For CMake 3.21, we are adding support for RUNTIME_DEPENDENCY_SET and RUNTIME_DEPENDENCIES keywords to install(TARGETS), but only RUNTIME_DEPENDENCY_SET for the new install(IMPORTED_RUNTIME_ARTIFACTS) subcommand. Should the latter command also support a RUNTIME_DEPENDENCIES keyword for consistency?
Cc: @kyle.edwards
RUNTIME_DEPENDENCIES is just a shortcut to create an anonymous RUNTIME_DEPENDENCY_SET. install(TARGETS) gets used a lot more frequently than install(IMPORTED_RUNTIME_ARTIFACTS), and so I think it’s more deserving of its own shortcut.
If there’s enough demand for RUNTIME_DEPENDENCIES with install(IMPORTED_RUNTIME_ARTIFACTS), we could consider adding it.
I get that, but my comments were intended to highlight an inconsistency in a new API. We have an opportunity to not create that inconsistency from the outset. In getting myself up to speed on these features, it was something that as a user I noticed relatively early. It didn’t make sense (again, to me as a user) why RUNTIME_DEPENDENCIES was supported for one and not the other. I found myself leaning toward basically ignoring RUNTIME_DEPENDENCIES and just focusing on learning how to use RUNTIME_DEPENDENCY_SET because I know that will work for both subcommands. We shouldn’t be placing such obstacles on the learning path for users exploring these new features if we can avoid it.