Added fixes which caused issues building on Apple Silicon

[mcbarton]

Stopped Cuda library been loaded when on Apple
Add .dylib option for when on MacOS

[vgvassilev]

Are the changes introduced in Is 4eac1ca not sufficient?

[mcbarton]

I didn't find it to be sufficient on my machine.

[vgvassilev]

cc: @alexander-penev

[vgvassilev]

@mcbarton, there is some discussion about .so vs .dylib: https://gitlab.kitware.com/cmake/cmake/-/issues/21189. They say that macOS distinguishes between shared libraries and loadable modules. CMAKE_SHARED_LIBRARY_SUFFIX is set to .dylib for shared libraries. CMAKE_SHARED_MODULE_SUFFIX is set to .so for loadable modules (e.g. dlopen). See the [add_library](https://cmake.org/cmake/help/v3.18/command/add_library.html#normal-libraries) command's SHARED and MODULE options..

CppInterOp is meant to be dlopened thus it means that we should keep probably the .so extension. We can do that by adding the MODULE option to https://github.com/compiler-research/CppInterOp/blob/4eac1caa49d86ce9e51da1d46bfe38675f57444c/lib/Interpreter/CMakeLists.txt#L87C1-L93

[mcbarton]

@vgvassilev I tried your suggestion and it didn't work on my machine. Adding the MODULE option caused the following to occur

1. The following error is produced during cmake configure stage

CMake Error at unittests/CppInterOp/CMakeLists.txt:16 (target_link_libraries):
Target "clangCppInterOp" of type MODULE_LIBRARY may not be linked into
another target. One may link only to INTERFACE, OBJECT, STATIC or SHARED
libraries, or to executables with the ENABLE_EXPORTS property set.

CMake Error at unittests/CppInterOp/CMakeLists.txt:26 (target_link_libraries):
Target "clangCppInterOp" of type MODULE_LIBRARY may not be linked into
another target. One may link only to INTERFACE, OBJECT, STATIC or SHARED
libraries, or to executables with the ENABLE_EXPORTS property set.

2. dylib extension stayed (determined by turning off building of unittests)

3. libclangCppInterop.dylib got renamed to clangCppInterop.dylib (determined by turning off building of unittests)

Any other suggestions?

[vgvassilev]

Ok, then we are back to your original patch. The part I do not understand is why then the CMAKE_SHARED_LIBRARY_SUFFIX variable is not set. I'd rather have include(CMakeSystemSpecificInformation) than setting it by hand. I checked on my platform and that does set the right suffix. Can we try that?

[mcbarton]

@vgvassilev I just performed a quick check on my system and this new solution works. I'll push this solution to my fork sometime this evening

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[vgvassilev]

LGTM, can you run git-clang-format to the diff? That is, squash the commits into one and then run git-clang-format HEAD~.

[codecov]

Codecov Report

Merging #170 (33a2b2c) into main (4eac1ca) will not change coverage.
The diff coverage is n/a.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #170   +/-   ##
=======================================
  Coverage   72.75%   72.75%           
=======================================
  Files           8        8           
  Lines        2885     2885           
=======================================
  Hits         2099     2099           
  Misses        786      786           

Files	Coverage Δ
lib/Interpreter/Compatibility.h	90.27% <ø> (ø)

Files	Coverage Δ
lib/Interpreter/Compatibility.h	90.27% <ø> (ø)

[mcbarton]

@vgvassilev I have squashed both commits into one, and then performed the command git-clang-format HEAD~, and committed this. I think you meant me to do this in such a way as to reduce everything to a single commit, but was unsure. Will squash both commits of this pull request into 1 once you confirm.

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[vgvassilev]

@vgvassilev I have squashed both commits into one, and then performed the command git-clang-format HEAD~, and committed this. I think you meant me to do this in such a way as to reduce everything to a single commit, but was unsure. Will squash both commits of this pull request into 1 once you confirm.

Now I can do it on my end.