Commits
-------
411a0cc [Validator] Added GroupSequenceProvider to changelog
815c769 [Validator] Renamed getValidationGroups to getGroupSequence
d84a2e4 [Validator] Updated test expectations
9f2310b [Validator] Fixed typos, renamed hasGroupSequenceProvider
e0d2828 [Validator] GroupSequenceProvider tests improved, configuration changed
c3b04a3 [Validator] Changed GroupSequenceProvider implementation
6c4455f [Validator] Added GroupSequenceProvider
Discussion
----------
[Validator] Added GroupSequenceProvider
Bug fix: no
Feature addition: yes
Backwards compatibility break: no
Symfony2 tests pass: ![](https://secure.travis-ci.org/blogsh/symfony.png?branch=dynamic_group_sequence)
As discussed in #3114 I implemented the "GroupSequenceProvider" pattern for the validator component. It allows the user to select certain validation groups based on the current state of an object. Here is an example:
/**
* @Assert\GroupSequenceProvider("UserGroupSequnceProvider")
*/
class User
{
/**
* @Assert\NotBlank(groups={"Premium"})
*/
public function getAddress();
public function hasPremiumSubscription();
}
class UserGroupSequenceProvider implements GroupSequenceProviderInterface
{
public function getValidationGroups($user)
{
if ($user->hasPremiumSubscription()) {
return array('User', 'Premium');
} else {
return array('User');
}
}
}
With this patch there are two mechanisms to define the group sequence now. Either you can use @GroupSequence to define a static order of validation groups or you can use @GroupSequenceProvider to create dynamic validation group arrays.
The ClassMetadata therefore has methods now which implement quite similar things. The question is whether it would make sense to interpret the static group sequence as a special case and create something like a DefaultGroupSequenceProvider or StaticGroupSequenceProvider which is assigned by default. This would cause a BC break inside the validator component.
---------------------------------------------------------------------------
by bschussek at 2012-01-28T13:39:54Z
I like the implementation, but I think we should differ a little bit from Java here.
1. `GroupSequenceProviderInterface` should be implemented by the domain classes themselves (`User`), not by a separate class.
2. As such, the parameter `$object` from `getValidationGroups($object)` can be removed
3. `ClassMetadata::setGroupSequenceProvider()` should accept a boolean to activate/deactivate this functionality. Also the check for the interface (does the underlying class implement it?) should be done here
Apart from that, special cases need to be treated:
* A definition of a group sequence and a group sequence provider in the same `ClassMetadata` should not be allowed. Either of them must not be set.
* Metadata loaders must take care of settings made by parent classes. If `Animal` is extended by `Dog`, `Animal` defines a group sequence (or group sequence provider) and `Dog` a group sequence provider (or group sequence), only the setting of `Dog` should apply
---------------------------------------------------------------------------
by blogsh at 2012-01-28T21:25:37Z
Changes of the latest commit:
- GroupSequenceProviderInterface has to be implemented by the domain class
- The annotation/configuration options let the user define whether the provider is activated or not (is this neccessary at all?)
- An error is thrown if the user wants to use static group sequences and the provider simultaneously
At the moment neither the static group sequence nor the provider is inherited from parent classes or interfaces. I don't know if it would make sense to enable this feature. There could be problems if a user wants to define a static group sequence in the parent class and a sequence provider in the child class.
---------------------------------------------------------------------------
by bschussek at 2012-01-30T13:07:04Z
> There could be problems if a user wants to define a static group sequence in the parent class and a sequence provider in the child class.
In this case, the setting in the child class should override the setting of the parent class.
But we can leave this open for now. As it seems, [this issue is unresolved in Hibernate as well](https://hibernate.onjira.com/browse/HV-467).
---------------------------------------------------------------------------
by blogsh at 2012-01-30T22:54:41Z
Okay, finally I managed to upload the latest commit. If you got a bunch of notifications or so I'm sorry, but I had to revert some accidental changes in the commit :(
I've rewritten the tests and have removed the "active" setting in the XML configuration.
---------------------------------------------------------------------------
by blogsh at 2012-02-02T15:24:01Z
Okay, typos are fixed now and `hasGroupSequenceProvider` has been renamed to `isGroupSequenceProvider`. I also had to adjust some tests after the rebase with master.
---------------------------------------------------------------------------
by bschussek at 2012-02-03T09:25:19Z
Looks good.
@fabpot 👍
---------------------------------------------------------------------------
by fabpot at 2012-02-03T09:46:52Z
Can you add a note in the CHANGELOG before I merge? Thanks.
---------------------------------------------------------------------------
by blogsh at 2012-02-09T12:31:27Z
@fabpot done
A new ExecutionContext is now created everytime that GraphWalker::walkConstraint() is
launched. Because of this, a validator B launched from within a validator A can't break
A anymore by changing the context.
Because we have a new ExecutionContext for every constraint validation, there is no point
in modifying its state anymore. Because of this it is now immutable.
Commits
-------
92f820a Renamed registerConstraints to loadDynamicValidatorMetadata
dd12ff8 CS fix, getConstraints renamed
09c1911 [Validator] Improved dynamic constraints
54cb6e4 [Validator] Added dynamic constraints
Discussion
----------
[Validator] Dynamic constraints
Bug fix: no
Feature addition: yes
Backwards compatibility break: no
Symfony2 tests pass: yes
By now the Validator component is based on a per-class configuration of
constraints, but in some cases it might be neccessary to add new constraints
dynamically at runtime.
This pull request adds a "ConstraintProviderInterface" to the Validator component. If an object is validated that implements this interface the method "getConstraints" is used to add dynamic constraints:
class User implements ConstraintProviderInterface
{
protected $isPremium;
protected $paymentInformation;
public function getConstraints(ClassMetadata $metadata)
{
if ($this->isPremium) {
$metadata->addPropertyConstraint('paymentInformation', new NotBlank());
}
}
}
---------------------------------------------------------------------------
by alexandresalome at 2012-01-15T11:20:04Z
Related to #1151
---------------------------------------------------------------------------
by canni at 2012-01-16T09:22:28Z
👍
---------------------------------------------------------------------------
by bschussek at 2012-01-16T12:32:44Z
I think this is a good addition. I think we still have a naming problem though. When constraints are loaded using a static method, the default name for the loader method is `loadValidatorMetadata`. Since the method for dynamic constraint loading is basically the same, I think the two names should be related.
Solution (1): Rename the method in your interface to `loadDynamicValidatorMetadata`. Ugly and long.
class MyClass implements ConstraintProviderInterface
{
public static loadValidatorMetadata(ClassMetadata $metadata) ...
public loadDynamicValidatorMetadata(ClassMetadata $metadata) ...
}
Solution (2): Rename the default method name in `StaticMethodLoader` to `registerConstraints` and adjust the docs. Breaks BC.
class MyClass implements ConstraintProviderInterface
{
public static registerConstraints(ClassMetadata $metadata) ...
public registerDynamicConstraints(ClassMetadata $metadata) ...
}
@fabpot: Are we allowed to break BC here? If not, we should probably stick to (1).
---------------------------------------------------------------------------
by fabpot at 2012-01-16T12:36:14Z
I would prefer to not break BC if possible.
---------------------------------------------------------------------------
by blogsh at 2012-01-16T15:25:46Z
So "loadDynamicValidatorMetadata" would be the best solution?
---------------------------------------------------------------------------
by althaus at 2012-01-17T13:39:19Z
>So "loadDynamicValidatorMetadata" would be the best solution?
Sounds fine for me based on @bschussek's comment.