Hello Mike, Thies,

i can only agree here, we just invented another inconsistency to violate
the beloved KISS approach once more for no particular reason. And this also
contradicts the arguments we use elsewhere.

marcus

Thursday, November 10, 2005, 1:35:56 AM, you wrote:

The problem does not cross class boundaries. However, here are some
test cases to better illustrate the question:

<?php
class Test {
function __get($offset) {
switch($offset) {
case 'hello1':
return 'there';
case 'hello2':
return $this->hello1;
case 'hello3':
return $this->__get('hello1');
case 'hello4':
return $this->testFunction();
}
}

 function testFunction() {
     return $this->hello1;
 }

}

$t = new Test();

// CASE 1: Code from outside class scope gets an undefined property
// which is proxied by __get(). This is the everyday usage.
// string(5) "there"
var_dump($t->hello1);

// CASE 2: Code inside __get() gets an undefined property,
// which generates a notice instead of being proxied by __get().
// NULL
var_dump($t->hello2);

// CASE 3: Code inside __get() calls __get()
// string(5) "there"
var_dump($t->hello3);

// CASE 4: Code inside __get gets an undefined property through a
// function, which generates a notice just like Case 2.
// NULL
var_dump($t->hello4);

// CASE 5: Code from inside a class function gets an undefined
// property which is proxied by __get()
// string(5) "there"
var_dump($t->testFunction());
?>>

I agree with Thies that this should be changed, on the basis that the
behavior is inconsistent and we don't protect against recursion
otherwise. Every test case should return the string.

The above examples were tested against 5.0.5 and the latest 5.1 with the
same results.

Mike

Andi Gutmans wrote:

Thies,

I think I might be wrong in my analysis. Seems like the problem is worse
than what I thought as it seems to cross boundaries between classes.
We'll look into it more.

Andi

At 02:15 PM 11/9/2005, Andi Gutmans wrote:

This is intentional and I think it makes it much more fool-proof for
developers. Adding recursion protection can be very expensive and
complicated, and it doesn't seem to make sense here. The idea is that
once you're in the __get() method, you know exactly how your
properties are mapped and can do the right thing. It's only this one
method where you have your mapping anyway, not an entirely different
place.

Andi

At 12:35 AM 11/4/2005, Thies C. Arntzen wrote:

hey,

in zend_object_handlers.c function zend_std_read_property we protect
against calling the __get function of an object if we're already in a
__get() function. (look for zobj->in_get) in that function.

i don't think we should be doing this.

simplyfied example:
<?php
class test {
function __get($offset) {
echo "__get($offset)\n";
switch ($offset) {
case 'name': return "name";
case 'length': return strlen($this->name);
}
}
}

$a = new test;
var_dump($a->length);
?>

outputs:
__get(length)
int(0)

but should say:
__get(length)
__get(name)
int(4)

i think recursive gets are highly useful and should be allowed. if we
really want to protect against recursion (here) we have two real
options: -a- protect against __get(ing) the same property more than
once in the same call-chain (mucho work), or -b- (preferred) have
some global infinite recursion protection in the engine (needs to be
done somewhen anyhow, but not now).

so, in one sentence: as we don't protect against recustion elsewhere,
can we please remove this unneded roadblock?

opinions?
-thies

--

--

--
Mike Naberezny
Senior PHP Developer

Zend Technologies, The PHP Company
19200 Stevens Creek Blvd, Suite 100
Cupertino, CA 95014
Tel: 408-342-8891
Fax: 408-253-8801
mike.n@zend.com

Best regards,
Marcus