Validation rules issues when using model's attribute as additional parameters

[rsn86]

As CModel._validators = CModel.createValidators() is called before any model's attribute be setted, the validation rules are compiled with attributes's default values.
This could lead to validation issues. To avoid that, CModel.validate() should always compile validation rules on runtime.

This is best observable when using CUniqueValidato, e.g.:

class MyModel extends CFormModel {

    public $someAttr;
    public $uniqueAttr;

    public function rules() {
        return array(
            array('uniqueAttr', 'unique', 'className'=>'AnotherClass', 'allowEmpty'=>false, 'skipOnError'=>true, 'criteria'=>array('condition'=>"notNullAttr=:someAttr",'params'=>array(':someAttr'=>$this->someAttr))),
        );
    }

}

In this example, CModel._validators will contain a CUniqueValidator validation rule
with :someAttr setted as NULL. CModel.validate() will always return TRUE regardless of MyModel->uniqueAttr 's value, since there isn't any AnotherClass->notNullAttr == NULL.

[resurtm]

Not an issue. This is by design. Treat rules() as a some sort of static method, it mustn't know anything about particular model instance. For your case consider making a new model method based validator.

[rsn86]

If this is the understanding about it at least it should be well documented.

Using protected attributes instead of private, or setter/getters for all vars, would allow a more flexible usage of the framework.

A resetValidators() { $this->_validators = null; } method would do the trick too.

All that need to be done is set CModel->_validators = null before validation.

[resurtm]

If this is the understanding about it at least it should be well documented.

Maybe it worth documenting it.

The rules() method is generally declarative thing and it should stick with KISS. If you want to somehow deal with the particular validated object you have to create your own class based or method based validator.

All that need to be done is set CModel->_validators = null before validation.

The problem is that if we'll use this way lazy validators initiation would become meaningless.

[rsn86]

I guess a better solution would be to use evaluateExpression() on validator's parameters

[cebe]

related: http://stackoverflow.com/a/9692637/1106908