Skip to content
/ walk Public

A fast, general purpose, graph based build and task execution utility.

License

Notifications You must be signed in to change notification settings

ejholmes/walk

Repository files navigation

Build Status Go Report Card Latest Version

walk is a fast, general purpose, graph based build and task execution utility.

Heavily inspired by make and redo.

Features

  • Fast parallel execution.
  • Graph based dependency management.
  • Maximum composability with existing UNIX tooling.
  • Describe targets and their dependencies as simple executables.
  • Universal execution; execute walk from any directory.

Installation

Using Go 1.7+:

$ go get -u github.com/ejholmes/walk

Or grab the latest release from https://github.com/ejholmes/walk/releases.

Usage

walk is built on top of a very simple concept; when you want to build a target, walk executes a file called Walkfile to determine:

  1. What other targets the given target depends on.
  2. How to build the target.

For example, if you wanted to build a program called prog from main.c and parse.c, you might write a Walkfile like this:

#!/bin/bash

# The first argument is the "phase", which will either be `deps` or `exec`. In
# the `deps` phase, the Walkfile should print the name of the targets that this
# target depends on.
phase=$1

# The second argument is the name of the target, like `prog`, `parse.o`, etc.
target=$2

case $target in
  prog)
    case $phase in
      # Prog depends on the object files we'll build from source. We simply
      # print each dependency on a single line.
      deps)
        echo main.o
        echo parse.o
        ;;
      exec) exec gcc -Wall -o $target $($0 deps $target) ;;
    esac ;;

  # A generic recipe for building a .o file from a corresponding .c file.
  *.o)
    case $phase in
      deps) echo ${target//.o/.c} ;;
      exec) exec gcc -Wall -o $target -c $($0 deps $target) ;;
    esac ;;

  # When invoking walk(1) without any arguments, it defaults to a target called
  # `all`.
  all)
    case $phase in
      deps) echo prog ;;
    esac ;;

  # In general, it's good practice to include a fallback rule like this, in
  # case someone tries to build a target that we don't know how to build (or
  # someone makes a typo).
  *.c|*.h) ;; # static files
  *) >&2 echo "No rule for target \"$target\"" && exit 1 ;;
esac

When you execute walk all, the following happens internally:

  1. walk resolves all of the dependencies, and builds a graph:

    $ Walkfile deps all
    prog
    $ Walkfile deps prog
    parse.o
    main.o
    $ Walkfile deps parse.o
    parse.c
    $ Walkfile deps main.o
    main.c
    $ Walkfile deps parse.c
    $ Walkfile deps main.c
  2. walk executes all of the targets, starting with dependencies:

    $ Walkfile exec parse.c
    $ Walkfile exec main.c
    $ Walkfile exec main.o
    $ Walkfile exec parse.o
    $ Walkfile exec prog
    $ Walkfile exec all

Ultimately, all of our targets end up getting invoked, and prog is built:

$ walk
ok	main.c
ok	parse.c
ok	parse.o
ok	main.o
ok	prog
ok	all

We can print the dependency graph to verify that our dependency chain is what we expect:

$ walk -p dot
digraph {
  "(root)" -> "all"
  "all" -> "prog"
  "prog" -> "main.o"
  "prog" -> "parse.o"
  "parse.o" -> "parse.c"
  "main.o" -> "main.c"
}

And that's it. Wait, that's it? That's it. walk is quite simply, just syntactic sugar over executing a binary as a graph.

See also man walk.