Integrating XCFrameworks into the development and release workflows - ios

I have an Xcode workspace with an iOS framework target and an iOS SampleApp target that depends on and embeds the framework. The framework is closed-source and thus gets delivered in binary form to customers, in a release package alongside a copy of the sources for the entire SampleApp project, so that they can test it out with our app and see how the integration is done.
With this setup, during the development phase, I can open the workspace in Xcode and build the SampleApp target, and because it has an explicit dependency on the framework, Xcode immediately triggers a build of the framework first and then the app is built and linked correctly.
Now I want to leverage the new XCFramework format to distribute this framework to customers instead of using lipo'ed fat frameworks. I update my shell script that produces the release package to create an xcframework bundle at the end with all the various platforms supported, and also update the SampleApp project to depend on this xcframework instead of on the individual frameworks as it did before.
But here comes the problem with the updated SampleApp: After removing the dependency on the frameworks, the workspace setup described above doesn't work anymore, i.e. Xcode will no longer automatically build the framework target when the SampleApp target is built, because the explicit dependency link between the two is now broken.
My first attempt at solving this involved making the xcframework part of my regular development workflow as well. For this I added a run-script phase to the framework build target that generates a current-platform-only xcframework bundle and copies it to the place where the SampleApp expects to find it, and then added the framework target to the targets of the SampleApp build scheme, in order for the framework to be built when the SampleApp is built. Unfortunately this approach does not work because Xcode 12 always seems to build both targets in parallel, no matter what the 'Parallelize Build' setting is set to, and because of that the app will fail to build simply because the xcframework is not yet ready when it's processed by the build system, which is very early in the build phases of the SampleApp. So unless there's a way to force Xcode to honor the order in which the build targets are defined in the scheme, this approach is a no-go.
If this approach cannot work then I'd have to ship the xcframework with the SampleApp unmodified, and instruct customers wanting to run the app to manually remove the references to the universal frameworks and then add/drag the xcframework to it. This would work but it's ugly and does not seem user friendly enough to me.
I'm also considering a different approach whereby the release script programmatically modifies the SampleApp.xcodeproj to depend on the xcframework instead of on the universal frameworks, but that seems like it could be a very fragile thing to do, so would rather avoid it if there's a better way.
This scenario does not seem unique to me so I'd guess someone else must have faced this problem too. If so, how did you solve it?

So I ended up solving this by programmatically modifying the SampleApp's project file at release time to link and embed the xcframework bundle instead of the universal frameworks.
Initially tried to use the tool pbxproj but it didn't quite work, the initial fat framework removal phase was not succeeding.
Then I figured I could try and diff the project.pbxproj file after having manually replaced the frameworks with an xcframework, and turns out it takes only a few small edits to get there, so they can be done programmatically with a simple shell script like the following:
#!/bin/bash
PROJECT_FILE=SampleApp/SampleApp.xcodeproj/project.pbxproj
FRAMEWORK_NAME=MyFramework.framework
XCFRAMEWORK_NAME=MyFramework.xcframework
sed -i '' "s|FileType = wrapper.framework; path = ${FRAMEWORK_NAME}; sourceTree = BUILT_PRODUCTS_DIR;|FileType = wrapper.xcframework; path = ${XCFRAMEWORK_NAME}; sourceTree = \"<group>\";|" ${PROJECT_FILE}
sed -i '' "s|${FRAMEWORK_NAME}|${XCFRAMEWORK_NAME}|g" ${PROJECT_FILE}
There might be small differences in other project setups but mostly it boils down to that. Just beware that if the xcframework is meant to be placed somewhere other than in the project root then you need both a name and a path properties in the first sed command, e.g.:
name = "MyFramework.xcframework"; path = "relative/path/to/MyFramework.xcframework"

