[feature] Clarify how to change location for conan_paths.cmake

As you all know, the typical CMake workflow is cd project && mkdir build && cd build && cmake .. && cmake --build .. In CMakeLists.txt of my projects I always prohibit in-source builds (build without creating build folder, directly in the project root). I used exports_sources approach with:

    def build(self):
        cmake = CMake(self)
        cmake.configure(source_folder=".", build_folder="build")
        cmake.build()

and spent hours trying to make it work because Conan generates cmake_paths.cmake outside of the build directory. Documentation for cmake_paths says:

The folder containing the conan_paths.cmake (self.install_folder when used in a recipe) is also included, so any custom file will be located too. Check cmake_find_package generator generator.

so I tried to adjust self.install_folder without any success. I don’t know if it’s possible at all, is it?

Then I found the layout() method. It says

self.folders.generators: To specify a subfolder where to write the files from the generators and the toolchains. (e.g the xx-config.cmake files from the CMakeDeps)

and suggests this:

     def layout(self):
         self.folders.source = "src"
         build_type = str(self.settings.build_type).lower()
         self.folders.build = "cmake-build-{}".format(build_type)
         self.folders.generators = os.path.join(self.folders.build, "conan")

I copy-pasted it and finally my build works. But I don’t understand how, it makes self.folders.generator to point into something like cmake-build-release/conan. I see no such folder after build and Conan files are located directly in cmake-build-release (otherwise it wouldn’t work because documentation suggests include(${CMAKE_BINARY_DIR}/conan_paths.cmake)). It’s very unclear what’s going on here. Actually, commenting that last line doesn’t break the build…

After all I found the cmake_layout(), now it works and my conanfile.py looks good. Can you please clarify what’s the proper way to do this? I also don’t understand why layout() is experimental if it’s absolutely vital to build canonical CMake projects, is it possible to use some non-experimental feature to achieve my goals? And one more thing, it would be so nice to have in docs a tutorial how to gracefully integrate canonical CMake projects with Conan. I’m now on this path and it’s pretty painful even though I watched all your courses and read the docs…

[memsharded]

I think that my main fallacy was that I thought Conan is already super-stable and super-integrated with everything because it’s been there for years so I tried to rely on non-experimental approaches 😃

Yes, you are right. The “stable” build system integrations were “frozen” like 4 years ago, and they were lacking features (like transparent integration) and sometimes had some bad defaults. But as they were frozen, they had some difficult to fix things without breaking, so they are stable, but not that good. So we tried to solve all those problems in the new integrations, so we could introduce them without breaking the others, but then the transition and mixing the two approaches is making the whole of it a bit messy and confusing.

We are very focused on 2.0, now in writing the new 2.0 docs, so I think we will better focus on getting that out as soon as possible, I think this one can be closed. Thanks very much for all the feedback! 😃

[OleksandrKvl]

I think that my main fallacy was that I thought Conan is already super-stable and super-integrated with everything because it’s been there for years so I tried to rely on non-experimental approaches 😃 I will continue to use cmake_layout() for now and will switch to CMakeToolchain in the future. I’m leaving this issue open but since my problem has been resolved feel free to close it if you don’t plan to update related documentation, thanks for you help 😃