PHP coding conventions
This document provides guidelines for code formatting and documentation to developers contributing to Silverstripe CMS core and supported modules.
These conventions will work well in most situations, though sometimes you may need to not follow them for practical reasons.
These coding conventions are for new code, and for situations where you are intentionally refactoring code for the purpose of improving coding standards (usually as part of a major release). Existing Silverstripe CMS code is inconsistent as to whether or not it follows these conventions.
General
- Aim for compliance with the PSR-12 coding standards.
- Prioritise the human readability of code over "clever" code that reduces the number of lines at the expense of readability.
- All symbols and documentation should use UK-English spelling, for example "behaviour" instead of "behavior" except when necessitated by third party conventions, for example using PHP's
Serializable
interface. - Avoid writing HTML in PHP code. Instead, put HTML in template files.
Naming conventions
Casing
- Follow PSR-1 for naming class names, method names, and class constants
- Use snake_case for configuration properties (i.e. any
private static
properties not annotated with@internal
) - Use camelCase for all other class properties
Accessors
- Prefix with
get
for getter methods andset
for setter methods, for examplegetValue()
andsetValue()
.
Class name suffixes and prefixes
Use an appropriate suffix or prefix for classnames when making a subclass or implementing an interface. Usually the suffix will be the name of the parent class or the interface. Sometimes the suffix/prefix is a shortened version of the name of the parent because it reads better while retaining easy comprehension. Here are some common examples with the parent class or interface in brackets:
Admin
(ModelAdmin
,LeftAndMain
if included in the CMS Menu)Block
(BaseElement
)DB
(DBField
) - use as a prefix, e.g.DBString
Controller
(Controller
)Exception
(Exception
)Extension
(Extension
)Factory
(Factory
)Filter
(SearchFilter
)Form
(Form
)Field
(FormField
)Handler
(RequestHandler
)Page
(SiteTree
)Job
(AbstractQueuedJob
)Middleware
(HTTPMiddleware
)Task
(BuildTask
)
Directory naming - src dir
- Follow the PSR-4 coding standard
- Use initial caps pluralised for the directory naming. For example, use
Controllers
rather thanController
orcontrollers
- Put similar types of classes in the same directory. For example, put Controllers in a
Controllers
directory - Put
DataObject
subclasses in a directory calledModels
- If you end up with a lot of classes of a specific type with some clear grouping then you can use subdirectories to group them. For example use
src/Models/Sports
if you have a lot of sports related models.
Properties and accessors
- For setter methods use a combination of a
static
return type andreturn $this
to allow for fluent programming. - Do not use dynamic properties
Method, property, and constant visibility
- Avoid
public
properties, instead use a combination ofprivate
(orprotected
) properties andpublic
getter and setter methods. - Prefer
private
overprotected
for all new properties. Subclasses can access the properties through the public getter and setter methods. -
Prefer
private
overprotected
for new methods, unless:- There is a clear use case for customising the specific functionality of that method on its own and the method implementation is unlikely to change in the near future
- The method is intended to be called by subclasses.
- Use
public
for constants that should be accessed outside the class hierarchy, - Prefer
private
overprotected
for non-public constants, unless they are intended to be used in subclasses. - If someone requests a
private
method, property, or constant be madeprotected
so that they can use it in their project or module code, that change should be made unless there's a really good reason not to.
Interfaces
- Create an interface for type hinting if there are (or are likely to be in the future) multiple classes implementing a specific feature. Use the interface for type hints (for example for parameters, properties, and return types), and as the service key for injector.
- If there is only a single implementation and no foreseeable expectation of more implementations being added, then do not create an interface for the single class implementation because this would only be adding unnecessary complexity.
Typing
- Strongly type all new properties, method parameters, and method return types. This includes files where there is no existing strong typing.
- Only provide docblock types for strongly typed parameters or return values if more context is provided.
Mixed types
- Avoid using mixed types, including type unions, because this leads to complicated and brittle code.
- Avoid using nullable scalars and instead use the default value of the scalar instead of null. One exception to this is for when
null
means 'uninitialised'. - Avoid using
func_get_args()
.
Object instantiation
- For classes with the
Injectable
trait, use the injector to instantiate objects using thecreate()
static method rather than thenew
keyword. - For classes without the
Injectable
trait, either add theInjectable
trait to the class or use thenew
keyword to instantiate it. - For third-party classes, use
Injector
directly to instantiate objects rather than thenew
keyword. - For unit tests, always use the
new
keyword when instantiating objects unless you are explicitly testing the injector logic.
Extensions and traits
- Use the
Extension
class for extending classes, includingDataObject
subclasses. - Do not use the
DataExtension
class, it will be deprecated in a future release.
