CMake Fetch Content

[dbonattoj]

It would be nice to be able to import Halide in a C++ project using the CMake package FetchContent,
eg:

cmake_minimum_required(VERSION 3.22)
project(HalideTests)

set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED YES)
set(CMAKE_CXX_EXTENSIONS NO)

include(FetchContent)

# Halide
FetchContent_Declare(halide
                     GIT_REPOSITORY https://github.com/halide/Halide
                     GIT_TAG        v15.0.1)
FetchContent_GetProperties(halide)
if(NOT halide_POPULATED)
    message("Cloning HALIDE")
    FetchContent_Populate(halide)
    add_subdirectory(
    ${halide_SOURCE_DIR}
    ${halide_BINARY_DIR})
endif()
FetchContent_MakeAvailable(halide)

#find_package(Halide REQUIRED)

add_executable(HalideTests 
    "${CMAKE_CURRENT_SOURCE_DIR}/src/main.cpp"
)
target_link_libraries(HalideTests PRIVATE Halide::Halide)

When doing that, it complains that it doesn’t find LLVM,

I tried to use fetchcontent on llvm (which works perfectly but requires a lot of space and compile time), but Halide still didn’t find it.

Is someone able to make Halide work with FetchContent?

[alexreinking]

The problem I consistently ran into when trying to make this work during my first CMake rewrite was that this statement,

I tried to use fetchcontent on llvm (which works perfectly but requires a lot of space and compile time)

is not true.

LLVM's build interface is pretty different from the interface it exposes via its CMake package. Unfortunately, this makes it very difficult to achieve good FetchContent support while also depending on LLVM.

Maybe one day LLVM's build will progress to a point where it can be consistently consumed via either interface, but regardless the user experience of using LLVM via FetchContent will always be so bad that it's hard to prioritize.

I have some more pressing build system improvements for Halide planned that should make it more amenable to FetchContent. We can re-evaluate in a few months once those are done.