Related

iOS Xcode SPM failed to demangle superclass

My app is composed of many projects (frameworks), one for each main feature and a common framework with all sorts of things that I need to access in multiple of my features.
I'm using Xcode 11's Swift Package Manager to add dependencies.
The common framework contains a RxSwift dependency, which I use throughout the whole project.
I'm facing problems when I try to use RxTest in any of my feature frameworks.
If I add RxTest via SPM to the test target directly and run the tests, I get
failed to demangle superclass of 'class name' from mangled name 'other class name'
and many
Class 'class name' is implemented in both 'common framework path' and 'test target path'
where all these classes are Rx related. The 'failed to demangle' error crashes the test and only occurs when I try to initialize a RxTest class.
If I add RxTest to the common framework, the tests run fine, but when I run the app, I get
dyld: Library not loaded: #rpath/XCTest.framework/XCTest
Which makes sense, because I'm adding a test framework to a non-test framework, and it's not something good to do.
So basically, I wasn't able to get a configuration where both the tests and the app run fine. Either the app runs or the tests run.
How can I get this working? Is there a way to include RxTest on the common framework only when I build it on a test target? Or should RxTest only be included on the test targets and I'm missing some configuration?
Xcode with SPM dependencies cannot handle same SPM dependency in multiple targets that are dependent on each other right now. Each dependency needs to be only in single target at the moment. I dont know why as of now, but I'll try investigate more and file bug if it is not filed yet.
Your issue is likely that the library is using static linking instead of dynamic linking. In SwiftPM you can specify a library as being static or dynamic if you want or you can just let the build system decide which is what most packages do. Xcode seems to favor the static approach when it builds with SwiftPM which results in the build issues you are experiencing.
If you modify the Package.swift to have RxTest be a dynamic library instead it should work. You can easily test this by cloning RxSwift and modifying this line:
.library(name: "RxTest", targets: ["RxTest"]),
into:
.library(name: "RxTest", type: .dynamic, targets: ["RxTest"]),
and then dragging the local copy of RxSwift into your Xcode Project Navigator.
It will then use your local copy of the package instead of the one cloned by Xcode.
Once you do this you can link it against any targets you need and it should work. If that does actually fix the problem then your long term solutions are likely:
1) Have a fork that simply changes it to a dynamic library.
2) Convince the RxSwift community to change their products to dynamic or to vend dynamic versions in addition to the default.
3) Don't use RxTest or similar things in multiple places.
It is also worth noting, that Xcode 11.3 and earlier do not support archiving with dynamic Swift Packages. So if you go down the dynamic route you will have to wait for Xcode 11.4.
Workaround:
I had the same issues. My project configuration is:
Workspace
ProjectAppOne
AppTargetOne (Embed and sign FrameworkTarget)
ProjectAppTwo
AppTargetTwo (Embed and sign FrameworkTarget)
ProjectCoreFramework
FrameworkTarget
testsTareget
SPM Dependencies (including RxSwift)
My workaround was:
Duplicate the FrameworkTarget and create a FrameworkTargetT (make sure that are both building with no issues).
FrameworkTarget is still used to build the applications targets, so be sure that is the only one used in the imports of the application targets.
Add RxTest to FrameworkTargetT
Remove RxTest from FrameworkTarget
Set in your tests #testable import FrameworkTagetT
Adjust the FrameworkTarget to not run the testsTarget
Set the FrameworkTargetT to run the testsTarget
Adjust your CI fastlane/scripts, that involve testing target and call the FrameworkTarget directly
At the end the project structure was looking like:
Workspace
ProjectAppOne
AppTargetOne (Embed and sign FrameworkTarget)
ProjectAppTwo
AppTargetTwo (Embed and sign FrameworkTarget)
ProjectCoreFramework
FrameworkTarget
FrameworkTargetT
testsTareget
SPM Dependencies (including RxSwift)
I do not think is the ideal solution, it is easy to import the wrong framework in code or test, so double check that you are happy with that. But while we are waiting for the SPM team to sort the problem, it could be used and removed easily. This is also easy if you have already a framework separated code.
PS: you need to remember to add every new file to both of the targets FrameworkTarget and FrameworkTargetT.

Why Am I Not Able To Embed My Custom Framework?

I have a very simple framework project, MyFramework, that builds successfully. The framework defines a single, global function, myFunction.
The framework also defines 2 global variables: MyFrameworkVersionNumber and MyFrameworkVersionString (These 2 variables were created for me by Xcode)
I have a very simple application project, MyApplication. I am adding the framework to the application project by dragging and dropping the framework package (i.e. the MyFramework.framework in DerivedData that was produced by building the framework) onto the Xcode Navigator and then selecting the framework in the Embedded Binaries section of the project's General tab.
If I add code to MyApplication that references the variable MyFrameworkVersionNumber then I am able to successfully build and run MyApplication.
If I add code to MyApplication that references the function myFunction then I am not able to build MyApplication.
First Update
I suddenly remembered that Carthage does exactly what I am trying to do. So, I used Carthage to build MyFramework and Voila! - I can drag/drop the Carthage build of the framework into the application project and successfully use it. So now my inquiry has become: What does Carthage know that I do not?
Oh for heaven's sake! It turned out to be so simple. MyFramework was being built for a Generic iOS Device (arm64 architecture). MyApplication was being built for a simulator (x86_64 architecture). As soon as I matched those two up all was well. The reason that the Carthage build worked so well is that it produces a universal binary (i.e. one that contains both architectures).
I still do not understand why the MyFrameworkVersionNumber global variable was able to be accessed regardless of the architecture. But I am okay with deferring that little mystery as some arcane bit of information that will be revealed in the goodness of time (perhaps something such as that it is in a header that is structured the same for both architectures).

How to build dynamic Framework for iOS with no Embedded Binaries?

