Suggestion: Package indirect.h and polymorphic.h inside a `xyz_value_types` directory

[jogerh]

When installing value_types (cmake install or vcpkg), the header files indirect.h and polymorphic.h end up in the top level include directory. It could be better to group these inside an xyz_value_types subdirectory, so that clients would include them like #include <xyz_value_types/indirect.h> instead of #include <indirect.h>. This is mostly a cosmetic topic, but it seems to be quite common practice for other libraries.

[jbcoe]

This seems like an excellent idea.

We could follow the approach taken by absl: https://github.com/abseil/abseil-cpp/blob/master/CMakeLists.txt

[jbcoe]

I'm tempted to use xyz/value_types/*. Can anyone envisage this causing problems?

[Twon]

The main concern here is that the CMake build has multiple ways for the repo to be used, and is must be consistent across all:

• An installed cmake package (find_package via a config)
• Via package managers such as Conan & VcPkg
• As part of a wider super build project (i.e. as a submodule or via FetchContent)
• Or a find module (find package via a module)

This means this folder structure will need to be reflected in the repository (adding a xzy/value_types) folder at the root, and must be added to the include directory of the installed package. This should be enough to ensure that the VcPkg and Conan packages reflect this change once they are packaged to pick up the release with these changes.

As an aside this is the reason when adding libraries we add an alias library, so the cmake library names will match regardless of if you include the library via (FetchContent, Git submodule or if you use the config generated by the export set when building against an installed cmake package:

value_types/cmake/xyz_add_library.cmake

Line 61 in 6f50aff

add_library(${XYZ_ALIAS} ALIAS ${XYZ_NAME})

Perhaps we should add some tests to CI against all of these method? At least we could then ensure all packaging methods are supported and working (as we've already had issues with the CMake installed package this week)

[jbcoe]

A packaging test for CI seems like a fine idea. Are there examples we can follow?