Makefiles are instructions (“rules”) for a computer on how to build “stuff”. Think of makefiles as a recipe you may know from cooking (“Baking a cake: First, take some flour, then add milk […]") - but then for computers.
Makefiles originate in software development, where they have been used to convert source code into software programs that can then be distributed to users.
Researchers can use makefiles to define rules how individual components (e.g., cleaning the data, running an analysis, producing tables) are run. When dependencies (e.g., to run the analysis, the data set first has to be cleaned) are well-defined, researchers can completely automate the project. When making changes to code, researchers can then easily “re-run” the entire project, and see how final results (potentially) change.
A rule in a makefile generally looks like this:
targets: prerequisites commands to build
The targets are things that you want to build - for example, data sets, outputs of analyses, a PDF file, etc. You can define multiple targets for one rule (separate targets by spaces!). Typically, though, there is only one per rule. Think of this as the dish (or part of the dish) that you want to create with your recipe.
dataset.csv(my final, cleaned dataset)
The prerequisites are the things that you need before you can build the target. It’s also a list of file names, separated by spaces. These files need to exist before the commands for the target are run. They are also called dependencies. The cool thing is that
makeautomatically checks whether any of the dependencies has changed (e.g., a change in the source code) - so it can figure out which rules to be run, and which ones not (saving you a lot of computation time!).You can view these as the ingredients of the recipe.
rawdata1.csv clean.R(before building
dataset.csv, the raw data and a specific script to clean the raw data need to exist)
The commands are a series of steps to go through to build the target(s). These need to be indented with a tab (“start with a tab”), not spaces. The commands can be seen as the recipe instructions.
- Example: the command
R --vanilla < clean.Ropens R, and runs the script
- Example: the command
dataset.csv: rawdata1.csv clean.R R --vanilla < clean.R
A makefile typically consist of multiple rules, which can depend on each other.
# rule to build target2 target2: target1 commands to build # rule to build target1 target1: prerequisite1 commands to build
Advanced Use Cases
Use directory names
You can easily use directory names in makefiles, e.g., to specify that a prerequisite is in one directory, and the target in another.
gen/data-preparation/aggregated_df.csv: data/listings.csv data/reviews.csv Rscript src/data-preparation/clean.R
Targets typically refer to output - such as files. Sometimes, it’s not practical to generate outputs. We call these targets “phony targets”.
- Creating a target
cleanis a convention of makefiles that many people follow.
- The phony target
allcalls all targets. Think of it as a “meta rule” to build it all!
- The target
cleantypically is used to remove generated/temp files, so you can start with a clean copy of your directory for testing.
all: one two one: touch one.txt two: touch two.txt clean: rm -f one.txt two.txt
Variables in a make script prevent you from writing the same directory names (or command to execute a program) over and over again. Variables are typically defined at the top of the file and can be accessed with the
$ command. Note that variables can only be strings.
INPUT_DIR = src/data-preparation GEN_DATA = gen/data-preparation $(GEN_DATA)/aggregated_df.csv: data/listings.csv data/reviews.csv $(INPUT_DIR)/clean.R