The problem I see with your proposal is that get/isset/unset could all
bypass the setter which I think is a bad idea.
If the problem is simply this "hidden state" that keeps being referred
to, I thought the solution I mentioned before would be better.
To re-iterate that prior idea is to make it behave identically to __get().
The first call to a getter would initiate the (calling) guard, any other
access outside of the getter from that point forward would return NULL,
but the access by the getter would return the property.
It would stop the infinite loop, stop invalid access, the author could
see where they've gone wrong (a warning could be emitted even) and we
maintain the current solution without introducing more complexity (which
would probably lead to further problems).
For your pro's listed
* I don't think a local variable named $prop (for example) is very clear
that it's going to modify a property of $this, I would expect
$this->prop and that $prop would be modifying a local value.
* That's a misconception that I accidentally let perpetuate because I
wasn't careful when I confirmed your alternate verbiage.
And then it adds the evil's listed.
-Clint
On 1/5/2013 3:03 PM, Steve Clay wrote:
On 1/5/13 2:05 PM, Clint Priest wrote:
I like the alternate idea here, but I'm not sure what advantage it
has over the current
situation?
See the "Pros" I listed. The primary being a clear differentiation
between calling accessors and handling of the storage value associated
with the property.
if an accessor is
already "in the call chain" then it will not be called again and
"standard property rules"
take effect.
I see that as not simpler, and in fact leading to small bugs whenever
getters/setters start becoming complex enough to call outside methods.
Consider this:
class Foo {
public $bar = 2 {
get { return $this->calcBar(); }
}
public function calcBar() {
return $this->bar * 3;
}
}
echo $foo->bar; // 6
Within calcBar, "$this->bar" is the raw property (2), because hidden
state has "unwrapped" (if you will) the property.
echo $foo->calcBar(); // 18
But here, within the first call of calcBar "$this->bar" triggers the
getter. Now, of course, this is a foolish implementation, but within
any method that could be called from the getter/setter, the symbol
$this->bar could mean two completely different things; I think this is
bad.
If, as I proposed, the storage var were only accessible as $prop in
the accessor scopes, that would force authors to pass $prop to any
supporting methods, clarifying intent. $this->bar would *always* be
accessor calls.
In this model, I think infinite-loop-causing recursions would be
easier to spot. If absolutely necessary we could always throw a fatal
error whenever a getter was called twice in the same call chain.
Steve
AFAICT C# strictly separates fields ("properties" in PHP) and
properties (a set of
accessors that emulate a field).
So the RFC provides 3 features (using C# terms):
1. A property API
2. A built-in storage variable so you don't need a separate field
3. Access to the storage variable as if it were a field of the same
name
I think #2 is useful, avoiding the need to make a separate field
just to make properties
read-only or type-hinted. However I think the complexity and
confusion we're running
into is mostly caused by #3.
I think we might be better served by having another way to access
this storage variable.
What if instead, we have the storage var available as $prop inside
all the accessors?
These would be the default implementations:
get { return $prop; }
set($value) { $prop = $value; }
isset { return $prop !== NULL; }
unset { $prop = NULL; }
Pros:
* Makes clear that $prop is regular var access, and that
$this->PropertyName *always*
goes through accessors
* Gives isset/unset full access to the storage var, which allows
doing things that can't
be done via setter/getter. E.g. you could actually implement a
property being "unset",
which would be different from having it set to NULL.
Cons:
* Allows "evil", like having reads affect the storage var.
* Allows authors to hang themselves with recursive accessor calls,
BUT those mistakes
would be apparent from looking at the code.
What functionality possible in the RFC would be lost by this?
Steve Clay
Steve Clay
--
-Clint
