Add android cross compiling support

[jcfr]

This corresponds to #280 being rebased against current master (along with extracting documentation into its own CROSS-COMPILING.rst document)

• Test
• Consolidate changes proposed here with the one originally suggested by @o01eg in Initial Android support #262
• Add continuous integration (possibly by leveraging dockcross ¹

Footnotes

1. https://github.com/dockcross/dockcross#readme ↩

[jcfr]

@iyadahmed When (and if) you have the bandwidth, let us know what you think 🙏

Alternatively if time is limited, there is no problem integrating the patch originally suggested in #320 (comment)

[esnosy]

I've just tried it, it works great, but it still does not execute the cmTC_* executables in a virtual device automatically, but not a problem, it can be done in a separate patch, also I was going to execute them manually anyways, I think this patch is good to go so far
@jcfr

[esnosy]

Here are the results of the binaries on an Android 5.0 arm64 virtual device, I think all of them succeeded except two

[esnosy]

Yep here are exit codes for each program, there's two extra programs that turned out to not return 0

[esnosy]

here's the configure command I'm using btw

cmake .. \
-D_download_3.10.9_md5="25eb3686327c157dcb8ac9f01105c306" \
-DPYTHON_VERSION=3.10.9 \
-DCMAKE_INSTALL_PREFIX=/home/iyad/Development/Blender_On_Android/lib/android_aarch64/python \
-DCMAKE_SYSTEM_NAME=Android \
-DCMAKE_SYSTEM_VERSION=21 \
-DCMAKE_ANDROID_ARCH_ABI=arm64-v8a \
-DCMAKE_ANDROID_NDK=/home/iyad/Android/Sdk/ndk/25.1.8937393 \
-DCMAKE_ANDROID_STL_TYPE=c++_static
make install

[esnosy]

tests are failing on mac though :/ @jcfr
Let's integrate the other patch instead as it succeeds on MacOS according to CI

[esnosy]

for some reason this patch produces different check binaries than master

[esnosy]

Here are results of cmTC_* binaries from the other patch

[esnosy]

the second run of my patch after inputting return codes outputs more binaries

cmake .. \
-D_download_3.10.9_md5="25eb3686327c157dcb8ac9f01105c306" \
-DPYTHON_VERSION=3.10.9 \
-DCMAKE_INSTALL_PREFIX=/home/iyad/Development/Blender_On_Android/lib/android_aarch64/python \
-DCMAKE_SYSTEM_NAME=Android \
-DCMAKE_SYSTEM_VERSION=21 \
-DCMAKE_ANDROID_ARCH_ABI=arm64-v8a \
-DCMAKE_ANDROID_NDK=/home/iyad/Android/Sdk/ndk/25.1.8937393 \
-DCMAKE_ANDROID_STL_TYPE=c++_static \
-DDOUBLE_IS_BIG_ENDIAN_IEEE754=0 \
-DDOUBLE_IS_LITTLE_ENDIAN_IEEE754=0 \
-DPLATFORM_RUN=0 \
-DPLATFORM_RUN__TRYRUN_OUTPUT="aarch64-linux-gnu" \
-DDOUBLE_IS_ARM_MIXED_ENDIAN_IEEE754=0 \
-DHAVE_BROKEN_NICE_EXITCODE=0 \
-DHAVE_BROKEN_POLL_EXITCODE=0 \
-DHAVE_MMAP_DEV_ZERO_EXITCODE=0

so I guess this patch has more fixes for Android, but it breaks MacOSX

returns codes and std output of those newly generated cmTC_* binaries

[esnosy]

ah for automatic running on virtual device my bad I should have used the instructions from unix docs,
still this CI macos failing hmm

[esnosy]

The run_on_android.sh script works great! got Python on the virtual device!

now we need to figure out why MacOSX CI fails 😄

[esnosy]

here's my final config command, now using the default Python version as Python 3.10 fails for some reason

cmake .. \
-DCMAKE_INSTALL_PREFIX=/home/iyad/Development/Blender_On_Android/lib/android_aarch64/python \
-DCMAKE_SYSTEM_NAME=Android \
-DCMAKE_SYSTEM_VERSION=21 \
-DCMAKE_ANDROID_ARCH_ABI=arm64-v8a \
-DCMAKE_ANDROID_NDK=/home/iyad/Android/Sdk/ndk/25.1.8937393 \
-DCMAKE_ANDROID_STL_TYPE=c++_static \
-DCMAKE_CROSSCOMPILING_EMULATOR=../run_on_android.sh \
-DANDROID_ALLOW_UNDEFINED_SYMBOLS=ON \
-DENABLE_DECIMAL=OFF \
-DENABLE_CTYPES=OFF

[esnosy]

Things are becoming more clear now, imo,

• We should not rely on CMAKE_CXX_COMPILER_ID MATCHES "Clang" let's use APPLE OR ANDROID instead, hopefully this fixes the MacOSX CI
• Let's also use ANDROID instead of ${CMAKE_SYSTEM_NAME} MATCHES "^Android" for sake of consistency and robustness

@jcfr