rfc:static_class

PHP RFC: Static class

Introduction

A static class is a class that cannot be instantiated, and whose members (properties and methods) are all static. Implicit static classes can be created today, by simply only including static members. However, explicitly marking a class as static has several advantages:

  • Clear expression of developer intent.
  • Language-level compile-time checks to verify no instance members are included anywhere in the inheritance chain.
  • Language-level runtime checks to verify the class is never instantiated by any mechanism.

Of these advantages, the first can be somewhat mitigated with a userland trait, but as such traits are non-standard, there is no consistent way to identify them between projects. The static keyword is a consistent way to identify static classes that could also be leveraged by static analysis and code-completion tooling. Equally importantly, the language-level checks cannot be replicated in userland, thus are better suited for inclusion in PHP. In particular, classes with a private constructor can still be instantiated via reflection and faux-deserialization hacks.

Proposal

A static class is so marked with the existing static keyword, adjacent to the class declaration. 

static class Foo {}

We introduce the static keyword at the class level to preclude the need to create a private constructor. That is, function __construct() is a compile-time error in a static class. Furthermore, attempting to instantiate a static by any means, whether with the new keyword, ReflectionClass::newInstance* methods or unserialize() hacks, is strictly forbidden and will result in a runtime error.

Whilst the goals of this RFC so expressed are fairly straightforward, we acknowledge there are many finer points to consider. In particular:

  1. Should a static class imply final (as in C#)?
  2. Should a static class prohibit inheritance (as in C#)?
  3. Should a static class permit state?
  4. Should traits be permitted in a static class?
  5. Can a static class be marked abstract?
  6. Which magic methods should be supported?

To answer these questions, we consider the principal reason to mark a class static is to move from an implicit declaration to the explicit. Ergo, any decision precluding the developer from doing so is invalid, since it coerces them to avoid explicit static, defeating its purpose. Thus we derive the following razor:

We cannot remove features from a static class that would otherwise be present in a standard class.

Following this razor we arrive at the following design.

  1. A static class may be marked final, but in the absence of such a mark, may still be extended.
  2. A static class may extend other static classes, provided the parents are also marked static.
  3. A static class may have static properties.
  4. A static class may use traits, provided they introduce no instance members.
  5. A static class may be marked abstract, since abstract static functions already exist in PHP.
  6. All static magic methods must be supported, i.e. just __callStatic().

Static keyword for class members

One question our razor cannot answer is whether static member declarations still require the static keyword in a static class.

Marking a class readonly makes marking properties with the same optional, thus one might argue static should follow suit. However, I opine it should not. Unlike readonly, which typically applies to properties declared near the head of the class (by convention) where the same modifier is also applied, static members appear throughout the full length of the class body. That is, it should be apparent when jumping anywhere into a static class that its members are static, without having to check the class declaration. Therefore this RFC still requires the static modifier for all members regardless of whether the class is also marked static, to prevent them appearing as instance members. Aside, this is commensurate with the C# approach.

Mutually exclusive modifiers

When static is applied to a class, the following other class-level modifiers become invalid and will raise compile-time errors.

  • readonly — Read-only properties are only supported for instance properties, but instance properties are forbidden by static.

Inheritance

  • A static class only permits extending a parent class that is also marked static.
  • A non-static class cannot extend a static class.
  • A static class can inherit from an abstract class, provided it is also marked static.

Anonymous classes

Anonymous classes are always instantiated and thus at odds with the concept of a static class. That is, the notion of an anonymous static class is inherently invalid and not supported.

Dynamic properties

Undeclared object properties can be added dynamically to object instances, but since static classes cannot be instantiated, there is nothing to attach such properties to. Static properties must always be declared explicitly.

Reflection extensions

A new ReflectionClass::isStatic method will be added, returning true when the class is so marked, otherwise false.

Backward Incompatible Changes

None known.

Future Scope

These are some possible future extensions, but we don't necessarily endorse them.

  1. As noted in Readonly properties 2.0, read-only static properties are not supported due to a technical limitation. If that limitation should ever be lifted, we could revisit lifting the restriction on mutual exclusivity with the readonly modifier.
  2. Static interfaces may be introduced later if there is such a demand.
  3. Static traits may be introduced later if there is such a demand.

