Package layout

Warning

This is an experimental feature subject to breaking changes in future releases. The layout() feature will be fully functional only in the new build system integrations (in the conan.tools space). If you are using other integrations, they might not fully support this feature.

Available since: 1.37.0

Before starting

To understand correctly how the layout() method can help us we need to recall first how Conan works.

Let’s say we are working in a project, using, for example, CMake:

<my_project_folder>
├── conanfile.py
└── src
    ├── CMakeLists.txt
    ├── hello.cpp
    ├── my_tool.cpp
    └── include
        └── hello.h

When we call conan create, this is a simplified description of what happens:

  1. Conan exports the recipe (conanfile.py) and the declared sources (exports_sources) to the cache. The folders in the cache would be something like:

    .conan/data/<some_cache_folder>
      ├── export
      │   └── conanfile.py
      └── export_source
          └── src
              ├── CMakeLists.txt
              ├── hello.cpp
              ├── my_tool.cpp
              └── include
                  └── hello.h
    
  2. If the method source() exists, it might retrieve sources from the internet. Also, the export_source folder is copied to the source folder.

    .conan/data/<some_cache_folder>
      ├── export
      │   └── conanfile.py
      ├── export_source
      │   └── src
      │       ├── CMakeLists.txt
      │       ├── hello.cpp
      │       ├── my_tool.cpp
      │       └── include
      │           └── hello.h
      └── source
          └── src
              ├── CMakeLists.txt
              ├── hello.cpp
              ├── my_tool.cpp
              └── include
                  └── hello.h
    
  3. Before calling the build() method, a build folder is created and the sources are copied there. Later, we call the build() method so the libraries and executables are built:

    .conan/data/<some_cache_folder>
     ├── export
     │   └── conanfile.py
     ├── export_source
     │   └── src
     │       ├── CMakeLists.txt
     │       ├── hello.cpp
     │       ├── my_tool.cpp
     │       └── include
     │           └── hello.h
     ├── source
     │   └── src
     │       ├── CMakeLists.txt
     │       ├── hello.cpp
     │       ├── my_tool.cpp
     │       └── include
     │           └── hello.h
     └── build
         └── <build_id>
             ├── say.a
             └── bin
                 └── my_app
    
  4. At last, Conan calls the package() method to copy the built artifacts from the source (typically includes) and build folders (libraries and executables) to a package folder.

    .conan/data/<some_cache_folder>
      ├── export
      │   └── conanfile.py
      ├── export_source
      │   └── src
      │       ├── CMakeLists.txt
      │       ├── hello.cpp
      │       ├── my_tool.cpp
      │       └── include
      │           └── hello.h
      ├── source
      │   └── src
      │       ├── CMakeLists.txt
      │       ├── hello.cpp
      │       ├── my_tool.cpp
      │       └── include
      │           └── hello.h
      ├── build
      │   └── <build_id>
      │       ├── say.a
      │       └── bin
      │           └── my_app
      └── package
          └── <package_id>
              ├── lib
              │   └── say.a
              ├── bin
              │   └── my_app
              └── include
                  └── hello.h
    
  5. The package_info(self) method will describe with the self.cpp_info object the contents of the package folder, that is the one the consumers use to link against it. If we call conan create with different configurations the base folder in the cache is different and nothing gets messed.

    conanfile.py
     import os
     from conans import ConanFile
     from conan.tools.cmake import CMake
    
    
     class SayConan(ConanFile):
         name = "say"
         version = "0.1"
         exports_sources = "src/*"
         ...
         def package_info(self):
             # These are default values and doesn't need to be adjusted
             self.cpp_info.includedirs = ["include"]
             self.cpp_info.libdirs = ["lib"]
             self.cpp_info.bindirs = ["bin"]
    
             # The library name
             self.cpp_info.libs = ["say"]
    

So, this workflow in the cache works flawless but:

  • What if I’m developing the recipe in my local project and want to use the local methods (conan source, conan build) and later call export-pkg to create the package?

    If you call conan build in your working directory, without specifying a --build-folder argument, you will end with a bunch of files messing with your project. Moreover, if you want to build more configurations you will need to create several build folders by hand, this is inconvenient, error-prone, and wouldn’t be easy for Conan to locate the correct artifacts if you want to call export-pkg later.

