Having a build system in a project is something many developers take for granted, never paying attention to the work involved in maintaining the build's quality and health. Can we identify whether a build system is good or not?
During my first years as a software engineer, I would feel like there was something wrong with the way we built our project; but being my first job and being mostly ignorant on build systems, I assumed we did things right. Until I decided to study a bit on the topic - oh was I wrong.
This post is meant as a high level introduction to building C++ software.
On a high level, building a project is a two-step process: first, a set of build files is converted into build artifacts. Then, a subset of those build artifacts is moved into some installation directory.
The first step is where the compiler and linker are involved, doing most of the work. But a lot more can happen in this stage, like running scripts to generate code, creating test inputs, generating documentation, etc.
In the second step, some build artifacts are moved into an installation directory, making the project available to anyone in the system. It's possible for developers to skip the installation step when developing, because developers can execute all components of the project using the build artifacts alone.
In the steps above, we have identified three important directories:
- Source directory, where the source files are located.
- Build directory, where the build artifacts are placed.
- Install directory, where the installed program is placed.
A wrong way to build
In some build methods, the source directory and the build directory are the same, this is known as an in-source build:
In other words, source code, scripts and test inputs are all in the same directory as object files, script-generated code, compiled libraries and executables, automatically generated documentation, etc.
Let's pause here for a moment and consider the implications of this build method:
- The version control system has to distinguish between files that should be committed and files that are build artifacts.
- Because the build process places files in the same directory where source code is located, it needs special care not to overwrite or delete important files1.
- Cleaning a build (deleting all build artifacts) becomes a non-trivial task.
- Sharing the same source files for independent builds becomes non-trivial (more on this later).
Note that it's possible to overcome each of those problems at the cost of added complexity. However, when striving for simplicity, we must ask if there are simpler ways of achieving the goals above; the answer is yes.
A cleaner approach
The build method in which the source and build directories are distinct is called an out-of-source build. The complete separation between source files and build artifacts makes for a simpler and more practical organization of the project:
- A file is under version control if and only if it is inside the source directory.
- The build process only ever outputs files to the build directory, therefore there is no risk of changing the source material.
- Cleaning a build is simple:
rm -r <build_directory>.
- Because the source directory is not aware of the build directory, the same source can easily be used for multiple builds.
This last point is so important that it deserves elaborating. Let's imagine a scenario where we have to develop an application on both Windows and Linux. We can place the source directory in network storage, and maintain build directories in the local (faster) storage of each system. One build does not affect the other by design.
Suppose that building on debug mode is expensive, so we don't want to do it all the time. When we want debug mode because of a faulty change, we simply create another build directory, configure it to debug, and build. There is no need to clone the repository and to apply the faulty changes on top of the new copy.
Suppose that we want to compare the performance of an application when built with two different compilers. We simply keep two distinct build directories, one for each compiler.
We've talked about what it means to build software, but this article started talking about build systems. What is a build system?
A build system is a description of how to build a particular software, combined with a program that reads this description and acts upon it. Here's the world's simplest build system:
In this example, our build system is a simple shell script that, when invoked,
main.cpp and installs the generated executable
/usr/local/bin2. It is an example of an in-source build. A better example
The new script can be invoked multiple times, with different values for
INSTALL_DIRECTORY. This is an example of an
There are many build systems out there, but they all follow this pattern: a file that describes the build process using some language is placed in the source directory, and a program is invoked to read that description and build the project.
It is the build system's job to:
- Figure out how to run code generation scripts.
- Find header files.
- Find external dependencies.
- Move files around.
- Determine the order in which build steps must happen.
- Determine which steps can be done in parallel.
- Determine where to find tools like the compiler.
- Understand how the compiler is invoked.
Depending on the expressiveness of the build system's language, the programmer might have to perform a lot of "hand-holding" for some or all of the steps above. In other words, the build description file might allow for higher level abstractions and be easy to reason about, or it might require low level commands to be spelled out (as the shell example showed).
Each build system has its own view on how to achieve those features. Is it possible to get all of them and still be operating system agnostic? For the C++ case, that's what CMake aims to do and we will explore how it does that in a future article.
Given all that we've discussed, it's possible to identify symptoms of a bad build system:
- How easy is it to spawn a second build from the same source directory?
- How easy is it to identify files that must be under version control?
- Can you identify the compiler that is used in a given build? How easily can that be changed?
- Can you build a single component of the project and its dependencies, without building anything unnecessary?
- If a source file is changed, how easy is it to incrementally build the affected components?
A good build system will also allow you to reason about individual components of your project and how they relate to each other, so that you can easily identify circular dependencies, or add new components with the correct dependencies.
Furthermore, a good build system is capable of inferring a lot of information given a description of the project. For instance, it should be able to find system libraries, understand how to invoke the compiler and automatically infer parallelism between build steps.
From the number of times I wrote easily in the text above, you can tell how much I value simplicity. When things are complicated, few developers know how to maintain a healthy build, and the build quality slowly deteriorates.
Once enough components are added, the project reaches a point where no one truly understands the dependencies between components, how to fix build breaks, or how to reduce the number of components built as result of dependencies. The consequence is a generalized loss in productivity as developers have to context switch while wainting for the subpar build.
Thanks for reading! Next time we'll talk about how CMake accomplishes most of the goals we talked about. In the meantime, try to identify how good the build system of your project is.
edit: Part 2 is out!
- If one has used Perforce as a version control system, one could argue that Perforce is so great it prevents the aforementioned issue from happening. Please don't use an objectively wrong build methodology as an excuse to use an objectively bad (and expensive) version control system.↩
- Note that
/usr/local/binis probably a write-protected directory. As such, sudo permissions might be needed to run this script.↩