Coding Standards

The rules are always up for debate. However, when debate is needed, it should happen outside the source tree. In other words, if the rules are wrong, first debate the rules in IRC etc., then fix the rules, then follow the new rules.

Git commits

Make sure your name and email address are correct.

  • Use git config --global user.email foo@example.com et al.
  • It's a little unfortunate to have commits with author foo@myworkstation.local but not bad enough to rewrite history, so fix this before you push!

Refer to a story number in the first (summary) line of each commit comment. This first line should be <80 chars long, and should be followed by a blank line.

  • 1234: Remove useless button.

When merging/committing to master, refer to the story number in a way Redmine will notice. Redmine will list these commits/merges on the story page itself.

  • closes #1234, or
  • refs #1234, or
  • no issue # if no Redmine issue is especially relevant.

Use descriptive commit comments.

  • Describe the delta between the old and new tree. If possible, describe the delta in behavior rather than the source code itself.
  • Good: "1234: Support use of spaces in filenames."
  • Good: "1234: Fix crash when user_id is nil."
  • Less good: "Add some controller methods." (What do they do?)
  • Less good: "More progress on UI branch." (What is different?)
  • Less good: "Incorporate Tom's suggestions." (Who cares whose suggestions -- what changed?)

If further background or explanation is needed, separate it from the summary with a blank line.

  • Example: "Users found it confusing that the boxes had different colors even though they represented the same kinds of things."

Every commit (even merge commits) must have a DCO sign-off. See Developer Certificate Of Origin.

  • Example: Arvados-DCO-1.1-Signed-off-by: Joe Smith <joe.smith@example.com>

Full examples:

commit 9c6540b9d42adc4a397a28be1ac23f357ba14ab5
Author: Tom Clegg <tom@curoverse.com>
Date:   Mon Aug 7 09:58:04 2017 -0400

    12027: Recognize a new "node failed" error message.

    "srun: error: Cannot communicate with node 0.  Aborting job." 

    Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curoverse.com>
commit 0b4800608e6394d66deec9cecea610c5fbbd75ad
Merge: 6f2ce94 3a356c4
Author: Tom Clegg <tom@curoverse.com>
Date:   Thu Aug 17 13:16:36 2017 -0400

    Merge branch '12081-crunch-job-retry'

    refs #12080
    refs #12081
    refs #12108

    Arvados-DCO-1.1-Signed-off-by: Tom Clegg <tom@curoverse.com>

Copyright headers

Each Arvados component is released either under the AGPL 3.0 license or the Apache 2.0 license. Documentation is licensed under CC-BY-SA-3.0. See the Arvados Licenses FAQ for the rationale behind this system.

Every file must start with a copyright header that follows this format:

Code under the AGPLv3 license (this example uses Go formatting):

// Copyright (C) The Arvados Authors. All rights reserved.
//
// SPDX-License-Identifier: AGPL-3.0

Code under the Apache 2.0 license (this example uses Python formatting):

# Copyright (C) The Arvados Authors. All rights reserved.
#
# SPDX-License-Identifier: Apache-2.0

Documentation under the Creative Commons Attribution-Share Alike 3.0 United States license (this example uses textile formatting):

###. Copyright (C) The Arvados Authors. All rights reserved.
....
.... SPDX-License-Identifier: CC-BY-SA-3.0

When adding a new file to a component, use the same license as the other files of the component.

When adding a new component, choose either the AGPL or Apache license. Generally speaking, the Apache license is only used for components where integrations in proprietary code must be possible (e.g. our SDKs), though this is not a hard rule. When uncertain which license to choose for a new component, ask on the IRC channel or mailing list.

When adding a file in a format that does not support the addition of a copyright header (e.g. in a binary format like an image), add the path to the .licenseignore file in the root of the source tree. This should be done sparingly, and must be discussed explicitly as part of code review. The file must be available under a license that is compatible with the rest of the Arvados code base.

When adding a file that originates from an external source under a different license, add the appropriate SPDX line for that license. This is exceptional, and must be discussed explicitly as part of code review. Not every license is compatible with the rest of the Arvados code base.

There is a helper script at https://github.com/curoverse/arvados/blob/master/build/check-copyright-notices that can be used to check - and optionally, fix - the copyright headers in the Arvados source tree.

The actual git hook that enforces the copyright headers lives at https://github.com/curoverse/arvados-dev/blob/master/git/hooks/check-copyright-headers.sh

Source code formatting

(Unless otherwise specified by style guide...)

No TAB characters in source files. Except go programs.

  • Emacs: add to ~/.emacs(setq-default indent-tabs-mode nil)
  • Vim: add to ~/.vimrc:set expandtab
  • See Git setup below

No inline comments: this = !desired; # we don't want to do it.

No long (>80 column) lines, except in rare cases when the alternative is really clunky.

No whitespace at the end of lines. Make git-diff show you:

git config color.diff.whitespace "red reverse" 
git diff --check

What to include

No commented-out blocks of code that have been replaced or obsoleted.

  • It is in the git history if we want it back.
  • If its absence would confuse someone reading the new code (despite never having read the old code), explain its absence in an English comment. If the old code is really still needed to support the English explanation, then go ahead -- now we know why it's there.

No commented-out debug statements.

  • If the debug statements are likely to be needed in the future, use a logging facility that can be enabled at run time. logger.debug "foo"

Style mismatch

Adopt indentation style of surrounding lines or (when starting a new file) the nearest existing source code in this tree/language.

If you fix up existing indentation/formatting, do that in a separate commit.
  • If you bundle formatting changes with functional changes, it makes functional changes hard to find in the diff.

Go

gofmt, golint, etc., and https://github.com/golang/go/wiki/CodeReviewComments

Ruby

https://github.com/bbatsov/ruby-style-guide

Python

PEP-8.

Tell Emacs you don't want a blank line at the end of a multiline docstring.

(setq python-fill-docstring-style 'pep-257-nn)

JavaScript

Follow the Airbnb Javascript coding style guide unless otherwise stated:
https://github.com/airbnb/javascript

We already have 4-space indents everywhere, though, so do that.

Git setup

Configure git to prevent you from committing whitespace errors.

git config --global core.whitespace tab-in-indent,trailing-space
git config --global apply.whitespace error

Add a DCO sign-off to the default commit message.

cd .../arvados
printf '\n\nArvados-DCO-1.1-Signed-off-by: %s <%s>\n' "$(git config user.name)" "$(git config user.email)" >~/.arvados-dco.txt
git config commit.template ~/.arvados-dco.txt

Add a DCO sign-off and "refs #xxxx" comment (referencing the issue# in the name of the branch being merged) to the default merge commit message.

cd .../arvados
cat >.git/hooks/prepare-commit-message <<'EOF'
#!/bin/sh

case "$2,$3" in
    merge,)
        br=$(head -n1 ${1})
        n=$(echo "${br}" | egrep -o '[0-9]+')
        exec >${1}
        echo "${br}" 
        echo
        echo "refs #${n}" 
        echo
        echo "Arvados-DCO-1.1-Signed-off-by: ${GIT_AUTHOR_NAME} <${GIT_AUTHOR_EMAIL}>" 
        ;;
    *)
        ;;
esac
EOF
chmod +x .git/hooks/prepare-commit-message