  • What if I don’t even want to call conan build but use my CLion IDE to build the project?

    By default, the CLion IDE will create the folders cmake-build-release and cmake-build-debug to put the build files there, so maybe your package() method is not able to locate the files in there and the export-pkg might fail.

  • What if I want to use my project as an editable package?

    If you want to keep developing your package but let the consumers link with the artifacts in your project instead of the files in the Conan cache, you would need to declare a yml file describing where are the headers, the libraries, the executables in your application.

So, the same we describe the package folder in the package_info() method, we can use the layout() to describe the source and build folders (both in a local project and in the cache) so we can:

  • Run the conan local commands (conan source, conan build, conan export-pkg) without taking care of specifying directories, always with the same syntax.

  • If you are using an IDE you can describe the build folder naming in the layout, so the libraries and executables are always in a known place.

  • In the cache, the layout (like a build subfolder) is kept, so we can always know where are the artifacts before packaging them.

  • It enables tools like the AutoPackager to automate the package() method.

  • It enables out-of-the-box to use editable packages, because the recipe describes where the contents will be, even for different configurations, so the consumers can link with the correct built artifacts.

Declaring the layout

In the layout() method you can set:

  • self.folders

    • self.folders.source: To specify a folder where your sources are.

    • self.folders.build: To specify a subfolder where the files from the build are (or will be).

    • 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)

    • self.folders.imports: To specify a subfolder where to write the files copied when using the imports(self) method in a conanfile.py.

    Check the complete reference of the “.folders” attribute.

  • self.cpp.source and self.cpp.build: The same you set the self.cpp.package to describe the package folder after calling the package() method, you can also describe the source and build folders.

  • self.cpp.package: You can use it as you use the self.cpp_info at the package_info(self) method. The self.cpp_info object will be populated with the information declared in the self.cpp.package object, so you can complete it or modify it later in the package_info(self) method.

Example: Everything together

Let’s say we are working in the project introduced in the section above:

<my_project_folder>
├── conanfile.py
└── src
    ├── CMakeLists.txt
    ├── hello.cpp
    ├── my_tool.cpp
    └── include
        └── hello.h

We are using the following CMakeLists.txt:

cmake_minimum_required(VERSION 3.15)
project(say CXX)

add_library(say hello.cpp)
target_include_directories(say PUBLIC "include")

add_executable(my_tool my_tool.cpp)
target_link_libraries(my_tool say)

# The executables are generated at the "bin" folder
set_target_properties(my_tool PROPERTIES RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/bin")

Let’s see how we describe our project in the layout() method:

conanfile.py
 import os
 from conans import ConanFile
 from conan.tools.cmake import CMake


 class SayConan(ConanFile):
     name = "say"
     version = "0.1"
     exports_sources = "src/*"
     ...
     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")

         self.cpp.package.libs = ["say"]
         self.cpp.package.includedirs = ["include"] # includedirs is already set to this value by
                                                    # default, but declared for completion

         # this information is relative to the source folder
         self.cpp.source.includedirs = ["include"] # maps to ./src/include

         # this information is relative to the build folder
         self.cpp.build.libdirs = ["."]        # maps to ./cmake-build-<build_type>
         self.cpp.build.bindirs = ["bin"]        # maps to ./cmake-build-<build_type>/bin

     def build(self):
         cmake = CMake(self)
         cmake.configure()
         cmake.build()
         # we can also know where is the executable we are building
         self.run(os.path.join(self.build_folder, self.cpp.build.bindirs[0], "my_tool"))

Let’s review the layout() method changes:

  • self.folders

    • As we have our sources in the src folder, self.folders.source is set to “src”.

    • We set self.folders.build to be cmake-build-release or cmake-build-debug depending on the build_type.

    • The self.folders.generators folder is where all files generated by Conan will be stored so they don’t pollute the other folders.

    Please, note that the values above are for a single-configuration CMake generator. To support multi-configuration generators, such as Visual Studio, you should make some changes to this layout. For a complete layout that supports both single-config and multi-config please check the cmake_layout() in the Conan documentation.

  • self.cpp

    Also, we can set the information about the package that the consumers need to use by setting the conanfile’s cpp.package attributes values:

    • Declaring self.cpp.package.libs inside the layout() method is equivalent to the “classic” self.cpp_info.libs declaration in the package_info() method.

