<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE html SYSTEM "http://www.w3.org/2002/04/xhtml-math-svg/xhtml-math-svg-flat.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<title>Sane Software Manifesto</title>
<link href="style.css" type="text/css" rel="StyleSheet" />
</head>
<body>
<h1>Sane Software Manifesto</h1>
<p>In respect to user freedoms, privacy, liberty, quality, mental health and world peace we create software according to these guidelines.</p>
<h2>free software</h2>
user has freedoms
user controls his/her computer and software
non-free software can not be trusted
must be buildable using free software toolchain (like GNU/Linux + GNU GCC or OpenJDK etc.)
should not promote non-free (proprietary) software or services
copyleft licenses (like GNU GPL or GNU Affero GPL) are recommended because they guarantee software freedoms to every single end-user and prevent possibility that freedom vanishes somewhere in the distribution chain a user can't benefit from the free software albeit the software is build on originally free source code
<h2>documented</h2>
at least basic documentation must be released under a free license (GNU FDL is recommended)
there might be also other documentation/books released under any license and price
but average software engineer must be able to build and operate the software with just basic free documentation
there must be a free documentation describing building and running of the software on a fresh operating system installation including description of all dependencies
documentation should focus on all target groups: users, administrators, developers
there must be a big picture and software architercure described
<h2>semantic versioning</h2>
strongly recommended
required when the package is suposed to be used as dependency by others
if there is need of some marketing or cool versioning/codenames like Ultrasonic Umbrella or 2016, they should be used in addition to semantic versioning, not instead it
once publicly released, the package must not be changed anymore – if a change (even small fix) is needed, new version number must be assigned
APIs, file formats and protocols might be semanticly versioned independently from the implementation
in such case, there should be a table documenting which API/format/protocol version matches which implementation version
<h2>compatibilible with itself</h2>
focus on backward compatibility
don't broke things
incompatible changes must be planned and announced in advance
upgrade scripts + upgrade documentation
<h2>compatibilible with others</h2>
use open standards (protocols, formats) if they exist
define own open standards if needed
also standards must be semantically versioned
should be written in machine readable format (WSDL, WADL, ASN.1, XSD, Diameter dictionary etc.) or at least formal language (Backus–Naur Form, EBNF etc.)
also configuration should have machine readable description and should be testable by executing a command
<h2>modular architecture</h2>
larger and multi-purpose software should be divided into smaller modules
modules must have defined dependencies (less = better)
particular modules should be compilable and executable separately
whole system should be compilable with only selected modules – must not require compilation or even distribution of all modules, if they are not necessary
<h2>extensible</h2>
able to be extended
by configuration (RegExp, SQL, XSLT, XPath etc.)
by scripting (Guile, ECMA Script etc.)
and/or third-party plugins/modules
it should be easy to create a third-party module and plug it in an existing system
dependencies needed to write an extension (i.e. header files, API classes/interfaces) should be as small as possible (do not require large codebase to write an extension)
there should be public directory of extensions/scripts
<h2>testable</h2>
there should be automated build-time complex tests for the package – feed the program with sample input and verify expected output
there should be also automated runtime/postinstall tests – in order to verify that software was installed properly, all required dependencies are met and basic function is guaranteed – the program should report problem during its start (as a warning if it is not fatal), instead of unexpected failures during operation
unit tests are recommended for code parts that are internally complex (algorithms, important business logic) and have simple interfaces
<h2>safe code and sustainability</h2>
correctness, safety and readability is prefered to performance
use strong data typing, declare preconditions and possible exceptions
data structures must be known and well documented – don't use undocumented map keys or properties
code, comments and analysis should be written in the same natural language
there should be a dictionary of used terms, so whole team and also users and customers will speak same language
<h2>small code footprint</h2>
less LOC (resp. complexity) = better
reduce boilerplate and unused code
use code generators (during build process, not to generate code to be manually edited and versioned)
<h2>sane dependencies</h2>
avoid NIH and reuse code but also avoid dependency hell
know your dependencies, know why they are required
reduce dependencies to only necessary ones
depend on small and useful libraries – not on bulky application packages
if dependency on bulky application package is inevitable, add a layer of abstraction – create a generic interface and connector and allow others to replace the bulky package with their own sane implementation
helper tools:
if you e.g. use Bash and Perl during the build process, don't add also Python dependency, write it in Perl – or use Python instead of Perl.