Vote

As per the voting RFC a yes/no vote with a 2/3 majority is needed for this proposal to be accepted.

Voting started on 2024-07-15 and ended on 2024-08-09 at 21:00 UTC.

Implement static classes as described?
Real name Yes No
alcaeus (alcaeus)  
asgrim (asgrim)  
ashnazg (ashnazg)  
bwoebi (bwoebi)  
crell (crell)  
derick (derick)  
devnexen (devnexen)  
dharman (dharman)  
ericmann (ericmann)  
galvao (galvao)  
girgias (girgias)  
jhdxr (jhdxr)  
kalle (kalle)  
kguest (kguest)  
levim (levim)  
mauricio (mauricio)  
mbeccati (mbeccati)  
nicolasgrekas (nicolasgrekas)  
ocramius (ocramius)  
patrickallaert (patrickallaert)  
pierrick (pierrick)  
ramsey (ramsey)  
rasmus (rasmus)  
reywob (reywob)  
sebastian (sebastian)  
seld (seld)  
sergey (sergey)  
thekid (thekid)  
theodorejb (theodorejb)  
trowski (trowski)  
weierophinney (weierophinney)  
Final result: 11 20
This poll has been closed.

Discussion

Though this is a fairly straightforward RFC, it is not without its detractors. Some view static classes as an anti-pattern; a namespace cheat that should instead be presented as namespaced functions (sans-class wrapper). That goes double for static classes including state, where static properties can be viewed as equivalent to global state, which is widely regarded as an anti-pattern. However, this RFC is not encouraging any particular patterns. On the contrary, we are merely returning a small but nevertheless useful tool to the developer's toolkit that could have been available since classes were introduced. In this humble author's opinion, static classes should just be a collection of pure functions, but as defined by our opening razor, we will not remove features from a static class that exist in a non-static class because we do not have the liberty of designing a new language, we're designing PHP, with all the weight of its past carried forward. Anyone wishing to remove features from PHP can submit a separate RFC, or perhaps more practically, just add a check to their favourite code style tool.

Some regard namespaced functions as the correct way to implement static classes. That is, a file of floating functions under a namespace, as in Amp. This author requested comments that would speak to any technical or philosophical reason for why this would be strictly better than a static class, but the only technical argument fielded was that classes can be autoloaded and functions cannot. Curiously, this fact only speaks in favour of static classes. One might argue this difference is negligible thanks to Composer; files of functions can be loaded by Composer, but in this case the file is always loaded, not autoloaded. Still, this could be splitting hairs since opcache presumably trivializes this difference. Interestingly, the current Amp maintainers commented that namespaced functions were something they inherited, and if they had to do it all over again, would probably elect for static classes. Thus we conclude those preferring one style over the other do so purely out of personal preference and not because one is technically nor even philosophically superior.

Most importantly, this proposal does nothing to promote any patterns or practices not already possible, nor does it block or inhibit development of orthogonal approaches, such as autoloading of namespaced functions, which some may regard as equivalent.

In the wild

Today, implicit static classes are all around us, in proprietary projects and some of the largest open source projects in the world, often called utils or helpers, all of these classes would benefit from being explicitly marked static. To name a few:

  • PHPUnit
    • Almost everything under Util including:
    • Interestingly, Assert would not be a static class because, despite being filled entirely with static methods, is intended to be extended by test class instances. In a (future) static class world, the absence of the static modifier would more clearly communicate this intent.
  • Symfony
  • Laravel – half this framework appears to be static, though fewer classes qualify as pure static, including:

References

Rejected Features

Forbidding static class type declarations

It should be regarded as an error to use a static class as a type declaration, since they cannot be instantiated and thus the requirement can never be fulfilled by a matching instance. However, it is not technically possible to forbid such declarations within PHP itself because type checking is done when an instance is passed at runtime (and we can never have such an instance). The engine does not support type checking at compile time for function/method signatures. Such a check would be a good candidate for third party static analysers instead.

Special thanks

I would feel remiss not to give thanks to the following list contributors whose high quality feedback made significant contributions to this RFC.

  • Mike Schinkel
  • Alexandru Pătrănescu
  • Claude Pache
  • Everyone else who engaged with this proposal.

Thank you!

rfc/static_class.txt · Last modified: 2024/08/10 13:30 by bilge

Page Tools



Table of Contents