[BUG] Metadata PhpDriver always beats AnnotationDriver in DriverChain

[akomm]

Problem:

Given classes C1 and C2, while C1 extends C2:
When I deserialize (fromArray) C1 I get an error "can not redeclare class C2".

Found cause:

The DriverChain (jms_serializer.metadata.chain_driver) has following drivers by default (order is important):

• YamlDriver
• XmlDriver
• PhpDriver
• AnnotationDriver

Both PhpDriver and AnnotationDriver accept an annotated .php file. But because PhpDriver is first in order, AnnotationDriver never comes into play. The PhpDriver requires the config resource and that triggers the explained error.

Overriding the service in the following way fixes the error:

  jms_serializer.metadata.chain_driver:
    class: "%jms_serializer.metadata.chain_driver.class%"
    arguments:
      -
        - '@jms_serializer.metadata.yaml_driver'
        - '@jms_serializer.metadata.xml_driver'
        - '@jms_serializer.metadata.annotation_driver' # moved up
        - '@jms_serializer.metadata.php_driver'

Not sure who really is responsible for the issue, the serializer bundle or library. Possibilities:

1. jms/serializer: The missing acceptance check API in Metadata\Driver\DriverInterface and thus missing checks using the API. Currently the use-cases I found always implicate that null returned from loadMetadataForClass call means "not accepted", but its not phpdoc-annotated and explained in DriverInterface, so I'm not sure.

2. jms/serializer: The bold require in PhpDriver without checking acceptance and returning null, before requiring. I see the problem not requiring here thought...

3. jms/serializer-bundle: Actually the root problem seems to be one of the above. When null return in 1. is meant to be "not accepted", then 2. is the problem. But I see some complexity when testing acceptance in case 2. before "require". So maybe rearranging the driver chain is in this package is the smarter solution.

[goetas]

Are you using annotations or php metadata files?

[akomm]

Annotation.

[goetas]

How are configured your paths? ( they should be not configured at all; and the php driver should never be invoked )

[akomm]

  metadata:
    auto_detection: false
    directories:
      AppMutationInput:
        namespace_prefix: "AppBundle\\GraphQL\\Resolver\\Mutation\\Input"
        path: "%kernel.root_dir%/../src/AppBundle/GraphQL/Resolver/Mutation/Input"

I use it to hydrate (fromArray) data transfer objects from graphql mutation queries. With default config it only checks AppBundle\Entities and throws, when I try to hydrate something else.

[goetas]

if you use annotations, you do not need to define any metadata path (annotations are extracted via reflection) from the classes.

if you are you using a mixed approach php + annotation, make sure to not have namespace collisions... (that is probably what activates the php driver...)

[akomm]

I started with zero configuration and got this exception. That is when I started check configuration and added the directory configuration for the desired namespace. Now I removed the configuration again and no exception. But something must have changed, because there is no magic.

[goetas]

replaceMetadataDir is being called. can you post your configs somewhere?

[akomm]

Its almost default. Added dunglas_action and overblog_graphql plus security config. None of them have even close something to do jms serializer. And none of them were changed a single time since I added the jms serializer bundle.

As you said it works with ZERO config - now. NO magic. I must have done something bloody stupid and did not notice it head => wall. If you want get sure that there is indeed no edge-case scenario which caused the method's invocation I can provide you the config, but I think it was just me confused by only devil knows what (I still can not reproduce the case).

[goetas]

it was just me confused by only devil knows what

👍