Jump to navigation Jump to search
This page is a translated version of the page Manual:Coding conventions and the translation is 31% complete.

Outdated translations are marked like this.
Other languages:
Deutsch • ‎English • ‎dansk • ‎español • ‎français • ‎italiano • ‎polski • ‎português • ‎português do Brasil • ‎русский • ‎العربية • ‎تۆرکجه • ‎සිංහල • ‎中文 • ‎日本語
ショートカット: CC

このページでは、MediaWiki コードベース内や、ウィキメディアのウェブサイト群で使用されることを意図している拡張機能内で使用される、コーディング規約を説明します。 Changes that do not comply with these conventions may be summarily -1'ed by code reviewers; this should be considered an invitation to fix the stylistic issues and update the patch.

このページでは、コードがどの言語で書かれているかにかかわらずすべての MediaWiki コードに適用される全般的な規約を列挙します。 MediaWiki の特定のコンポーネントやファイルの種類に適用されるガイドラインについては、以下を参照してください:

On wikitech (applies at least to operations/puppet):




Lines should be indented with a single tab character per indenting level. You should make no assumptions about the number of spaces per tab. Most MediaWiki developers find 4 spaces per tab to be best for readability, but many systems are configured to use 8 spaces per tab and some developers might use 2 spaces per tab.

For vim users, one way to establish these settings is to add the following to $HOME/.vimrc:

autocmd Filetype php setlocal ts=4 sw=4

with similar lines for CSS, HTML, and JavaScript.

However, for Python, instead follow the whitespace guidelines from PEP 8, which recommends spaces for new projects.


All files should use Unix-style newlines (single LF character, not a CR+LF combination).

  • git on Windows will (by default) convert CR+LF newlines to LF during committing.


  • It makes sense since all other lines have a newline character at the end.
  • It makes passing data around in non-binary formats (like diffs) easier.
  • Command-line tools like cat and wc don't handle files without one well (or at least, not in the way that one would like or expect).


All text files must be encoded with UTF-8 without a Byte Order Mark.

Do not use Microsoft Notepad to edit files, as it always inserts a BOM. A BOM will stop PHP files from working since it is a special character at the very top of the file and will be output by the web browser to the client.

In short, make sure your editor supports UTF-8 without BOM.


When using an IDE, pressing the Home and End keys (among other keyboard shortcuts) usually ignores trailing whitespace and instead jumps to the end of the code, which is intended. In non-IDE text editors, though, pressing End will jump to the very end of the line, which means the developer must backspace through the trailing whitespace to get to the spot where they actually want to type.

Removing trailing whitespace is a trivial operation in most text editors. Developers should avoid adding trailing whitespace, primarily on lines that contain other visible code.

Some tools make it easier:

  • nano: craigbarnes/nanorc;
  • Komodo Edit: in the Save Options from menu "Edit > Preferences", enable "Clean trailing whitespace and EOL markers" and "Only clean changed lines";
  • Kate: you can see trailing spaces by enabling the option "Highlight trailing spaces". This option can be found in "Settings > Configure Kate > Appearance". You can also tell Kate to cleanup trailing spaces on save in "Settings > Configure Kate > Open/Save".
  • vim: various automatic cleanup plugins;

キーワード =

Do not use parentheses with keywords (e.g. require_once, require) where they are not necessary.



MediaWiki's indenting style is similar to the so-called "One True Brace Style". Braces are placed on the same line as the start of the function, conditional, loop, etc. The else/elseif is placed on the same line as the previous closing brace.

function wfTimestampOrNull( $outputtype = TS_UNIX, $ts = null ) {
    if ( $ts === null ) {
        return null;
    } else {
        return wfTimestamp( $outputtype, $ts );

Multi-line statements are written with the second and subsequent lines being indented by one extra level:

Use indenting and line breaks to clarify the logical structure of your code. Expressions which nest multiple levels of parentheses or similar structures may begin a new indenting level with each nesting level:

$wgAutopromote = [
    'autoconfirmed' => [ '&',
        [ APCOND_EDITCOUNT, &$wgAutoConfirmCount ],
        [ APCOND_AGE, &$wgAutoConfirmAge ],

In switch statements indent within the braces and between cases:

switch ( $word ) {
    case 'lorem':
    case 'ipsum':
        $bar = 2;
    case 'dolor':
        $bar = 3;
        $bar = 0;


Avoid vertical alignment. It tends to create diffs which are hard to interpret, since the width allowed for the left column constantly has to be increased as more items are added.

