lightning-kmp-app moves to d87af50, which carries Kotlin 2.4.21 down to secp256k1-kmp, and engine-interop/new follows the app as its comment asks. The declared Kotlin now matches the stdlib quartz already resolved, and the resolved graphs are unchanged. secp256k1-kmp and bitcoin-kmp tests, lightning-kmp-app's library compile and engine-interop all pass. The Kotlin/Native workaround stays until a mac links without it. The submodule commits are on claude/dependency-update branches, not pushed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
engine-interop
The MLS engine the app ships against the one it shipped before, in two JVMs. docs/quartz-mls-marmot-update.md, Phase 3.
old/is one Marmot member played by quartz 1.15.2's engine, the one a Bennu peer on that version still runs. It reads one command per line on stdin and answers on stdout (seeOldPeer.kt).new/is the app's engine at the versions ingradle/libs.versions.toml: marmot-kmp over mls-kmp, through Magpie. Its driver starts the old peer as a child process and runs two scenarios: the new engine creates the room and the old one joins, then the reverse. In each, the creator commits two more adds after the Welcome, and both sides must land on the same epoch and kind:445 key and read each other's messages at every epoch.
They are two builds because they cannot be one: both depend on the com.vitorpamplona.quartz
group, at 1.15.2 and 2.20.0. Magpie serves that group exclusively, and its relay has nothing
older than 2.19.0, so old/ resolves from Maven Central with no Magpie at all.
Not part of ./gradlew check. It is a probe to run by hand when the engine moves.
../../../gradlew -p old installDist
../../../gradlew -p new run --args="$PWD/old/build/install/engine-interop-old/lib"
It prints ok per check and ALL PASSED, or FAIL with the step and exits 1. Move
new/build.gradle.kts's versions with the catalogue's, and let the first run add them to
new/gradle/magpie.lock.