The correct solution here would have been a system-level, open source, provider-agnostic push notification API built into AOSP. The Web Push API is a great example of how a push API can be provider-agnostic: https://developer.mozilla.org/en-US/docs/Web/API/Push_API. It makes sense that push notifications should be consolidated to one provider -- this way, your device only needs to maintain one 24/7 TCP connection to a server which it receives all notifications from all apps through, rather than having each app run its own notification service in the background, killing your data and battery life.
Google's push service works by providing an HTTP endpoint to apps that they can use on the backend to deliver notifications. The Web Push API also works this way, but the returned endpoint can be for any provider. If you use Chrome, the domain will be `google.com`, but on Firefox, it's `mozilla.org`. But each provider's endpoint uses the same standardized API, so the backend doesn't have to care what the URL is. And this isn't a security risk, because everything sent through the service is encrypted.
This could have been done for Android, but that would have given Google much less control over the platform, so they decided to do it in a monopolistic way. This is one of the many ways Google aims to maintain their monopolistic control over "open source" Android. Another example is SafetyNet hardware-backed attestation.
Projects like GrapheneOS are really interesting because they are finally providing a real secure, private, de-Googled OS option, with excellent app compatibility thanks to sandboxed Play Services (which allows you to run Play Services without giving it root access to your device [1]). But will we ever be able to fully decouple Google from Android? I'm not sure. I expect most users of custom ROMs to continue installing Play Services on top of them for a long time, if they want something as basic as push notifications to work.
By the way, while I personally have qualms with Google, I believe users should have the choice to use Google apps, if they are comfortable with the inherit privacy risks. This belief is shared by the GrapheneOS developers. [2] My problem with Android is that it was not built to function without Play Services. Community projects that enable it to do so, like microG, will always be a cat-and-mouse game with Google. I think sandboxed Play Services is the most sane approach the problem of Play dependencies the community has come up with so far, but I think if we want real change to happen, we need to target app developers. Each developer has the choice of whether to include Google or not within their app. We just need to convince them that they don't need Google.
Google owns `developer.android.com` and points developers towards using Play Services APIs wherever possible. Perhaps the community should create their own open-source alternatives to the most commonly used APIs (e.g. push notifications) and developer documentation that provides instructions on how to switch from Google (ideally making the transition as easy as possible by emulating the Google API syntax). It would probably be possible to offer an installable platform-agnostic push notification API that developers can use. Providers could be installed as separate apps. Google could be one such provider. Perhaps we could implement a Web Push compatible API, and reverse engineer Chrome's Firebase integration to implement it as an option from the get-go, hoping other providers show up over time. We could perhaps allow users to self-host their push service as well. Modify Gotify, an open source self-hosted push server [3], to accept push notifications in the Web Push format.
In the hacker circles it seems there's two groups of people: those who think Android is a lost cause because it will always be controlled by Google, and think Linux phones are the only real alternative, and those who believe we can actually "steal" Android back from Google and make it into a true open source project. I fall into the latter, but I think a lot more work still needs to be done if we want to achieve this.
[1] https://grapheneos.org/usage#sandboxed-play-services
[2] https://grapheneos.org/features#features
> We aren't against users using Google services but it doesn't belong integrated into the OS in an invasive way. GrapheneOS won't take the shortcut of simply bundling a very incomplete and poorly secured third party reimplementation of Google services into the OS. That wouldn't ever be something users could rely upon. It will also always be chasing a moving target while offering poorer security than the real thing if the focus is on simply getting things working without great care for doing it robustly and securely.