Most diff tools provide options to ignore whitespace changes.
Git: git diff -w

When needed, create mid-line vertical alignment with spaces rather than tabs. For instance this:

$namespaceNames = [
    NS_MEDIA            => 'Media',
    NS_SPECIAL          => 'Special',
    NS_MAIN             => '',

Is achieved as follows with spaces rendered as dots:

 →  NS_MEDIA············=>·'Media',
 →  NS_SPECIAL··········=>·'Special',
 →  NS_MAIN·············=>·'',

(If you use the tabular vim add-on, entering :Tabularize /= will align the '=' signs.)


Lines should be broken at between 80 and 100 columns. There are some rare exceptions to this. Functions which take lots of parameters are not exceptions.

The operator separating the two lines should be placed at the end of the preceding line.

return strtolower( $val ) == 'on' ||
    strtolower( $val ) == 'true' ||
    strtolower( $val ) == 'yes' ||
    preg_match( "/^\s*[+-]?0*[1-9]/", $val );
    Xml::fieldset( wfMessage( 'importinterwiki' )->text() ) .
        Xml::openElement( 'form', [ 'method' => 'post', 'action' => $action, 'id' => 'mw-import-interwiki-form' ] ) .
        wfMessage( 'import-interwiki-text' )->parse() .
        Xml::hidden( 'action', 'submit' ) .
        Xml::hidden( 'source', 'interwiki' ) .
        Xml::hidden( 'editToken', $wgUser->editToken() ),

When continuing "if" statements, a switch to Allman-style braces makes the separation between the condition and the body clear:

if ( $.inArray( mw.config.get( 'wgNamespaceNumber' ), whitelistedNamespaces ) !== -1 &&
    mw.config.get( 'wgArticleId' ) > 0 &&
    ( mw.config.get( 'wgAction' ) == 'view' || mw.config.get( 'wgAction' ) == 'purge' ) &&
    mw.util.getParamValue( 'redirect' ) !== 'no' &&
    mw.util.getParamValue( 'printable' ) !== 'yes'
) {

Opinions differ on the amount of indentation that should be used for the conditional part. Using an amount of indentation different to that used by the body makes it more clear that the conditional part is not the body, but this is not universally observed.

Continuation of conditionals and very long expressions tend to be ugly whichever way you do them. So it's sometimes best to break them up by means of temporary variables.


Do not write "blocks" as a single-line. They reduce the readability of the code by moving important statements away from the left margin, where the reader is looking for them. Remember that making code shorter doesn't make it simpler. The goal of coding style is to communicate effectively with humans, not to fit computer-readable text into a small space.

// No:
if ( $done ) return;

// No:
if ( $done ) { return; }

// Yes:
if ( $done ) {

This avoids a common logic error, which is especially prevalent when the developer is using a text editor which does not have a "smart indenting" feature. The error occurs when a single-line block is later extended to two lines:

if ( $done )

Later changed to:

if ( $done )

This has the potential to create subtle bugs.


In emacs, using php-mode.el from nXHTML mode, you can set up a MediaWiki minor mode in your .emacs file:

(defconst mw-style
  '((indent-tabs-mode . t)
    (tab-width . 4)
    (c-basic-offset . 4)
    (c-offsets-alist . ((case-label . +)
                        (arglist-cont-nonempty . +)
                        (arglist-close . 0)
                        (cpp-macro . (lambda(x) (cdr x)))
                        (comment-intro . 0)))
        (defun-open after)
        (block-open after)

(c-add-style "MediaWiki" mw-style)

(define-minor-mode mah/mw-mode
  "tweak style for mediawiki"
  nil " MW" nil
  (tabify (point-min) (point-max))
  (subword-mode 1)) ;; If this gives an error, try (c-subword-mode 1)), which is the earlier name for it

