Fix crashes for RealmAny on ARM 32 bit architectures

[rorbech]

Closes #7626

[rorbech]

This fix isn't verified by CI, but I have executed our full test suite as 32-bit tests on an arm devices. Dump of important output is:

$ ./gradlew connectedObjectServerDebugAndroidTest

[...]

[CXX5202] This app only has 32-bit [armeabi-v7a] native libraries. Beginning August 1, 2019 Google Play store requires that all apps that include native libraries must provide 64-bit versions. For more information, visit https://g.co/64-bit-requirement

> Task :kotlin-extensions:connectedObjectServerDebugAndroidTest
Starting 61 tests on SM-G973F - 11

[...]

> Task :realm-library:connectedObjectServerDebugAndroidTest
Starting 8163 tests on SM-G973F - 11

[...]

BUILD SUCCESSFUL in 31m 54s
117 actionable tasks: 7 executed, 110 up-to-date

[cmelchior]

Great find 🥇 ... but would still be interesting to understand the root cause.

[rorbech]

Great find 🥇 ... but would still be interesting to understand the root cause.

Guess the pointer size that can differ based on compiler/platform, etc, so forcing it into jlong (int64_t) ensures that it isn't truncated. But if the platform somehow only used smaller types you would think that it could also fit inside the jlong passed to java. Maybe something skew issue around the va_list convention as it seemed like the erroneous pointer was pointing to memory containing (J)V which is the signature argument to the Java call.

I have tried to reach out to the core team if anybody has some insight.

[clementetb]

Great find ⭐️!

[rorbech]

The cast is needed due to https://en.cppreference.com/w/cpp/language/reinterpret_cast point 2 and 3 (thanks James)