Return to index

This document is extremely new, and may change rapidly.

For context, please also read the main lbmk maintenance manual.

You should read the logic in lbmk yourself, to really know what is meant by some of the concepts explained here. This article will no doubt be incomplete, and several practises may persist in spite of it; nonetheless, this article shall serve as a reference for lbmk development.

Libreboot’s build system was previously written in Bash, and actually used Bash-specific behaviour. This was later corrected, thanks largely to work done by Ferass El Hafidi.

Here is an excellent introduction to posix sh scripting: https://pubs.opengroup.org/onlinepubs/009604499/utilities/xcu_chap02.html

and an even more excellent introduction: https://vermaden.wordpress.com/ghost-in-the-shell/ (seriously, it’s good. Read it!)

Libreboot’s build system design is very simple: put as much as possible under config/, and keep actual logic to a minimum.

You can read about that design in the lbmk maintenance manual.

We have Makefiles in some C programs, under util/, and projects that we import may use Makefiles, but lbmk itself does not contain any Makefiles. Instead, we do everything in shell scripts.

This approach has certain drawbacks, but for the most part it ensures that the code is more readable. It’s easier to implement a cleaner coding style, which the next sections will cover.

Read https://man.openbsd.org/style.9 and go read a few userland program source trees in OpenBSD’s main CVS tree. This is the style that inspires the lbmk coding style; OpenBSD’s style pertains to C programming, and it has been adapted for shell scripts in the Libreboot build system, lbmk.

You should read the OpenBSD style and go read OpenBSD utils, especially userland programs like cat or ls in the OpenBSD src tree.

Libreboot scripts, and also C programs like nvmutil, are heavily inspired by this style. We insist on its use, because this style is extremely readable and forces you to write better code.

In every lbmk script, it is our intention that there be a main() function. All logic should be inside a function, and main() should be the function that executes first; at the bottom of each script, insert this line:

main $@

This will execute main(), passing any arguments (from the user’s shell) to it.

Every function called from main should always be below the calling function. Therefore, if multiple functions call a given function, that function should be below the final one that called it. Here is an example (please also pay attention to how the functions are formatted, e.g. where { and } go:

#!/usr/bin/env sh

. "include/err.sh"

main()
{
	foo
	bar
	do_something_else
}

foo()
{
	printf "I'm a function that does stuff.\n"
	bar || err "foo: an error occured"
	do_something_else
}

bar()
{
	printf "I'm another function that does stuff.\n"
	some_other_command || printf "WARNING: bar: something something" 1>&2
}

do_something_else()
{
	complicated_function bla bla bla || \
	    err "do_something_else: something happened that wasn't nice"
}

complicated_function()
{
	printf "I'm a complicated function, provided as helper"
	printf " function for do_something_else()\n"

	do_some_complicated_stuff || return 1
}

main $@

In any script executed by lbmk, under script/, the work directory is relative to the main lbmk script. In other words, all scripts under script/ also assume this.

This is actually one of the reasons for that design, as also alluded to in the main lbmk maintenance manual.

The main() function should not implement much logic itself. Each script in lbmk is its own program. The main() function should contain the overall structure of the entire logic, with subfunctions providing actual functionality.

Subfunctions can then have their own subfunctions, declared below themselves, in this top-down style. For example, a function that builds SeaBIOS payloads might be below a function that builds ROM images with SeaBIOS payloads inside them, when building coreboot ROM images.

Not literally one task, but one theme, one kind of overall task. For example, script/build/roms builds final ROM images of coreboot, containing payloads; that same script does not also build cross compilers or tell you the current weather forecast. This is an analog of the Unix design philosophy which says: write one program that does one thing well, and then another program that does another thing very well; programs communicate with each other via the universal method, namely text.

Where feasible, a script should do:

set -e -u

If -e isn’t feasible, perhaps try just -u - if neither is feasible, then that is OK. Judge it case by case.

However, neither of these should be relied upon exclusively. When a script runs any kind of command that could return with error status, that error status must be handled.

The general rule is to call err(), which is provided in lbmk by the file include/err.sh. This is inspired by the way err() is called in BSD programs (from err.h, a non-standard BSD libc extension).

Where a script must perform certain cleanup before exiting, the script should implement its own fail() function that performs cleanup, and then calls err(). The err() function takes a string as argument, which will be printed to the screen.

If err is being called from main(), just write the error message. However, if it’s being called from another function, you should write the function name. For example:

err "function_name: this shit doesn't work. fix it."

Please try to use err for all error exits.

The main lbmk script has its own exit function, for handling zero or non-zero exits. Zero means success, and non-zero means error.

A script should either return zero status, or call err().

An individual function may, in some cases, return 1 or 0 itself, which would then be handled accordingly by the calling function.

There are some instances where errors should be ignored, in which case you might do:

command || :

The || means: if command exits with non-zero (error) status, do this, and then after the || is what to do: similarly, && instead would mean: if the command succeeded, then do this.

Never mix && and ||

Keep these simple, and where possible, maybe don’t use them at all! For example:

if [ "${var}" = "foo" ]; then
	do_something
fi

You might instead do:

[ "${var}" != "foo" ] || \
	do_something

or

[ "${var}" = "foo" ] && \
	do something

In C, the stderr file is 2 as represented by int fd style. In shell scripts, it’s the same: 1 for standard output, 2 for errors/warnings. The err function in lbmk writes to 2 (stderr).

If you want to output something that is a warning, or otherwise an error that should not yield an exit, you should do something like this:

printf "function_name: this is dodgy stuff. fix it maybe?\n" 1>&2

In functions, use of arguments passed to them can be useful, but in general, they should be avoided; use global variables when feasible.

See: RFC 3676

Excessively long code lines are really annoying to read.

A new line should begin with tab indentation, in a function.

Use \ at the end, as you would, but use four spaces to indent on the follow-up line. For example:

function_name()
{
	really stupidly long command that may also return error state || \
	    err "function_name: you fucked up. try again."
}

Don’t use echo unless there’s some compelling reason to do so.

The printf functionality is more standard, across various sh implementations.

Don’t do:

#!/bin/sh

Do:

#!/usr/bin/env sh

This is more portable, between various Unix systems.

In addition to not using bashisms, commands that lbmk uses must also be portable; where possible, third party projects should be tweaked.

This is actually something that is currently lacking or otherwise untested in Libreboot; it’s currently assumed that only Linux (specifically GNU+Linux) will work, because many of the projects that Libreboot makes use of will use bashisms, or other GNUisms (e.g. GNU-specific C extensions or GNU Make specific behaviour in Makefiles).

Work+testing is sorely needed, in this area. It would be nice if Libreboot could be built on BSD systems, for example.

Don’t over-engineer anything. Write as simply as you can, to perform a single task. This is basically the same as what has been written elsewhere, but it’s re-stated this way to illustrate a point:

Libreboot’s build system is designed to be as efficient as possible. It intentionally avoids implementing many things that are unnecessary for the user. The purpose of Libreboot is to provide coreboot ROM images as efficiently as possible, with desirable configurations that users want. Do that in as few steps as possible, in the most streamlined way possible, while still providing a degree of configurability - this is the mentality behind lbmk design.

Markdown file for this page: https://libreboot.org/docs/maintain/style.md

Subscribe to RSS for this site

Site map

This HTML page was generated by the Untitled Static Site Generator.