;; Add other sniffers as needed
(defun mah/sniff-php-style (filename)
  "Given a filename, provide a cons cell of
   (style-name . function)
where style-name is the style to use and function
sets the minor-mode"
  (cond ((string-match "/\\(mw[^/]*\\|mediawiki\\)/"
         (cons "MediaWiki" 'mah/mw-mode))
         (cons "cc-mode" (lambda (n) t)))))

(add-hook 'php-mode-hook (lambda () (let ((ans (when (buffer-file-name)
                                                 (mah/sniff-php-style (buffer-file-name)))))
                                      (c-set-style (car ans))
                                      (funcall (cdr ans) 1))))

The above mah/sniff-php-style function will check your path when php-mode is invoked to see if it contains “mw” or “mediawiki” and set the buffer to use the mw-mode minor mode for editing MediaWiki source. You will know that the buffer is using mw-mode because you'll see something like “PHP MW” or “PHP/lw MW” in the mode line.

Data manipulation

URL の構築

Never build URLs manually with string concatenation or similar. Always use the full URL format for requests made by your code (especially POST and background requests).

You can use the appropriate LinkerManual:Linker.php or TitleManual:Title.php method in PHP, the fullurlHelp:Magic words#fullurl magic word in wikitext, the mw.util.getURL() method in JavaScript, and similar methods in other languages. You'll avoid issues with unexpected short URL configuration and more.


Files which contain server-side code should be named in UpperCamelCase. This is also our naming convention for extensions.[1] Name the file after the most important class it contains; most files will contain only one class, or a base class and a number of descendants. For example, Title.php contains only the Title class; WebRequest.php contains the WebRequest class, and also its descendants FauxRequest and DerivativeRequest.

Access point files

Name "access point" files, such as SQL, and PHP entry points such as index.php and foobar.sql, in lowercase. Maintenance scripts are generally in lowerCamelCase, although this varies somewhat. Files intended for the site administrator, such as readmes, licenses and changelogs, are usually in UPPERCASE.

Never include spaces in file names or directories, and never use non-ASCII characters. For lowercase titles, hyphens are preferred to underscores.

JS, CSS, and media files

For JavaScript, CSS and media files (usually loaded via ResourceLoader) file naming should match module naming. For example:

  • Module might have files resources/src/ and resources/src/
  • module mediawiki.Title has file resources/src/mediawiki.Title.js

Modules names registered by extensions should follow name like ext.myExtension, for instance:

  • extensions/FooBar/resources/ext.fooBar/init.js

This keeps it easy to find files, even if you divide up a module into smaller files for editing convenience and then bundle them together into a single module.


The language-specific subpages have more information on the exact syntax for code comments in files, e.g. comments in PHP for doxygen. Using precise syntax allows us to generate documentation from source code at

Some elements of MediaWiki are documented in core's /docs folder. For instance, if you add a new hook, you should update /docs/hooks.txt with the name of the hook, a description of what the hook does, and the parameters used by the hook.

テキスト ファイル

Developers can keep documentation files in Git alongside code. This can be good for detailed documentation of extension architecture, database design, etc. that you should update with each code commit that changes behavior. Pages on that relate to documentation in Git should link to it using {{git file}}.

(The possibility to transclude text from Git files into wiki pages is tracked in phab:T91626.)

Note that many technical documentation pages on pages document the evolution of MediaWiki code over many releases. Either describe changes in your document or state it only describes the latest codebase in "master".

テキスト ファイルの書式

If your text file is wikitext, give it a .wiki extension. GitHub can parse a subset of wikitext, so files mirrored on GitHub will display some formatting (a .mediawiki extension also works, but is longer). For example, the Wikibase extension's docs/ in GitHub.

Doxygen supports Markdown formatting, so you can put lightly-formatted documentation in .md files. Diffusion and GitHub also support .md files. Name the explanatory file for a directory or project; Diffusion and GitHub will display this file when you view that directory or project (e.g. RESTbase's doc/development/, in Diffusion and on GitHub).
no extension and .txt
Doxygen by default parses these as C language files (!!, tracked in タスク T106116). You can take advantage of this by making the file mimic a C comment, and then add doxygen directives to the file. For example, includes/filebackend/README generates File backend design in doxygen, and begins with:
\ingroup FileBackend
\page file_backend_design File backend design

Some notes on the FileBackend architecture.
Special:Version/Credits assumes AUTHORS and CREDITS (with no extension) are wikitext files.


