Fix find_headers to also look for C++ headers and Fortran modules

[michaelkuhn]

This is likely caused by #10623. Currently, only C headers are considered, causing build failures for packages depending on, e.g., netcdf-fortran and xerces-c. Additionally, the regex used to look for the include path component did not consider word boundaries, causing false matches.

[alalazo]

An alternative approach for the netcdf-fortran problem would be to extend the list of suffixes considered in find_headers.

I'd like that better, as there could be more Fortran packages suffering from this. Curious what @adamjstewart and @scheibelp think about that.

[scheibelp]

@alalazo is this expected? I would think that the logic of #10623 would add <prefix>/include if any subdirectory in <prefix>/include contained a header file.

@michaelkuhn could you add the error message here (or point me to the relevant issue)?

[michaelkuhn]

@michaelkuhn could you add the error message here (or point me to the relevant issue)?

Sorry, I did not capture the error output. esmf failed to build because the compiler could not find netcdf.mod and the xerces header. I then checked the build environment and saw that SPACK_INCLUDE_DIRS did not contain the prefix/include for these two packages. I can reproduce it if you want.

[alalazo]

@scheibelp

is this expected? I would think that the logic of #10623 would add /include if any subdirectory in /include contained a header file.

It's expected in the sense that find_headers (not modified in #10623) will recognize as headers only *.h files, while as far as I get from the comment here the directory only contains Fortran *.mod files. My suggestion was to add *.mod to the list of files recognized as headers as this will possibly solve the same issue in other similar packages.

[scheibelp]

It's expected in the sense that find_headers (not modified in #10623) will recognize as headers only *.h files, while as far as I get from the comment here the directory only contains Fortran *.mod files.

OK thanks I did not realize that. So

An alternative approach for the netcdf-fortran problem would be to extend the list of suffixes considered in find_headers.

I'd like that better, as there could be more Fortran packages suffering from this. Curious what @scheibelp think about that.

That sounds good

[michaelkuhn]

I have pushed an updated version that now makes both package-specific changes unnecessary. I missed that xerces-c ships C++ headers the first time around and have now updated find_headers to also look for .hpp headers.

[scheibelp]

Thanks!