Make Alamofire an optional dependency

[benasher44]

Issue Link 🔗

This fixes #32, making Alamofire optional.

Goals ⚽

• Allow using AlamofireNetworkActivityIndicator without Alamofire

Implementation Details 🚧

• This uses the new (in Swift 4.1) #if canImport to conditionally import Alamofire and setup the default notification listening behavior that is currently used by Alamofire users.
• If Alamofire cannot be imported, no notification listening is setup by default, and clients of this library can now register notifications used by other libraries for this library to listen to.

Testing Details 🔍

• Add tests for the public notification method
• Setup travis to test with and without Alamofire

[benasher44]

As discussed in #32, this will have to wait until we're ready to require Swift 4.1.

For testing "how does this library behave in an Alamofire environment," I was thinking of moving forward with using Carthage to test one environment and CocoaPods to test the other. We can pod lib lint to verify the build without Alamofire, and we can use Carthage and the existing xcodeproj to verify the build with Alamofire. Thoughts? I think the Cartfile will need to be moved into a subdirectory made private, so that Carthage users stop automatically picking up the Alamofire dependency. Is that accurate? I'm less familiar with Carthage. This seems to be the case after reviewing Carthage documentation.

[benasher44]

I've also updated the version number. I believe this is a backwards incompatible change: users that pulled in this as a dependency and were taking advantage of its Alamofire dependency (as a result) will no longer get that benefit.

[benasher44]

This approach gathers the separation of Alamofire symbols into one place.

[benasher44]

This function works particularly well with AFNetworking; you can hook this library up to AFNetworking in one call.

[benasher44]

Looks like Xcode 9.3 has known problems with running tests with iOS 8.4: http://www.openradar.me/39335367

[benasher44]

After attempting to write tests for this, it has become clear that accomplishing the goal of making Alamofire optional for a single file adds a lot more complexity than makes sense to try to support. The issue boils down to this: in order to ensure that if #canImport(Alamofire) works properly, you have to be sure that Alamofire is compiled before this library.

The only way to really ensure that is to declare 2 modes of integration for each dependency manager: 1 podspec/Cartfile that is explicit about the Alamofire dependency and 1 that declare no dependency. For Carthage, that may not even be possible to do in one repo; because Carthage picks up the Cartfile at the root of the repo, maintaining a separate fork (forgive my lack of Carthage knowledge if there's something I missed there) may be required.

The actual lines of Swift here seem simple enough, but I think the maintenance burden this adds is too high. I'm going to close this PR and issue. Thanks for entertaining this!