Why is access and visibility mixed up in PHP?

[Terje Slettebø]

Well, as is customary when you're new to a group, you tend to post quite a
bit, :) so here's one more. Some things I've been wondering about with PHP
(5).

Today, I worked on an implementation of a finite state machine. Unlike the
pear::fsm, this one supports hierarchical states, and I intend to make it
available, as well. It exists in both PHP 4 and 5 versions, but only the
PHP5 version is relevant for this example. The base class defines some
(virtual) functions that may be overridden in a derived class, but they are
only called from the base class. My original code was as follows (I only
quote the bare minimum needed for illustration):

class fsm
{
...
public function process($signal)
{
// null_action() is called from here (unless another action function is
specified for a transition)
}

private function null_action($from_state,$to_state)
{
}
}

class my_fsm extends fsm
{
private function null_action($from_state,$to_state)
{
echo "Transition from $from_state to $to_state";
}
}

In C++ this would work just fine: As null_action() is called from fsm, and
it's private in fsm, this works fine. It gets overridden in my_fsm, but
being private in fsm, it can only be called there (not in my_fsm), which is
as intended. This is because access and visibility are orthogonal concepts
in C++: The access specifiers only specify who are allowed to access (as in
calling, taking the address of, etc.) a function, but it doesn't affect
overriding.

The reason for this is as follows (from "The Design and Evolution of C++"):
By not letting the access specifiers affect visibility (including
overriding), changing the access specifiers of functions won't affect the
program semantics.

However, this is not so for PHP...

The above won't work, or at least not work as intended: The function
null_action() will only be visible in the class it's defined, and therefore
the derived class version won't override the base class version. In order to
get it to work, the access specifiers have to be changed to protected. This
means that derived classes may also _call_ the function, something that is
not desired. This means I can't enforce this design constraint of having
this function private.

Why is it done like this?

Regards,

Terje

[Tim Van Wassenhove]

["Followup-To:" header set to comp.lang.php.]
On 2005-01-22, Terje Slettebø <ts*******@hotmail.com> wrote:

[snip code]

The above won't work, or at least not work as intended: The function
null_action() will only be visible in the class it's defined, and therefore
the derived class version won't override the base class version. In order to
get it to work, the access specifiers have to be changed to protected. This
means that derived classes may also _call_ the function, something that is
not desired. This means I can't enforce this design constraint of having
this function private.
Actually, you can have the constraint. But it acts as expected in PHP
and not the way you would like it to act.
Why is it done like this?

I think you can better ask this on a php-internals ml.
My guess is that this is done because OOP in PHP5 seems to have been inspired by Java.
(Would also explain why we don't have friend visibility)
--
Met vriendelijke groeten,
Tim Van Wassenhove <http://www.timvw.info>

[Terje Slettebø]

"Tim Van Wassenhove" <ti***@users.sourceforge.net> wrote in message
news:35*************@individual.net...

["Followup-To:" header set to comp.lang.php.]
On 2005-01-22, Terje Slettebø <ts*******@hotmail.com> wrote:

The above won't work, or at least not work as intended: The function
null_action() will only be visible in the class it's defined, and therefore the derived class version won't override the base class version. In order to get it to work, the access specifiers have to be changed to protected. This means that derived classes may also _call_ the function, something that is not desired. This means I can't enforce this design constraint of having
this function private.
Actually, you can have the constraint.

No. The constraint was that it should be possible to override the function
in a derived class, but not to call it, there. Anyway, from my experience
with the FSM, it seems PHP behaves sufficiently different from C++ (and
Java, actually) in this context, that the issue is kind of moot. To
illustrate:

class base
{
public function g()
{
$this->f(); // Succeeds!
}
}

class derived extends base
{
protected function f() { echo "f()"; }
}

$object=new derived();

$object->g(); // Prints "f()"

IOW, it's possible to call a derived class function from the base class,
even if it's not defined in the base class (this is rather unlike Java, but
actually very useful for the FSM :) ), as long as the actual object being
used is of class derived.
But it acts as expected in PHP and not the way you would like it to act.

It seems that "what is expected" may be a tad subjective. :) For a C++
programmer, the way I described (being able to override, but not call, a
private virtual function) makes complete sense, at least to those that know
that access (being able to call) and inheritance/visibility (being able to
override) are orthogonal in C++.

For those interested, here's an article about it (C++):
http://www.gotw.ca/publications/mill18.htm However, the advice to prefer
public nonvirtual functions, delegating to private virtual functions, is not
something that is generally agreed upon (but also has nothing to do with my
example, as it doesn't do this).

However, I guess the Java/PHP way of doing it is reasonable, as well.

Regards,

Terje