Le 08/07/2026 à 14:07, Michael Morris a écrit :
On Wed, Jul 8, 2026 at 1:08 AM Holly Schilling
<[email protected]> wrote:
Following the discussion in Discord yesterday, I started working
on the feasibility of implementing class extensions. I believe
I’ve proven out the concept and viability of it in a 2-phase
approach using an extension syntax similar to that of Swift.
Before I go into much more detail, what is the general interest in
adding this to the language?
We have inheritance (extends keyword, parent / child classes)
We have traits (an implementation of multiple classes taken from Ruby)
Looking at Swift's extension syntax I fail to see anything it adds not
covered by the above.
I think that Swift's extension system goes the other way round: the
class itself doesn't declare its extensions, but they are declared from
userland.
It's similar to how Rust's traits system can allow extending other types
(though there's kind of a safeguard to avoid extending nonsense) , you
can check it there: https://doc.rust-lang.org/book/ch10-02-traits.html
PHP with inheritance goes this way:
- Declare Class A
- Class A internally defines which class it inherits from
- Class A internally defines which traits it uses
- Everything is now defined for Class A and its structure is finite as
of the end of its declaration
With an "extension" system, it might go this way:
- Declare Class A
- Class A internally defines which class it inherits from
- Class A internally defines which traits it uses
- Class A is defined, but extensions can update it somehow.
- An extension is later declared for Class A: it can customize the inner
structure of Class A, like new traits or implemented interfaces
- Maybe (depending on how the RFC goes) other extensions can be declared
later, in order to customize other things.
Conceptually, I'm not against it, but just like Michael said: I can't
find use-cases that would be better than what we actually have.
Extending a class "from the outside", whatever the way to do it, implies
that a new safeguard has to be added. We could use the "final" keyword,
but "final" is not at all the same thing as "no extension allowed": a
class could be declared as not-final while still disallowing
extensibility from the outside. That's the case when you have a
non-final class but some of its methods are "final". Disallowing
extensibility would require an entirely new keyword for that, and it
adds another burden to code maintainers, especially if extensibility is
opt-out.
So, IMO, such RFC should contain this:
- Thorough examples of the actual problems it solves for package
maintainers and application developers
- Clear views on how extensibility impacts the current ecosystem (I
guess it shouldn't have any impact on Composer and the autoloading
system, because if it's similar to Swift, an extension should just
behave like a class in terms of autoload, but I might have not thought
about it completely)
- Extensibility must be either opt-in by design, or opt-out by design,
but the decision must be made **after** having analysed the impact on
the ecosystem. I can't imagine extensibility being opt-out and people
starting to wildly override frameworks core classes. (yeah, I'm talking
about some Laravel nerds that tend to bend and break fast (pun intended))
- Extension points must be defined on both the extender and the extended
structures. Like, for example, making sure it's impossible to extend
something that's "final" by default, but making a non-final structure
being non-extensible should also be possible. That's part of the concept
of encapsulation in general, but it also respects the Open/Close
principle (from SOLID) and several other programming practices that make
sure we don't do bullcrap in all codebases. And trust me, PHP is already
full of bullcrap, so adding more isn't a good idea at all. PHP hasn't
become "more and more strict" over time for nothing... and still, that
strictness is opt-in (by the eyes of PHP, not from framework
maintainers, which can be seriously opinionated).
Hope this helps :)