Use a trait instead of an Extension
when the composable functionality:
- Relies on per-instance properties, because they cannot be used with extensions
- Does not add or modify configuration properties or values
- Is not expected to be applied to third-party code.
Use an Extension
instead of a trait when the composable functionality:
- Does not rely on per-instance properties
- Adds or modifies configuration properties or values
- Could be expected to be applied to third-party code.
Extension hooks
- For hooks that update the value that's returned, name the hook
update{MethodName}
whereMethodName
has any prefixget
removed. For examplegetMyValue()
would have an extension hook namedupdateMyValue()
. - For hook that update a variable in the middle of a method, name the hook
update{Variable}
whereVariable
is the name of the variable being updated. For example if a method has a variable called$myValue
then the hook would be namedupdateMyValue()
. - For hooks that update method parameters before running the body of the method, name the hook
onBefore{MethodName}
. For instance if a method is calledprocessSomething()
call the hookonBeforeProcessSomething()
. - If the method has an
onBefore
hook and also has a hook at the end of the method, then second hook should be calledonAfter{MethodName}
.
Class member ordering
Order code in classes in the following order:
- Class constants
- Configuration properties (i.e. private static properties that don't have the
@internal
annotation) - Static properties
- Member properties
- Static methods
- Controller action methods
- Commonly used methods like
getCMSFields()
- Accessor methods (
getMyField()
andsetMyField()
) - Template data-access methods (methods that will be called by a
$MethodName
or<% loop $MethodName %>
construct in a template somewhere) - Object methods
Inline comments
- If there is a non-obvious reason for a code operation, include inline comments that explain why we are doing this operation. Remember that while something may be obvious to you now, it will not be obvious to someone else in 6 months.
- Self explanatory code with well-named methods and variables is preferred to inline comments that explain what the code does, though it's still OK to include these sorts of comments if you think it makes the code easier to follow.
Unit testing
- Unit test method names should match the method that they're testing with the first letter of the method uppercased and the word
test
prefixed (for example for the methodmyMethodName()
the unit test method should be calledtestMyMethodName()
). - Use the
@dataProvider
annotation to provide test case data when you are testing the same method in multiple scenarios. It makes code much cleaner and it makes it very easy to add further test cases. - Data provider method names should be same as the test case method name they're providing for with the leading work
test
substituted forprovide
(for example fortestSomething()
the DataProvider isprovideSomething()
). - Data provider array keys should describe the scenario they're testing.
- Unless you need to explicitly create dynamic fixtures, fixtures should be added via YAML fixture files.
- It's usually OK to use reflection to change the visibility of a private or protected method so that it can be tested independently. It is generally preferable to keep a smaller public API surface and use reflection to test functionality, rather than making the API public just so it can be tested.
Raw SQL
- Avoid writing raw SQL and instead use ORM methods.
- If you do write raw SQL, use double quotes around table/column names and use parameterised queries, for example
->where(['"Score" > ?' => 50])
. - Use ANSI SQL format.
PHPDoc
PHPDocs are not only useful when looking at the source code, but are also used in the API documentation at https://api.silverstripe.org.
- All public API should have a PHPDoc to describe its purpose, unless the API name and signature describe everything about its usage and purposes without any additional context.
-
All
DataObject
andExtension
subclasses must have@method
annotations in their PHPDoc for relations (has_one
,has_many
, etc).- Do not add the annotation if a real method is declared with the same name.
- Include the relevant
use
statements for all classes mentioned in the annotation - Return types should include generics, e.g.
@method HasManyList<Member> Members()
Other conventions
- Prefer the identical
===
operator over the equality==
operator for comparisons. - If you directly reference a third-party dependency in code, then ensure the dependency is required in the module's
composer.json
file. - Avoid hardcoding values when there is a method available to dynamically get a value, for example use
DataObjectSchema::tableName()
to get the table name for aDataObject
model rather than hard coding it. -
Do not use the
self
keyword, e.g.self::myMethod()
instead use the short-name of the current class e.g.MyClass::myMethod()
which is functionally equivalent. This avoids a specific class of bug that is introduced when using late static binding downstream fromself
.- The one exception to this rule is traits as there is no way to know the class the trait is applied to. In that case, it's fine to use for getting the class name (e.g.
self::class
) but should only be used for calling static methods (i.e.self::class::myMethod()
) if you can't refactor the code to avoid needing to useself
. - The use of
static
keyword is perfectly OK and is often appropriate.
- The one exception to this rule is traits as there is no way to know the class the trait is applied to. In that case, it's fine to use for getting the class name (e.g.