[RFC] Anonymous Classes

I think the proposal is a bit incomplete.
It's possible to instantiate an anonymous class, but currently it's not
possible to do with them anything else (assign to variable, pass to
function, etc). Something similar to Closure objects should be introduced.

Thanks. Dmitry.

On Tue, Feb 24, 2015 at 4:52 PM, Philip Sturgeon pjsturgeon@gmail.com
wrote:

Good day!

https://wiki.php.net/rfc/anonymous_classes

There's a little RFC + patch that Joe Watkins put together, and as
before with the ArrayOf RFC, I'll be helping out.

So, lets get this discussion rolling.

It was declined for PHP 5 some time ago, and has returned now to try for
PHP 7.

The usage of anonymous classes to some will be instantly offensive,
but really it comes down to the use case. The usage of anonymous
functions compared to declared functions is pretty much the exact same
thing as anonymous classes.

Other than examples on the RFC, Fractal would certainly be happy to
have them: http://fractal.thephpleague.com/transformers/

Defining a class properly is certainly still going to be the majority
of uses of classes in PHP. That helps with the Optimizer, and helps
code reuse.

Sadly due to the way in which people have had ONE CLASS ONE FILE
drilled into their head since PEAR and continuing through Zend and
PSR-0, it can become a PITA to add some simple functionality if a
small class is needed for one tiny thing.

Anonymous functions alleviate that annoyance with a simple and
consistent feature that just give people a nice simple option to get
their jobs done, without hitting autoloaders and file systems to do
it.

Hi Phil,

Am 24.02.2015 um 14:52 schrieb Philip Sturgeon:

https://wiki.php.net/rfc/anonymous_classes

I like the idea of having anonymous classes, it is very helpful during
development to just try something out without having the burden of
creating a new file and a complete class including namespace and use
declarations, etc.

A great feature of anonymous classes in Java is their ability to access
the private/protected properties of the object they are defined in,
similar to what Closures do in PHP. The thing is, in Java if you access
a variable (without this.), it may be of the current scope, a member of
the current class or a member of the class where the anonymous class was
defined in. In PHP, you have to use $this-> to access class members of
the anonymous class but there is some $outer-> or similar variable
missing for the access. You can as a work around pass all members by
reference via the constructor, but that is really not nice.

In addition, if you define an anonymous class inside a "real" class and
pass another instance of the outer class, the anonymous class can still
access the private/protected methods imho. Any thoughts how to handle
this or should we just ignore it for now?

Second thing is serialization. PHP closures can not be serialized, does
the same apply for anonymous classes? It would be really nice if such
anonymous classes could e.g. send via a command bus or something to be
executed on another host.

Thanks
Dennis

P.s.: there is a footnote in the RFC that does not belong there is think

Good day!

https://wiki.php.net/rfc/anonymous_classes

There's a little RFC + patch that Joe Watkins put together, and as
before with the ArrayOf RFC, I'll be helping out.

So, lets get this discussion rolling.

Anonymous classes open up slightly new ways to do things. I like to
write a lot of small classes for the purposes of DI and re-usability.
Because of that, the code often looks like:

$foo = new Foo($depend1, $depend2, function() {
return 'my custom behavior';
});

That's fine when there's only one callback needed, but beyond that it
breaks down. This becomes more readable:

abstract class Foo
{
...
}

$foo = new class($depend1, $depend2) extends Foo {
public function doFoo() {
}
public function doBar() {
}
}

These cases for me are always single-use for the scope of the
application, and often it's inside configuration blocks that are
custom per environment. I have no need to name the class other than
PHP doesn't support anonymous ones.

I'm +1 on the concept, depending on what the maintenance impact that
this would bring to PHP's core. There are enough other ways to
accomplish the same goals that I don't think this is "must have", even
though I'd definitely use it.

Regarding this syntax (from other discussions on the list):

$foo = class { };

I think if such a thing were supported, maybe $foo would just be a
ReflectionClass object. (I haven't really thought that one through...)

--
Matthew Leverton

here is an example for opening a mysqli connection only when the first query is executed:

$c = new class extends mysqli {
public function query($query, $resultmode = MYSQLI_STORE_RESULT) {
if (empty($this->host_info)) parent::real_connect('127.0.0.1', 'user', 'pwd', 'db');
return parent::query($query, $resultmode);
}
}

// do sth for some time

$->query('...'); // triggers real_connect
$->query('...'); // connection is already established

I haven't tested this, so I'm only intending to present the idea.

Regards
Thomas

Philip Sturgeon wrote on 24.02.2015 14:52:

Good day!

https://wiki.php.net/rfc/anonymous_classes

There's a little RFC + patch that Joe Watkins put together, and as
before with the ArrayOf RFC, I'll be helping out.