    • Also, as you may know, self.cpp.package.includedirs is set to [“include”] by default, so there’s no need in declaring it but we are leaving it here for completeness.

    We can describe also the source and build folders with the cpp.source and cpp.build objects:

    • We are setting self.cpp.source.includedirs = ["include"]. The self.folders.source information will be automatically prepended to that path for consumers so, for example, when working with an editable package, Conan will try to get the include files from the ./my_project_folder/src/include folder.

    • We set the self.cpp.build.libdirs to [“.”], so we are declaring that, if we make the package editable, the libraries will be at the ./cmake-build-<build_type> folder.

    • We set the self.cpp.build.bindirs to [“bin”], because the CMakeLists.txt file is changing the RUNTIME_OUTPUT_DIRECTORY to that directory.

There is also an interesting line in the build(self) method:

conanfile.py
   def build(self):
      ...
      # we can also know where is the executable we are building
      self.run(os.path.join(self.build_folder, self.cpp.build.bindirs[0], "my_tool"))

We are using the self.cpp.build.bindirs[0] folder to locate the my_tool. This is a very recommended practice especially when our layout depends on the build system, for example, when using CMake with Visual Studio, the binaries are typically built at Release/ or Debug/ (multiconfiguration) but in a regular Linux or Macos the output folder will be “.”, so it is better to declare the layout self.cpp.build.bindirs following that logic and then just access to the correct path if we need to know where are the resulting files of our build. If you check the cmake_layout() you can see that the predefined cmake_layout is doing exactly that when using a multiconfiguration build system.

So, now we can run the conan local methods without taking much care of the directories where the files are or the build files should be, because everything is declared in the layout:

# This will write the toolchains and generator files from the dependencies to the ``cmake-build-debug/generators``
$ conan install . -if=my_install -s build_type=Debug

# In case we needed it (not the case as we don't have a source() method), this would fetch the sources to the ./src folder.
$ conan source . -if=my_install

# This will build the project using the declared source folder and ``cmake-build-debug`` as the build folder
$ conan build . -if=my_install

Note

Maybe you are wondering why the install folder is not parametrized and has to be specified with the -if argument. Currently, Conan generates several files like the graph_info.json and the conanbuildinfo.txt that are read to restore the configuration saved (settings, options, etc) to be applied in the local commands. That configuration is needed before running the layout() method because the folders might depend on the settings like in the previous example. It is a kind of a chicken-egg issue. In Conan 2.0, likely, the configuration won’t be stored, and the local methods like conan build . will compute the graph from arguments (–profile, -s, -o…) and won’t need the --if argument anymore, being always trivial to run.

Our current folder now looks like this:

<my_project_folder>
├── conanfile.py
├── src
│   ├── CMakeLists.txt
│   ├── hello.cpp
│   ├── my_tool.cpp
│   └── include
│       └── hello.h
└── cmake-build-debug
    ├── libsay.a
    └── bin
        └── my_tool

We could put the package in editable mode and other packages that require say would consume it in a completely transparent way, even locating the correct Release/Debug artifacts.

$ conan editable add .  say/0.1

Note

When working with editable packages, the information set in self.cpp.source and self.cpp.build will be merged with the information set in self.cpp.package so that we don’t have to declare again something like self.cpp.build.libs = ["say"] that is the same for the consumers independently of if the package is in editable mode or not.

And of course we can run also a conan create command. When the build(self) method is run in the conan cache, it is also able to locate the my_tool correctly, because it is using the same folders.build:

.conan/data/<some_cache_folder>
  ├── source
  │   └── src
  │       ├── CMakeLists.txt
  │       ├── hello.cpp
  │       ├── my_tool.cpp
  │       └── include
  │           └── hello.h
  ├── build
  │   └── cmake-build-debug
  │       ├── say.a
  │       └── bin
  │           └── my_app
  └── package
      ├── lib
      │   └── say.a
      ├── bin
      │   └── my_app
      └── include
          └── hello.h

Warning

The conan package local command has been disabled (will raise an exception) when the layout() method is declared. If the package can be consumed “locally” in a handy way, the use case for the conan package method is only testing that the method is correctly coded, but that can also be done with the conan export-pkg method. This responds to the migration to Conan 2.0, where the conan package method will disappear.