Cake is a really thin, drop-in replacement/wrapper around make that runs all of your targets inside of a development Docker/Podman container.

• Though Cake supports more complex workflows, most projects that currently have a Makefile at their root should also place a developer-focused Dockerfile there for convenience and portability
  • The Makefile is the single source of truth for the build process
  • The Dockerfile is the single source of truth for the build environment
• A container runtime should not be a hard dependency to build the project.
• Choosing between containerized and "naked" builds should be as easy as typing make or cake interchangeably
• CI/CD pipelines should be able to reuse the instructions from the Makefile in an ergonomic way without having to keep the build context in mind

Because I found myself constantly writing Makefiles that run their targets in a container, then adding in add-hoc ways for people not to use the container through environment variables, followed by a half-hearted attempt at optimizations through bind-mounts and less frequent restarts, and some faulty logic to avoid name and tag clashes. I figured it was time to extract this into a script. Despite its simplicity, the script covers 99% of my use cases for tools like act without being tied to a specific forge.

Just use cake instead of make. The defaults should fit most use cases.

If you really have to, you can specify additional docker/podman arguments using $CAKE_RUNTIME_ARGS. I recommend placing these in your .envrc if you need them to stick around due to the specific needs of your project.

If you're building/testing your software against multiple environments, you can always set $CAKE_DOCKERFILES (defaults to Make's ${PWD}/Dockerfile - which is not necessarily the same as your shell's ${PWD}/Dockerfile). This will run your Make targets in one container per Dockerfile. If $CAKE_DOCKERFILES is a directory, all Dockerfiles in that directory (and all of its sub-directories) will be used. This is the one area in which Cake diverges from Make. You have to specify Cake-relevant environment variables before the command, not after. You can take a look at some of my test cases for example invocations:

cake
cake all
cake -C subdir
CAKE_DOCKERFILES='subdir/' cake
CAKE_DOCKERFILES='subdir/Dockerfile' cake
CAKE_DOCKERFILES='subdir/one.dockerfile subdir/Dockerfile' cake

If I want to debug my development container, I like to add a shell target to my Makefile like so:

shell:
    /bin/sh

It's more ergonomic then copying the container name.

The same goes for dealing with things like ./autogen.sh and the ./configure script (often managed directly by the user). I tend to call those through a Makefile as well. Take this snippet from the GNUMakefile in the Emacs source tree as an example:

configure:
	@echo >&2 'There seems to be no "configure" file in this directory.'
	@echo >&2 Running ./autogen.sh ...
	./autogen.sh
	@echo >&2 '"configure" file built.'

Makefile: configure
	@echo >&2 'There seems to be no Makefile in this directory.'
	@echo >&2 'Running ./configure ...'
	./configure
	@echo >&2 'Makefile built.'

# 'make bootstrap' in a fresh checkout needn't run 'configure' twice.
bootstrap: Makefile
	$(MAKE) -f Makefile all

Because additional dependencies are a problem, especially in corporate environments. just curl/copy this script into a directory on your $PATH and you're good to go.

I might provide them for convenience later, but in principle all you need to do is reuse existing make completions. In zsh that looks something like this:

compdef _make cake