Why I use Make

Posted on May 31, 2015 - Subscribe - Home

I've changed many things in my software development workflow over the past year or so. I've tweeted about bits and parts of it. There is one thing change that stands out above all else. I started to use make and it changed everything about how I build and test my software.

It all started by using docker exclusively. This required a new workflow that I hadn't encountered before. I needed a binary based tool (read: no language stack required), could map files to other files (e.g. these files are used in this Dockerfile), and could run arbitrary commands as part of the wider development workflow. I also wanted something simple so I could create a similar development experience across all projects (e.g. there is a similar workflow for building the project and running its tests). Naturally this is a solved problem. Make solved it some 30 years ago.

Make is a tool for building projects--most commonly for compiled languages like C. The Makefile declares the rules for producing files and dependencies between files. For example you can have a rule that *.c should produce *.o to create a complex graph. Make also provides functionality to execute arbitrary tasks that do not produce any files (e.g make install). Together make has enough proven functionality to handle most projects.

Changes happened slowly. First I needed something to handle building multiple docker images. This was before docker introduced docker build -f. Turns out with a pattern rule and symlinks solved it easily. make images/app would take the appropriate file from dockerfiles/app and coordinate the process in the root folder. Then I need something to pull dependencies and setup the environment. make environment took care of that. Things really started rocking as I iterated on a generic Makefile.

Then something strange and interesting started. I started to use make to launch docker containers as part of the environment target. I combined those with -d & --name, save container ids to a file with a redirect and voilla. There's a way to start long running processes with a single command. This is not a revolutionary idea but it allowed me to do things I'd never done before. I think docker is also responsible because -d makes it simple to manage (start/stop) processes. I started to see the entire codebase as processes and only processes. Processes have their owned defined public interfaces (ports) and can be tested as such. Naturally it would be possible to start another process that could communicate over the public interface as a smoke tests. Does the process speak HTTP? Great, fire some curl commands. Does the process speak thrift? Great, start a thrift client and make some RPCs. Does the code base have some CLI utils to go with it? Great, then write a make target to call those things.

This unrelated workflow change has completely changed how I view my software and the verification of process deliverables. This is something I never expected. I stopped thinking in a whitebox and shifted to more on black box deliverables. There are only so many things you can do if you only focus on internals. If you never start the process and communicate with it your testing is a lie--pure and simple. These workflows have also drastically increased software quality more than anything I've done since I started doing TDD.

At this point my workflow goes like this:

  1. make dependencies - build any dependencies (may pull down external libraries, compile things, or anything like that)
  2. make - build the project
  3. make environment - build the project; start processes delivered by the project and any other long running processes needed in development
  4. make test - quick unit tests; should execute as quickly as possible and may not by exhaustive
  5. make test-ci - run every possible test
  6. make clean - remove project artifacts and stop anything

Then there are various other test targets. Ruby projects always have make test-boot. This actually start the process in a given environment. If the process has some CLI utils there is make test-cli. The list goes on and depending on the project.

This workflow dramatically increased my work quality. I suggest you look into these things as well. They may be as transformative for you as they are for me.

If you want concrete examples, then wait for the next post. That post is about replacing rake with make for great good.

— Adam Hawkins