[vcpkg-ci-fortran] New test port

[dg0yt]

WIP

[dg0yt]

As expected (from octave CI logs): vcpkg_find_fortran unexpectedly switches C and CXX compiler from MSVC to MinGW.

[Neumann-A]

That is probably intentional since otherwise you cannot combine fortran and other languages object files into a single lib. At least if you use default gfortran in vcpkg

[dg0yt]

That is probably intentional

This is what I assume...

since otherwise you cannot combine fortran and other languages object files into a single lib [...] if you use default gfortran in vcpkg

... but I wonder if this is the right goal? This seems to clearly contradict the policy of "the triplet defines the (C/C++) compiler." It isn't allowed to change the compiler even for packages which more or less insist on clang(-cl) for windows.

And with MinGW C/C++, the implications are even more tricky IMO.

(AFAIU CMake promises some limited use of GFortran with MSVC in form of fortran libs being transformed to what is accepted by MSVC, but not to "combine fortran and other languages object files into a single lib".)

[Neumann-A]

"the triplet defines the (C/C++) compiler."

historic reasons. The correct approach is displayed in #42129 and uses per port customization which avoids the rebuild of the whole tree via a conditional include and VCPKG_HASH_ADDITIONAL_FILES. However, that approach wasn't implementable in the past and would always require a whole rebuild of the tree if a single port required special behavior. As such the special behavior has been implemented into the portfile instead of the triplet to avoid a rebuild of the tree. Maybe it is time to remove the port special compiler behavior and move it into the triplet instead.

CMake promises some limited use of GFortran with MSVC in form of fortran libs being transformed to what is accepted by MSVC, but not to "combine fortran and other languages object files into a single lib

It has GNUtoMS which works not only for Fortran.

However, I wouldn't spend too much time on it since there seems to be some historic decision in the team to switch to flang as soon as it becomes readily available from llvm. (and then the GNUtoMS hack is no longer necessary)