Skip to content


DDC-115: Two ClassMetadata instances for an entity in global namespace #1750

doctrinebot opened this Issue · 7 comments

2 participants


Jira issue originally created by user @beberlei:

The global namespace creates problems for entities that are retrieved via getClassMetadata($className) when $className is not prepended by "\". This is only a problem in the global namespace, because in other instances a slash might be present and the metadata factory does not find the relevant previously generated ClassMetadata instance.

This problem specifically occours with Annotations Mapping, since ReflectionClass::getName() returns the class name unqualified.

php -r 'namespace Foo { $r = new \ReflectionClass("\stdClass"); echo $r->getName();} '

I attached a Test-Case that proves this behaviour. It requires the additional attached Global namespace model data.


Comment created by @beberlei:

[20:20:55] <beberlei> guilhermeblanco: the problem is that ReflectionClass->getName() returns the name without prepended namespace operator for global classes
[20:27:43] <guilhermeblanco> beberlei: so it's a bug in ReflectionClass
[20:27:51] <guilhermeblanco> can't you just do get_class($instance)?
[20:28:11] <beberlei> no its the annotations mapping
[20:28:22] <beberlei> its using ReflectionClass::getName()
[20:29:13] <guilhermeblanco> hm...
[20:29:22] <guilhermeblanco> give me a second... I wanna test something
[20:31:33] <beberlei> i all attach test and classes to a bug
[20:32:43] <guilhermeblanco> weird
[20:32:54] <guilhermeblanco> beberlei: let me show you something
[20:33:37] <beberlei>[DDC-115](
[20:33:50] <beberlei> the problem is ReflectionClass->getName() has to be backwards compatible
[20:34:19] <guilhermeblanco> beberlei:
[20:34:54] <beberlei> oh
[20:34:57] <beberlei> they all strip it?
[20:35:04] <beberlei> we need a normalize method
[20:35:41] <guilhermeblanco> let me go further
[20:35:47] <guilhermeblanco> I wanna check if it's not a PHP bug
[20:37:53] <guilhermeblanco> beberlei:
[20:48:22] <beberlei> its not a bug yeah, but then we need to extend ClassMetadataFactory to prepend each $className which has none with the namespace separator
[20:48:34] <beberlei> hasMetadata getMetadata setMetadata

Comment created by @beberlei:

Committing a patch that adds the following snippet:

if($className[0] == "\\") {
    $className = substr($className, 1);

to each of the following methods of ClassMetadataFactory:

setMetadataFor($className, $class)

Comment created by @beberlei:

Applied to trunk, Can someone verify this is ok?


Issue was closed with resolution "Fixed"


Comment created by romanb:

Not sure I understand this. As I said in another issue class names in strings are always considered fully qualified by php. A leading backslash in class names in strings should never be used it is completely unnecessary.

The patch is problematic because the methods they're applied to are one of the most-called methods and they were already "slow enough" previously.

If there is a problem and the problem is only with models in the global namespace then this should not be fixed. User-defined classes dont belong in the global namespace, ever.


Comment created by @beberlei:

We found the true issue.

Several methods in ClassMetadataInfo assumed when no slash was prepended that a shortcut is used and prepends its own namespace plus a NS. This caused the problem which is now fixed by additionally checking for strlen($this->namespace) on this positions.

@beberlei beberlei was assigned by doctrinebot
@doctrinebot doctrinebot closed this
@doctrinebot doctrinebot added the Bug label
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.