Ensure module output directory is generated in configure stage

[awvwgk]

• dependent projects using the stdlib's build interface will resolve the module output directory before it is created and error without this fix

[ivan-pi]

Is this something I can test, by forking stdlib-cmake-example and trying to build a project with several modules?

[awvwgk]

I'm already fixing this behaviour in the CMake example, you would have to remove my fix from there first to check:

https://github.com/fortran-lang/stdlib-cmake-example/blob/61e5fb0a1df900d8ebfe216b0f60f1a5eeefcdf9/config/cmake/Findfortran_stdlib.cmake#L149-L151

[awvwgk]

Won't happen with a single dependency, so this can't be tested with the CMake example.

I usually hit issues with non-existent include directories when working with five to six interdependent CMake projects and multiple levels of dependencies, looks something like this:

CMake Error in CMakeLists.txt:
  Imported target "toml-f::toml-f" includes non-existent path

    "/home/awvwgk/projects/src/git/mctc-tools/tblite/__build/_deps/toml-f-build/include"

  in its INTERFACE_INCLUDE_DIRECTORIES.  Possible reasons include:

  * The path was deleted, renamed, or moved to another location.

  * An install or uninstall procedure did not complete successfully.

  * The installation package was faulty and references files it does not
  provide.

This error is usually gone on the second CMake run since all module directories are generated after the first run, even if it was unsuccessful because one target tried to access the module directory of a dependency before it was created.

I'm usually fixing this in my find modules by moving the module directory generation to an earlier point. Did cost me some time to figure this out and it is a quite unintuitive error. I guess users of stdlib have a low chance of ever encountering this issue and if they are able to build the required complexity in CMake they might also be able to figure out the hack to fix this. Still, it would be preferable if we can mitigate this issue directly in stdlib.

[ivan-pi]

I'm not well versed in CMake, but the change is small and I trust you have tested this so +1.

As a side note: do you think it makes sense to add a link in your comment, e.g. "see PR #521", or is this unnecessary because it can always be queried using git?

[milancurcic]

Thanks!

[jvdp1]

Sounds good to me. Thank you.