step that processes the Python standard library, and your ownPython binary dependencies. In the "Build Phases" tab, add a new "RunScript" build stepbefore the "Embed Frameworks" step, butafter the"Copy Bundle Resources" step. Name the step "Process Python libraries",disable the "Based on dependency analysis" checkbox, and set the scriptcontent to:

    set-esource$PROJECT_DIR/Python.xcframework/build/build_utils.shinstall_pythonPython.xcframeworkapp

    If you have placed your XCframework somewhere other than the root of yourproject, modify the path to the first argument.

  8. Add Objective C code to initialize and use a Python interpreter in embeddedmode. You should ensure that:

    Your app's bundle location can be determined using[[NSBundlemainBundle]resourcePath].

Steps 7 and 8 of these instructions assume that you have a single folder ofpure Python application code, namedapp. If you have third-party binarymodules in your app, some additional steps will be required:

  • You need to ensure that any folders containing third-party binaries areeither associated with the app target, or are explicitly copied as part ofstep 7. Step 7 should also purge any binaries that are not appropriate forthe platform a specific build is targeting (i.e., delete any device binariesif you're building an app targeting the simulator).

  • If you're using a separate folder for third-party packages, ensure thatfolder is added to the end of the call toinstall_python in step 7, andas part of thePYTHONPATH configuration in step 8.

  • If any of the folders that contain third-party packages will contain.pthfiles, you should add that folder as asite directory (usingsite.addsitedir()), rather than adding toPYTHONPATH orsys.path directly.

7.2.3.Testing a Python package

The CPython source tree containsa testbed project thatis used to run the CPython test suite on the iOS simulator. This testbed can alsobe used as a testbed project for running your Python library's test suite on iOS.

After building or obtaining an iOS XCFramework (seeApple/iOS/README.mdfor details), create a clone of the Python iOS testbed project. If you used theApple build script to build the XCframework, you can run:

$pythoncross-build/iOS/testbedclone--app<path/to/module1>--app<path/to/module2>app-testbed

Or, if you've sourced your own XCframework, by running:

$pythonApple/testbedclone--platformiOS--framework<path/to/Python.xcframework>--app<path/to/module1>--app<path/to/module2>app-testbed

Any folders specified with the--app flag will be copied into the clonedtestbed project. The resulting testbed will be created in theapp-testbedfolder. In this example, themodule1 andmodule2 would be importablemodules at runtime. If your project has additional dependencies, they can beinstalled into theapp-testbed/Testbed/app_packages folder (usingpipinstall--targetapp-testbed/Testbed/app_packages or similar).

You can then use theapp-testbed folder to run the test suite for your app,For example, ifmodule1.tests was the entry point to your test suite, youcould run:

$pythonapp-testbedrun--module1.tests

This is the equivalent of runningpython-mmodule1.tests on a desktopPython build. Any arguments after the-- will be passed to the testbed asif they were arguments topython-m on a desktop machine.

You can also open the testbed project in Xcode by running:

$openapp-testbed/iOSTestbed.xcodeproj

This will allow you to use the full Xcode suite of tools for debugging.

The arguments used to run the test suite are defined as part of the test plan.To modify the test plan, select the test plan node of the project tree (itshould be the first child of the root node), and select the "Configurations"tab. Modify the "Arguments Passed On Launch" value to change the testingarguments.

The test plan also disables parallel testing, and specifies the use of theTestbed.lldbinit file for providing configuration of the debugger. Thedefault debugger configuration disables automatic breakpoints on theSIGINT,SIGUSR1,SIGUSR2, andSIGXFSZ signals.

7.3.App Store Compliance

The only mechanism for distributing apps to third-party iOS devices is tosubmit the app to the iOS App Store; apps submitted for distribution must passApple's app review process. This process includes a set of automated validationrules that inspect the submitted application bundle for problematic code. Thereare some steps that must be taken to ensure that your app will be able to passthese validation steps.

7.3.1.Incompatible code in the standard library

The Python standard library contains some code that is known to violate theseautomated rules. While these violations appear to be false positives, Apple'sreview rules cannot be challenged; so, it is necessary to modify the Pythonstandard library for an app to pass App Store review.

The Python source tree containsa patch file that will removeall code that is known to cause issues with the App Store review process. Thispatch is applied automatically when building for iOS.

7.3.2.Privacy manifests

In April 2025, Apple introduced a requirement forcertain third-partylibraries to provide a Privacy Manifest.As a result, if you have a binary module that uses one of the affectedlibraries, you must provide an.xcprivacy file for that library.OpenSSL is one library affected by this requirement, but there are others.

If you produce a binary module namedmymodule.so, and use you the Xcodebuild script described in step 7 above, you can place amymodule.xcprivacyfile next tomymodule.so, and the privacy manifest will be installed intothe required location when the binary module is converted into a framework.