deps: land safe Dependabot bumps; triage jni & gotatun

[kasnder]

Summary

Triage of the five open Dependabot PRs (#629, #630, #631, #632, #633). Each was investigated in an isolated worktree and its fix compiled + tested. This PR carries the three that are verified green and documents the two that need manual work (left as their own Dependabot PRs).

All five were red in CI. The common thread: four of them (#629–#632) touch the shared Rust crate wgbridge-rs, and #633 was blocked by an unrelated compileSdk floor. None were safe to merge blind.

✅ Included in this PR (verified green)

#633 — gradle-minor-patch group (8 updates)

• androidx.core:core-ktx 1.18.0 → 1.19.0
• com.google.android.material 1.13.0 → 1.14.0
• androidx.lifecycle 2.10.0 → 2.11.0
• glide 5.0.7 → 5.0.9
• okhttp 5.3.2 → 5.4.0
• AGP 9.0.1 → 9.2.1, Gradle wrapper 9.2.1 → 9.6.1

Why CI failed: core-ktx:1.19.0 and glide:5.0.9 declare minCompileSdk=37 (confirmed from their AAR metadata), but the project was on compileSdk = 36, so :app:checkGithubDebugAarMetadata failed. AGP ≥9.1.0 is also required — the group's bump to 9.2.1 satisfies that.

Fix: bump compileSdk/targetSdk 36 → 37. Platform 37 is released and installed; SDK 38 is not needed (correcting an earlier assumption). targetSdk moved in lockstep with compileSdk, matching this file's existing convention — this is the one judgment call, since only compileSdk is strictly forced by the AAR metadata.

Verified: ./gradlew :app:assembleGithubDebug → BUILD SUCCESSFUL, including the native wgbridge-rs build via cargo-ndk.

Note: the checked-in gradlew/gradlew.bat/gradle-wrapper.jar bootstrap regen from Dependabot's diff is not included here — only distributionUrl was bumped. The old bootstrap correctly downloads and runs Gradle 9.6.1. If you prefer the full wrapper regen, take those files from #633.

#631 — android_logger 0.14 → 0.15

Not breaking for our single call site in jni_bindings.rs (Config::with_max_level/with_tag/init_once are unchanged in 0.15). Cargo.toml + lock only, no code change. Verified via cargo ndk arm64 cross-compile.

#630 — getrandom 0.2 → 0.3

0.3 renamed the free function getrandom::getrandom() → getrandom::fill(). Updated the single call site in keys.rs (WireGuard private-key generation). getrandom 0.3.4 was already in the tree via gotatun → rand 0.9 → rand_core 0.9, so this adds no new version; 0.2.17 remains for ring/chacha20poly1305. Not blocked by any other dep. Verified cargo build + cargo test (18/18) + arm64 cross-compile.

⚠️ NOT included — need manual work (kept as separate Dependabot PRs)

#629 — jni 0.21 → 0.22 · fix exists, needs on-device test

jni 0.22 is a ground-up API redesign, not a signature tweak: JNIEnv split into EnvUnowned/Env (closure-based with_env upgrade + ErrorPolicy), closure-based thread attachment (attach_current_thread_permanently removed), GlobalRef → Global<T>, call_method now takes compile-time jni_str!/jni_sig!, and jboolean changed from u8 to bool.

A full rewrite of jni_bindings.rs was produced that compiles cleanly for all four ABIs and preserves the exported FFI symbols (verified with llvm-nm against the Kotlin native declarations). But it changes runtime semantics (exception propagation via ThrowRuntimeExAndDefault, worker thread attach/detach lifecycle) on the most safety-critical FFI path. Recommendation: review + on-device test (tunnel start/stop, key gen, DNS recording) before merging. Too large for a rubber-stamp. The ready fix can be dropped in when someone can validate on a device.

#632 — gotatun 0.7.2 → 0.8.0 · defer / file upstream

Root cause: 0.8.0 made UdpSocket's inner an opaque enum (Socket / DisabledIpv6) to support IPv4-only fallback, and removed the public impl AsFd for UdpSocket. There is now no public API to obtain the raw fd that our code hands to Android's VpnService.protect(fd) — this is our loop-prevention / socket-protection path.

A workaround compiles (bind our own tokio::net::UdpSocket, protect it, implement gotatun's UdpSend/UdpRecv on it), but it swaps gotatun's optimized Linux datapath (sendmmsg/recvmmsg + UDP GRO) for per-packet send/recv — more syscalls/wakeups under load, which is a direct throughput and battery/radio-wakeup concern for this app. Recommendation: don't merge as-is. Preferred path: file an upstream request for a public fallible fd accessor (e.g. UdpSocket::as_raw_fd() -> Option<RawFd>) so the optimized datapath is retained, and defer #632 until then. gotatun also migrated internal logging log → tracing (lock churn only) and added DeviceBuilder::suspended (additive).

Verification summary

PR	Status	Verification
#633 gradle group	✅ included	assembleGithubDebug passes (incl. native build)
#631 android_logger	✅ included	arm64 cross-compile clean
#630 getrandom	✅ included	host build + 18/18 tests + arm64 cross-compile
#629 jni	⚠️ documented	fix compiles 4 ABIs, FFI symbols match — needs on-device test
#632 gotatun	⚠️ documented	workaround compiles but regresses UDP datapath — defer

Suggested follow-up

• Merge this PR; close deps: bump the gradle-minor-patch group with 8 updates #633, deps(rust): bump android_logger from 0.14.1 to 0.15.1 in /wgbridge-rs #631, deps(rust): bump getrandom from 0.2.17 to 0.3.4 in /wgbridge-rs #630 as superseded.
• Leave deps(rust): bump jni from 0.21.1 to 0.22.4 in /wgbridge-rs #629 open until someone can on-device test the jni rewrite.
• Leave deps(rust): bump gotatun from 0.7.2 to 0.8.0 in /wgbridge-rs #632 open (or close with a note) pending an upstream fd accessor.

🤖 Generated with Claude Code