[Android] Use CFFIXED_USER_HOME for bundle path

[Molanda]

For Android, the main bundle path is set to the executable directory, which is something like /system/bin/app_process32. As a result, things like localized strings do not work.

This PR allows the bundle path to be set by the CFFIXED_USER_HOME environmental variable.

In my case, I set this in the Kotlin (or Java) initialization code as follows...

Os.setenv("CFFIXED_USER_HOME", mainActivity.filesDir.path, true)

I then copy the files from the asset manager into the files directory. This includes Info.plist which configures the bundle identifier and development region, and en.lproj/Localizable.strings which contain the English strings.

Once these are in place, I can call NSLocalizedString to get the localized string.

[Molanda]

Courtesy ping.

[Molanda]

Any comments on this one? It should be consistent with other areas configured for Android.

[drodriguez]

I never seen the /system/bin/app_process32 when starting executables directly. Does that only happen when the Swift code is embedded into an apk and the code is used through JNI?

@swift-ci please test

[Molanda]

Here is a good description from StackExchange:

However, for all the android apps, that's just /system/bin/app_process32 or /system/bin/app_process64 ie the same executable as the main/zygote process (itself started by init).

That process will just fork children process which will then load the java classes making up an Android app, without exec()ing any extra binary.

The command lines you see in the output of ps (/proc/PID/cmdline) are just "faked" as usual by overwriting the argv block -- see here.

This commit is written so if there is no wrapper Java/Kotlin code setting CFFIXED_USER_HOME and loading Foundation, the original code path will determine the bundle path.

[drodriguez]

Yes, that's what I imagined, that it was the original process for zygote. I don't see anything wrong in the code, and should only affect Android. The failure in macOS seems unrelated. Let's see if it goes away.

@swift-ci please test macOS platform

[Molanda]

Thank you for looking the code over and running the tests.

[drodriguez]

I think it is good to merge. I think the problem in macOS is unrelated, or I cannot see how this change can affect SwiftPM at all.