Reconsider using the `jni` crate

[wuwbobo2021]

The newly released version 0.22 of jni-rs has fixed some previously known soundness bugs and introduced binding macro usages, see https://docs.rs/jni/0.22.0/jni/macro.bind_java_type.html. It is reasonable to reconstruct java-spaghetti based on this jni crate, probably in a manner of avoiding the arbitrary_self_types nightly feature.

I think the remaining reason of not using jni directly in bluest is about the proxy feature not provided by the current jni.

[wuwbobo2021]

Well, I just discovered https://github.com/jni-rs/jbindgen. I needs some time to digest this.

[simlay]

Well, I just discovered https://github.com/jni-rs/jbindgen. I needs some time to digest this.

Oh nice. This is new! I've been throwing an occasional patch on jaffi and have been adding various features to my fork of jaffi to better support nested modules. https://simlay.net/android-bindings-rs/target/aarch64-linux-android/doc/android_bindings/android/widget/struct.TextView.html are some of the generated docs from https://github.com/simlay/android-bindings-rs

[wuwbobo2021]

Notes about why generated jni Java method bindings requires the &mut Env as the first parameter: jni-rs/jni-rs#763. This is necessary for managing multiple local reference frames, which is more efficient than dropping every local reference by JNI DeleteLocalRef (see java_spaghetti::Local's drop glue).