Make Host NSUnimplemented for Android

[ephemer]

It is currently impossible to load Foundation on Android 5.0 because getifaddrs is not available there.

In the absence of #if available(Android 24, *) (which would be the ideal solution here), we need to avoid compiling any mention of getifaddrs, which is what this PR aims to achieve.

[spevans]

If ResolveType.current cant be supported on Android its probably better to mark the .current value as unavailable using @available. Then the call to _resolveCurrent() can be removed on Android as well. Might also be worth making _resolveCurrent() private since it isn't used outside of this file.

Also, if functionality cant be supported, it should call NSUnsupported() not NSUnimplemented()

[spevans]

@swift-ci test

[spevans]

cc @drodriguez

[millenomi]

I hate to regress. Can we provide a dynamic shim in CF for getifaddrs for Android use only that errors out?

[millenomi]

By that I mean: that errors out if you're running on an old-enough Android OS.

[drodriguez]

It would be better to not disable it completely. I was aware of this usage, and there should be a way of fixing it. Lily proposal might be a way (even if the alternative for pre-5.0 is just something like NSUnimplemented.

In any case, if this is all the implementation, it also should include disabling the tests, or they will fail. Host functionality is not that interesting in any case (also not being available in iOS makes it a less used type than others in Foundation).

[millenomi]

For context, I've provided some guidance on what the preferred way to deal with these situations on the forums.

[ephemer]

Hi @drodriguez and @millenomi, I agree that it doesn't feel good to regress on available functionality. But I don't see a better way around this. As I outlined on the forums here I don't think it's possible to make a shim for this on Android: I'm pretty sure we need a compile-time API Level check. Otherwise we have no Foundation at all on Android 5.0, which is a much worse regression than not having Host.

I have no idea how often this functionality would be used on Android (have never used the API myself) but if it's not available on iOS I think it's reasonable for it not to be available on Android either.

We'd need to disable the Host tests on Android, yes.

And I agree that NSUnsupported() would be better than NSUnimplemented(). Even better than either of those would be an @available annotation so people don't think it'd be a good idea to try it for Android. From what I understand @available isn't available for Android yet though. In addition, it alone wouldn't suffice to fix the error we're having here - we'd still need to ensure that getifaddrs doesn't enter the symbol table for Android by using a conditional compilation.

[ephemer]

@spevans sorry, that last message about NSUnimplemented was in response to your message too. I appreciate your thoughts on this.

I agree, it'd probably be ideal to mark Host as not @available on Android, but I don't think we have this feature in Swift yet (for Android). Without that, I'd rather only hit NSUnsupported() in the cases where it really is not supported. That way we'd make it clear to other s-c-f devs where the problem lies, and decrease the amount of unavailable API to a minimum.

But I'm happy to go with your guidance on this if you have a strong opinion- I personally have never used the Host API- I just want to get Foundation working on Android 5.0 again.

[ephemer]

@millenomi looks like I can take your suggestion after all using the approach from #2228.

What do you think the best way of implementing the NSUnsupported() behaviour would be? Is something available in CoreFoundation (in C) along these lines?

[spevans]

Could getifaddrs be implemented as a weak function that just returns failure? Then shouldn't it be replaced by the version in the SDK if it exists otherwise the weak version would be used as a fallback?

[millenomi]

I'm not quite sure what the support for that is on Android. @ephemer, thank you for going a different route.

[ephemer]

@millenomi I'm just getting to making a PR now (I have very little idea what I'm doing though- as I mentioned I've never used this API before). I am just dynamically checking for getifaddrs and freeifaddrs which I should be able to figure out.

What I'm missing is a way to fatal error out from C in the NSUnsupported case. Do I just NSLog and abort, or is there a preferred way to do this in CFFoundation?

Edit: it currently looks like this:

static void _CFIfaddrsUnsupported() {
    CFLog(__kCFLogAssertion, CFSTR("get/freeifaddrs is unsupported on Android < 7.0"));
    abort();
}

[spevans]

Could you make the shim function return errno = EOPNOTSUPPif the functions don't exist and check this result in the swift code?

[Molanda]

I found an Android implementation of getifaddrs/freeifaddrs that might be helpful...

https://github.com/morristech/android-ifaddrs

However, for myself, I have just used the NSUnimplemented solution, as Android 5.0 support is much more important to me than a working address lookup in Host.

[shahmishal]

The Swift project moved the default branch to main and deleted master branch, so GitHub automatically closed the PR. Please re-create pull request with main branch.

More detail about the branch update - https://forums.swift.org/t/updating-branch-names/40412