[vcpkg script] Test pkg-config integration

[dg0yt]

• What does your PR fix?

  Test user-perceived pkg-config integration in vcpkg (find_package(PkgConfig), pkg_check_modules(...)).

• Which triplets are supported/not supported? Have you updated the CI baseline?

  all, no

• Does your PR follow the maintainer guide?

  yes

• If you have added/updated a port: Have you run ./vcpkg x-add-version --all and committed the result?

  not needed.

Related

#22392 (comment)
#25529

[dg0yt]

Host pkgconf can be found reliably with CMake 3.22 FindPkgConfig.cmake (but only after system pkg-config). But there is room for problems in user projects:

• Our tests do set VCPKG_HOST_TRIPLET. Users might miss this parameter, so that CMAKE_PROGRAM_PATH wouldn't be set to the host tools. Solution for manifest mode: Extend triplet detection in CMake toolchain #25529.
• User's CMake before 3.22 doesn't look for pkgconf. Possible solutions:
  • Rename pkgconf[.exe] to pkg-config[.exe]. This would bring vcpkg's port in front of system tools, for all triplets, increasing decoupling from system libraries.
  • Add a pkgconfig.bat wrapper which forwards to pkgconf, for Windows.

[Neumann-A]

User's CMake before 3.22

Don't need to care about it. System pkg-config is ok on !Windows. Windows users should update cmake any way.

Main problem with current FindPkgConfig in vcpkg is that it doesn't support multi-config generators yet. We could do that with a wrapper installed alongside with pkgconf which does overrides the functions cmake normally uses and setup the targets correctlyy dependening on config.

[dg0yt]

User's CMake before 3.22

Don't need to care about it. System pkg-config is ok on !Windows. Windows users should update cmake any way.

I mostly agree, but I don't want to decide for all users.
A possible issue with system pkg-config on !windows is that it can hide problems with vcpkg pc modules.

Main problem with current FindPkgConfig in vcpkg is that it doesn't support multi-config generators yet. We could do that with a wrapper installed alongside with pkgconf which does overrides the functions cmake normally uses and setup the targets correctlyy dependening on config.

The problem with wrappers vs. port pkgconf is that wrappers are loaded for the target triplet, but pkgconf is used as a host dependency.
More generally, the problem is that you cannot know how the results of pkg_check_modules are used. As with CMake find modules (and wrappers), it is probably easy to carry multi-config with imported targets, but it is error-prone to handle multi-config for the variables. A modified behaviour wouldn't only affect the current project, but also its transitive dependencies.

(I have a local branch which additonally uses the Ninja multi-config generator for this test-port, but I didn't see a benefit for find_package tests.)

[JackBoosY]

I think vcpkg has the ability to provide usage information for pkg-config by now.
Next we should add usage printing code about pkg-config to vcpkg-tool.