So, lets get this discussion rolling.

It was declined for PHP 5 some time ago, and has returned now to try for PHP 7.

The usage of anonymous classes to some will be instantly offensive,
but really it comes down to the use case. The usage of anonymous
functions compared to declared functions is pretty much the exact same
thing as anonymous classes.

Other than examples on the RFC, Fractal would certainly be happy to
have them: http://fractal.thephpleague.com/transformers/

Defining a class properly is certainly still going to be the majority
of uses of classes in PHP. That helps with the Optimizer, and helps
code reuse.

Sadly due to the way in which people have had ONE CLASS ONE FILE
drilled into their head since PEAR and continuing through Zend and
PSR-0, it can become a PITA to add some simple functionality if a
small class is needed for one tiny thing.

Anonymous functions alleviate that annoyance with a simple and
consistent feature that just give people a nice simple option to get
their jobs done, without hitting autoloaders and file systems to do
it.

On Tue, Feb 24, 2015 at 7:52 AM, Philip Sturgeon pjsturgeon@gmail.com
wrote:

Good day!

https://wiki.php.net/rfc/anonymous_classes

There's a little RFC + patch that Joe Watkins put together, and as
before with the ArrayOf RFC, I'll be helping out.

So, lets get this discussion rolling.

It was declined for PHP 5 some time ago, and has returned now to try for
PHP 7.

The usage of anonymous classes to some will be instantly offensive,
but really it comes down to the use case. The usage of anonymous
functions compared to declared functions is pretty much the exact same
thing as anonymous classes.

Other than examples on the RFC, Fractal would certainly be happy to
have them: http://fractal.thephpleague.com/transformers/

I would also find myself to be a very happy consumer of this. It can
provide the opportunity of getting rid of a large amount of boilerplate.
In addition, there has been several times when implementing single use
extensions for say view libraries (Plates extensions comes to mind) or
Plugin handling for things like Zend Framework 2 and there has been several
times I utilize a closure with an array to get around having an object
simply due to you want to have a callback but with some form of
definition. I think there is a very large use case here and something that
would be a great addition to the language. Sure, I could make a class and
extend it just like this would do but for the convenience of the
implementation it is sometimes better to do a simple closure.

Defining a class properly is certainly still going to be the majority

of uses of classes in PHP. That helps with the Optimizer, and helps
code reuse.

Sadly due to the way in which people have had ONE CLASS ONE FILE
drilled into their head since PEAR and continuing through Zend and
PSR-0, it can become a PITA to add some simple functionality if a
small class is needed for one tiny thing.

I see the point here, but the argument to me seems void. Nothing wrong
with a simple class one file for a small class.

Anonymous functions alleviate that annoyance with a simple and
consistent feature that just give people a nice simple option to get
their jobs done, without hitting autoloaders and file systems to do
it.

Hi!

https://wiki.php.net/rfc/anonymous_classes

There's a little RFC + patch that Joe Watkins put together, and as
before with the ArrayOf RFC, I'll be helping out.

So, lets get this discussion rolling.

This is a nice and clear proposal, however I'm not sure I'm completely
convinced about the necessity of it. In Java this kind of pattern is
used very frequently, because Java is a statically typed language where
it is the only way to achieve dynamic interface, and has other
capabilities, see below. In PHP, there are other ways (and can be made
more ways, such as more support for duck typing).

Also, the question arises when comparing this to Java - would anonymous
class functions be closures connected to the environment? In Java, they
are and that is what they are most frequently used for - to make
contextualized callbacks. In PHP, you need special syntax to capture the
environment, so how that would work? Also, what about scope access, etc.?

Sadly due to the way in which people have had ONE CLASS ONE FILE
drilled into their head since PEAR and continuing through Zend and
PSR-0, it can become a PITA to add some simple functionality if a
small class is needed for one tiny thing.

One class/one file may be a good idea in general for a big project, but
as they say, any feature that can not be turned off when needed is a bug.

Stas Malyshev
smalyshev@gmail.com

Hi Philip

-----Ursprüngliche Nachricht-----
Von: Philip Sturgeon [mailto:pjsturgeon@gmail.com]
Gesendet: Dienstag, 24. Februar 2015 14:52
An: PHP Internals
Betreff: [PHP-DEV] [RFC] Anonymous Classes

Good day!

https://wiki.php.net/rfc/anonymous_classes

There's a little RFC + patch that Joe Watkins put together, and as before with the ArrayOf RFC, I'll be helping out.

[Robert Stoll]
I like the RFC +1 :)

Can I also use traits in combination with anonymous classes?=3D20

Hi Phil.