I try to generate framework named AfTestFramework for ios. I created through Xcode Cocoa Touch Framework with some basic code, ran it and got:
So far so good.
I created new Test Application and drag # dropped AfTestFramework.framework into the project, like crashlytics does (see below)
After building project and running I got Library not loaded ... Reason: image not found:
dyld: Library not loaded:
#rpath/AfTestFramework.framework/AfTestFramework Referenced from:
/var/containers/Bundle/Application/9B12D0AD-94AA-4119-A5DE-6BCAA806FA9F/TestAvoidEmbedded1.app/TestAvoidEmbedded1 Reason: image not found
I know the solution is to attach this framework to Embedded Binaries but I look for the way to avoid this approach.
Further, It demands to create 2 Frameworks: for Simulator and for Release otherwise I cannot release application with my framework, that looks a bit messy and weird.
I found that following frameworks work without Embedded Binaries:
crashlytics
firebase
facebook
Chartboost
ans so on.
So far I didn't find any description, I know also that its something new.
I started to play with xcodebuild like:
xcodebuild -sdk "iphoneos10.1" "ARCHS=arm64 armv7 armv7s" "VALID_ARCHS=arm64 armv7 armv7s" "ONLY_ACTIVE_ARCH=NO" -target AfTestFramework -configuration "Release" clean build
but no success.
Can somebody spread the light on this problem?
If you add a dynamic Framework to your iOS project, you must add it to the Embedded Binaries. In opposite to static libraries are dynamic libraries loaded at runtime. Thus the dynamic linker must access them on the device.
The only possible way to add the code inside of the framework without embedding the framework is to bind it statically to your app. For that you should create a static library from the framework. (Probably, libtool(1) may help you with that.) But anyway you may run into other problems, if the framework needs its bundle structure for loading resources etc.
I'm not quite sure what you are trying to achieve but it sounds like you are trying to create a framework which has a dependency on another framework.
So when used, it would be something like app -> your-framework -> some-other-framework.
I would strongly recommend looking at carthage as a solution to our problem. Carthage is an OS X native dependency manager that is designed to manage dependencies for you.
Firstly you need to check that the framework your framework uses is Carthage friendly. For now I'll assume so.
In your frameworks project you would add a Cartfile where you declare your dependency on the other framework, then use Carthage to download and build the other framework. You then add the built framework to your project as you would with any Apple supplied framework. You DO NOT need to embed the framework and generally speaking it's not recommended to do so. Carthage will handle this issue.
In the app project you would then also add a Cartfile, but you only need to specify your project in it. When Carthage builds the framework, it will automatically locate the other framework, download and build it as well. Then in the apps build phases you simple specify linking agains both frameworks and add a carthage-copy script phase to include them in the Frameworks directory of the finished app.
With regard to building for simulator and release. You don't need to build multiple targets. For a start, using Carthage means that the frameworks are only built when you need them, so this removes any requirement for you to supply binaries.
When telling Carthage to build the dependencies, by default it builds Release binaries. Building Debug frameworks for testing is just a matter of add --configuration Debug to Carthage's command line arguments.
Finally - to be fair, there are also two other package managers out there: Ruby based CocoaPods which is an older manager and for Swift, the Swift Package Manager. Personally I prefer Carthage for reasons I explain in a blog post here.
this is old but if its useful for future reference:
you can do
$lipo -create <path to simulator framework> <path to device framework> -output <path to the output framework>

Force Xcode to build with specific .framework

I have a target that links with a framework (both of which I develop). If I build the target it builds the framework before hand for use in the target.
What I want during certain builds is to bring in an existing already built version of that framework and use it during the build of that target. Is there a way to force xcode to build using an already compiled version of a dependant framework
(The reason I'm doing this is because I have multiple targets that are built separately and want them all to be built using the same framework binary)
I finally solved this, this can be done by turning off the "Find Implicit Dependencies" under your build scheme.
Some good reading can be found here on the topic here: https://pewpewthespells.com/blog/managing_xcode.html#scheme-action

How to handle Debug/TestFlight/AppStore releases without targets?

I use Cocoapods for my project and I need a specific set of frameworks for each Debug/TestFlight/AppStore release:
Debug (simulator or device): common frameworks + Calabash - Google Analytics
TestFlight: common frameworks + TestFlight SDK
App Store: common frameworks only
I could use targets, but they are annoying:
Every time you add a new file (or create one) you have to remember to add it to all targets. It makes it really easy to produce builds that don't have a required file.
Changes in build settings of one target don't propagate to other targets (i.e: changing a llvm flag).
On the other hand, cocoapods doesn't support weak frameworks, so I can't weak-link to Calabash in my main target and force-load it when building in Debug mode (because it'll still be in the framework set when building for the app store)
What would be a good balance between the very segregated option that are targets, and "all in" solution of a single target?
I have a makefile that removes certain pods then runs Pod install. We're making it much easier in the next release to do config based pods, so you will be able to do all of what your asking in version 0.34+. This is in the current HEAD on github.

Resources