Make for a Better World

Towards a robust compilation process

GNU Make

There are many different types of make system, among which the most famous one is GNU make. make is a tool designed for compiling programs into executables, but it also can be used for executing consecutive commands which have internal dependencies. The essense of make is dependency: it aims at figuring out the dependencies in your workflow with minimal effort. To some extent, I usually think of it as shell scripts with dependency control.

To take advantage of this powerful tool, we need a basic understanding of the wildcard and the dependency logic. When you first start with Makefiles, you would tend to explicit state every task as do on command line. However, this is exactly the redundancy make wants to free you from. Take a look at Makefile in Practice, which is a great introduction to most commonly used features.

Tips for make

• By default make looks for the makefile in the current level of the directory. You can use -C to specify the location of the target makefile.

CMake

Although make has been very popular for compiling programs on a Unix-like system, it certainly cannot do everything alone. People who write code on multiple different platforms became annoyed by the fact that they need to create different Makefiles for different platforms, given it Linux, Mac, or Windows. ¹ This is where Kitware’s CMake comes into play. In essence, cmake is a cross-platform tool for generating Makefiles. It also adds the capability of testing and deploying which may often be better than the make equivalence. When you automate multiple things together, magic can happen. Someone even says that using cmake force you to build better modular project. I would tend to use this for my next large C++ project.

Basic tutorials on CMake can be found: Hello-World-CMake and Modern Simple CMake.² For some latest features, it would be better to check the official website. For installing cmake, checkout modern cmake.

Here I list my conceptual understanding of cmake.

• The top level CMakeLists.txt is the configuration file where we specify everything globally.
• -S specifies where to find the CMakeLists.txt source file
• -B specifies where to store the generated makefile, together with some other generated configuration files.
• Must-have in the CMakeLists.txt
  • cmake_minimum_required(VERSION x.xx.x);
  • project(MYCOOLPROJECT);
  • add_executable(${PROJECT_NAME} main.cpp): executable name and source codes.
• Similar to the makefiles, CMakeLists.txt can be hierarchy, meaning that for each component/library, you can have a separate CMakeLists.txt. At the upper level, use add_subdirectory(SUBDIR) to specify where the underlying CMakeLists.txt is, targe_include_directories(${PROJECT_NAME} PUBLIC MYLIB) to specify the library headers, target_link_directories(${PROJECT_NAME} MYLIB) to locate the library objects, and target_link_libraries(${PROJECT_NAME} MYLIB) to specify the linking stage. At the lower level, the most important command to compile an object file is add_library(MYLIB lib.cpp lib.h).
• cmake has this industry standard trick to add code version into the executable. Check this video for a quick demo.³

Meson

There are even newer build systems, e.g. Meson. Meson is written in Python, targeting at a simpler, faster CMake.

• “Simpler”, thanks for the robustness and cleaness in Python.
• “Faster”, thanks for the underlying low-level assembler Ninja.

Meson only depends on core Python libraries. It is recommended to install Meson through pip.

Footnotes

1. I remember when I first arrived at Michigan, Gabor told me that our Fortran code BATSRUS is only available on Linux and Mac. Well, that’s because we are using only make!

2. Even better than the official tutorial!

3. In early 2020, Gabor implemented something like this in SWMF by himself. Vlasiator also has similar stuff, but is more brute-force like.