Or if you use Java as your main language, consider not using Python/Perl for scripting and use Java for it
if possible, always depend on abstract interfaces, not on particular implementations
<h2>easily auditable</h2>
small code footprint and minimal dependencies makes it easy to do security audit
avoid ungrounded refactoring and reformatting – they make mess and noise in the version control system and impede the audit
refactoring/reformatting changesets should be separated from substantive changes
<h2>reproducible builds</h2>
builds should be reproducible: same code/version → same binary package
if not, it should be documented, why and how build products mihgt differ, and there should be plan/task to make it reproducible
<h2>trustworthy packages and sources</h2>
every released version (binary or source) is cryptographically signed by the authors (GnuPG/OpenPGP is strongly recommended)
if HTTP is supported, HTTPS should also be – the attacker/eavesdropper should not even know what software/package/update is downloaded by the user
the attacker should not be able to suppress updates – the program must not be silent in such case and must warn the user that something possibly nasty and dangerous is happening
releases should be downloadable also (or exclusively) over BitTorrent or other P2P network
there should be also checksums/hashes for every package
source code repository is accessible through an encrypted connection
<h2>network interactions</h2>
no network connection is needed during build – build must be possible completely offline, all dependencies must be downloadable and documented including secure hashes or better cryptographic signatures
if dependencies are optionally automatically downloaded during/before build, the packaging system must cryptographically verify that that they are undamaged
avoid unwanted network interactions during runtime – no „call home“ or update-checks without user's explicit consent
if any network connection is used, it must be cryptographically secured against MITM attacks
<h2>localized/internationalized</h2>
is is strongly recommended that it should be possible to localize the user interface independently from the original author by writing a language pack
GNU Gettext or other standard framework (like Java resource bundles) should be used
error messages should have assigned unique error codes, so it is possible to find relevant information regardless current locale
GEC is recommended for such unique error identifiers
data formats and protocols must be language/locale independent
e.g. use decimal point instead of comma and no thousand separators for numbers, use standardized date formats
in general: everything that is expected to be machine-readable or machine-generated must be independent from current locale
character encoding:
always be aware of it, don't just blindly use current platform's default (because the other side might run on different platform with different default)
if given software/format/protocol has some default encoding, it must be clearly defined in its specification and this default should not be changed without changing the major version number
if there is no default, the encoding must be specified in the metadata attached (e.g. protocol headers, extended attributes on filesystem) to the actual data or at least at the begining of the data (like declaration in XML format)
the metric system should be used as default
<h2>communication channels</h2>
RSS/Atom or other machine readable format for:
security announcements
new version announcements
infrastructure outage announcements
blog, documentation, how-tos etc.
AFK events (conferences, meetings, hackatons etc.), for calendar data iCal format is strongly recommended
mailing list
e-mail/smtp
use TLS
use DKIM/ADSP
use signed and encrypted messages (GnuPG or X.509)
avoid spam and viruses, don't spam the users, don't push them to subscribe your „newsletter“ – always offer also anonymous channel like RSS/Atom
Jabber MUC or IRC
discussion forum
don't push users to register at a proprietary social networks resp. at particular company like Facebook – users without such account must not be discriminated – use open and decentralized networks/protocols instead
Q&A tool + FAQ
there should be a second-level internet domain for the project or its team
but don't buy an internet domain if you are not prepared to mainain it for decades – rather use third level domain under some reliable second level domain maintained by a credible group or person – think of that every expired domain helps spammers and scammers and hurts the users
URLs should be as stable as possible (don't broke old links, set up redirections if needed)
the website must be independent and must contain everything needed – any content (JavaScripts, CSS, fonts etc.) downloaded from othed domains must not be required to browse/use the website
authors should publish their public keys (GnuPG/OpenPGP or X.509)
crpyptographically secured e-mail address or web form for receiving security vulnerabilities report
every security incident must be clearly documented and investigated – don't obscure it
<h2>open development – has public:</h2>
source code repository (versioning system), not just source code releases
description of the process of accepting external patches
feature/bug tracking system
roadmap of future releases
plan of supported versions/branches
every release/version/branch must clearly declare the status (alpha, beta, prototype, stable, retired, deprecated…)
</body>
</html>