I really like this feature and am keen to see it go in. Developers

often create classes (which are typically new files) just to extend and
implement a small method. This feature could see codebases become much
lighter. So a massive thanks for your and Joe’s efforts on this.

I have two quite small points I want to make.

Firstly,

When a error occurs on an anonymous class the class description isn’t very
useful (see http://3v4l.org/9Nbea/rfc#tabs). Instead of
“class@0x7f9e33870fd0” Is it possible it could be prefixed with “anonymous”
or something a bit clearer. Further to this if the anonymous class extends
anything it omits the parent class name in the error message (see
http://3v4l.org/UFEbo/rfc#tabs). Is this expected? Could it be something
more helpful?

Secondly, I think we need to either allow the behaviour or generate an
error (similar to anonymous closures) when an anonymous class is
serialized. Please observe the following behaviour..

$foo = new class {
public function Bar () {
echo 'bar';
}
};

// This succeeds. I'd argue an error should be generated at this point
echo $serFoo = serialize($foo);

// Unserialize fails and a notice is generated
$unFoo = unserialize($serFoo);

// Error generated on call
$unFoo->Bar();

see http://3v4l.org/AJY6d/rfc#tabs

That’s all. Again, thanks for your efforts.

/@leedavis81

On Tue, Feb 24, 2015 at 1:52 PM, Philip Sturgeon pjsturgeon@gmail.com
wrote:

Good day!

https://wiki.php.net/rfc/anonymous_classes

There's a little RFC + patch that Joe Watkins put together, and as
before with the ArrayOf RFC, I'll be helping out.

So, lets get this discussion rolling.

It was declined for PHP 5 some time ago, and has returned now to try for
PHP 7.

The usage of anonymous classes to some will be instantly offensive,
but really it comes down to the use case. The usage of anonymous
functions compared to declared functions is pretty much the exact same
thing as anonymous classes.

Other than examples on the RFC, Fractal would certainly be happy to
have them: http://fractal.thephpleague.com/transformers/

Defining a class properly is certainly still going to be the majority
of uses of classes in PHP. That helps with the Optimizer, and helps
code reuse.

Sadly due to the way in which people have had ONE CLASS ONE FILE
drilled into their head since PEAR and continuing through Zend and
PSR-0, it can become a PITA to add some simple functionality if a
small class is needed for one tiny thing.

Anonymous functions alleviate that annoyance with a simple and
consistent feature that just give people a nice simple option to get
their jobs done, without hitting autoloaders and file systems to do
it.

On Tue, Feb 24, 2015 at 5:19 PM, Philip Sturgeon pjsturgeon@gmail.com
wrote:

I think the proposal is a bit incomplete.
It's possible to instantiate an anonymous class, but currently it's not
possible to do with them anything else (assign to variable, pass to
function, etc). Something similar to Closure objects should be
introduced.

Thanks. Dmitry.

1. You can absolutely assign the instantiated classes to variables.

Check out this test in the patch:

https://github.com/krakjoe/php-src/compare/anon#diff-25e330fb5a98810de178a5b798102d01R1

In tests you assign instantiated objects.

$a = new class {...};
$b = new class {...};

I'm talking about classes as first class objects.

$c = class {...};
$a = new $c;
$b = new $c;

2. Why do you say they cannot be passed to a function? I can add a
   test if you can give me an example of what you're suggesting doesn't
   work.

3. Not sure why we'd need a Closure-alike object. Anonymous classes
   are just a class, and classes have all the types and hinting
   functionality of regular classes. You don't need to implement a class
   to let people know its a class.

Maybe you could expand on that a bit? :)

I tried it in the example above.

Also, classes may be useful without objects at all (just static properties
and methods).

$c = class {...};
$c::static_foo();

Thanks. Dmitry.

Thank you for the example! Your example explains my question 1, but
doesn't seem to explain 2 or 3.

I'll play around with static usage and show it off in the tests.

Am 24.02.2015 um 14:52 schrieb Philip Sturgeon:

https://wiki.php.net/rfc/anonymous_classes

I like the idea of having anonymous classes, it is very helpful during
development to just try something out without having the burden of
creating a new file and a complete class including namespace and use
declarations, etc.

I don't buy this argument. If your class loader or PSR prevents you from
temporarily adding a 2nd bogus class to the same file (not talking about
best practices here, obviously) then there's the culprit and it's
nothing the language needs to fix.

Your reply seems to assume that avoiding rules in a PSR is the only
benefit of the RFC, which is weird when there is a whole RFC full of
benefits of this RFC.

My reply was simply re: the quoted paragraph in the mail I directly
replied to, not yours. I simply think "trying something out" is not a
valid use case of anonymous classes.

I can generally like the RFC and still refute arguments in favor of it,
right? :)

~Florian