In order to be compliant with most licenses you should have something similar to the following (specific to GPLv2 PHP applications) at the top of every source file.

 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * GNU General Public License for more details.
 * You should have received a copy of the GNU General Public License along
 * with this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 * @file


Licenses are generally referred to by their full name or acronym as per SPDX standard. See also Manual:$wgExtensionCredits#license.


You must document all significant changes (including all fixed bug reports) to the core software which might affect wiki users, server administrators, or extension authors in the RELEASE-NOTES-N.NN file. RELEASE-NOTES-1.32 is in development; on every release we move the past release notes into the HISTORY file and start afresh. RELEASE-NOTES-N.NN is generally divided into three sections:

  • Configuration changes is the place to put changes to accepted default behavior, backwards-incompatible changes, or other things which need a server administrator to look at and decide "is this change right for my wiki?". Try to include a brief explanation of how the previous functionality can be recovered if desired.
  • Bug fixes is the place to note changes which fix behavior which is accepted to be problematic or undesirable. These will often be issues reported in Phabricator, but needn't necessarily.
  • New features is, unsurprisingly, to note the addition of new functionality.

There may be additional sections for specific components (e.g. the Action API) or for miscellaneous changes that don't fall into one of the above categories.

In all cases, if your change is in response to one or more issues reported in Phabricator, include the task ID(s) at the start of the entry. Add new entries in chronological order at the end of the section.


When creating a new system message, use hyphens (-) where possible instead of CamelCase or snake_case. So for example, some-new-message is a good name, while someNewMessage and some_new_message are not.

If the message is going to be used as a label which can have a colon (:) after it, don't hardcode the colon; instead, put the colon inside the message text. Some languages (such as French which require a space before) need to handle colons in a different way, which is impossible if the colon is hardcoded. The same holds for several other types of interpunctuation.

Try to use message keys "whole" in code, rather than building them on the fly; as this makes it easier to search for them in the codebase. For instance, the following shows how a search for templatesused-section will not find this use of the message key if they are not used as a whole.

// No:
return wfMessage( 'templatesused-' . ( $section ? 'section' : 'page' ) );

// Yes:
$msgKey = $section ? 'templatesused-section' : 'templatesused-page';
return wfMessage( $msgKey );

If you feel that you have to build messages on the fly, put a comment with all possible whole messages nearby:

// Messages that can be used here:
// * myextension-connection-success
// * myextension-connection-warning
// * myextension-connection-error
$text = wfMessage( 'myextension-connection-' . $status )->parse();

See Localisation for more conventions about creating, using, documenting and maintaining message keys.


It is just as important to have consistent spelling in the UI and codebase as it is to have consistent UI. By long standing history, 'American English' is the preferred spelling for English language messages, comments, and documentation.


プレースホルダー (placeholder: 入力欄内のテキスト)
ツールチップ テキスト (tooltip text)
利用者の個人設定内のトグル (toggle) オプション (訳注: チェックボックス)


Non-title error messages are considered as sentences and should have punctuation.


Module names should match the main definition of the scripts they load. For example a module defining the mw.util object is named "mediawiki.util" and the module for the mw.Title object constructor is named "mediawiki.Title".

コア モジュール

If you need some additional functionality of a core module (or you need a function that does something similar but a different), actually improve the core module. Don't just copy+paste and modify the code in another place.


Refactor code as changes are made: don't let the code keep getting worse with each change.

However, use separate commits if the refactoring is large. See also Architecture guidelines (draft).


Given HTML can be generated by MediaWiki's PHP code as well as wikitext created by users, especially templates in the "Template:" namespace, HTML going through the parser is subject to sanitization.

While most data-... attributes can be added by on-wiki templates, specific names are disallowed to allow client JavaScript code to determine whether the HTML is coming from a trusted source:

  • data-ooui attribute is present for HTML generated by OOUI
  • data-mw and data-parsoid are reserved for Parsoid
  • data-mw-... are reserved for extensions or core

JavaScript when selecting elements, can thus check the existence of one of these data attributes to verify the HTML it is working with has come from a trusted source. Example: Don't trigger 'wikipage.diff' hook for content that looks like a diff.