Thanks for the feedback, I've cleaned up as requested (wanted a minimal source diff initially).
Please review again, thanks !

@moberhuber Thanks for the update! 👍

@jcfr This LGTM.

I've added 2 more fixes for cross-compiling VxWorks, could these be considered please? - Thx :)

@moberhuber thanks!

LGTM

CC @jcfr

@moberhuber Thanks again for your patches

Fixes related to libutil have been pushed into their own PR. See #165

Fix related to sys/select.h has been pushed into its own PR. See #166

Now for de1ffdf and fdd2c56, the exclusion of MSVC is already done when the version is empty or doesn't match, the issue happen because CMAKE_SIZEOF_VOID_P is not set.

With a working toolchain, I would expect this variable to be set to either 4 (32-bit) or 8 (64-bit), I suspect something is missing from the toolchain and/or companion files.

For reference this variable is set within CMake here using the content of the variable CMAKE_C_SIZEOF_DATA_PTR originally set here: https://github.com/Kitware/CMake/blob/60d80bca4afcfb3a38b588a5a2060cc275b4afbc/Modules/CMakeDetermineCompilerABI.cmake#L41-L59

Looking at the CMakeOutput.log should help understand better.

@moberhuber Let us know what you find out.

Hi @jcfr , regarding fdd2c56 I can confirm that the issue does not occur with the GNU toolchain which does set CMAKE_SIZEOF_VOID_P as part of its compiler detection.

But for VxWorks, we also want to build with the Diab toolchain which isn't supported by stock CMake so far. So we use our own (very basic) cross-toolchain information. This works fine everywhere else, in my fork I could build a basic Python that works even without CMAKE_SIZEOF_VOID_P being set (I suspect because any important compiler / platform information is discovered directly through other mechansims).

May be you could add the following to the toolchain file:

set(CMAKE_SIZEOF_VOID_P 8)

Or we could add a statement like this in the python cmake build system:

if(NOT DEFINED CMAKE_SIZEOF_VOID_P AND CMAKE_CROSSCOMPILING)
  set(CMAKE_SIZEOF_VOID_P 8)
  message(STATUS "Defaulting CMAKE_SIZEOF_VOID_P to 8 because CMakeDetermineCompilerABI failed to do so")
endif()

Let me know what you think.

Jc

To summarize:

• Fixes related to libutil have been pushed into their own PR. See Support system not expecting libutil #165
• Fix related to sys/select.h has been pushed into its own PR. See ConfigureChecks: VxWorks: Disable _XOPEN_SOURCE due to "sys/select.h" failing #166

If the last issue related to CMAKE_SIZEOF_VOID_P is still relevant, consider creating a new issue and evaluating the approach proposed in #160 (comment)