Allow multiple nested serializer calls

[xavierlacot]

In case of serializer recursion, the serializer send an exception: "Can't pop from an empty datastructure", as noted by @damienalexandre in #319

This PR allows to stack the navigator association to visitors, thus making it possible to nest serializer calls. It contains two commits:

• the first one contains the required changes for the bug to be fixed;
• the second one adds a test which shows up the bug resolution (tested against the fixed testsuite proposed in Fix test suite on travisci #292)

[ogizanagi]

Any news about this PR ?

[schmittjoh]

I'm not sure whether we need to support this use-case, i.e. nested calls. I think it might be better to create another serializer instance if you need this instead of using the same instance.

[ogizanagi]

Thanks for the quick answer 😃

This could be out of the scope of the current debate, but may I ask you the proper way to do that, as I need to register my custom handlers too, in the context of a Symfony 2 application ?
This basically means redoing the same work as in the jms/serializer-bundle, right ?

However, for my use case, using the SerializerBuilder directly, and registering the handlers I need seems sufficient.

[damienalexandre]

I find it disturbing that we need a fresh new Serializer to perform nested calls. A new SerializationContext should be enough. I use Serializer as a service most of the time, where one instance is supposed to fit all serialization needs.

Isn't it like saying Serializer is not stateless?

[digitalkaoz]

yeah ran into the same issue here...sadly the serializer is somewhat stateful and thats a bad design here in my eyes

[digitalkaoz]

@schmittjoh why are you blocking this valid use case?

[digitalkaoz]

if you have a more complex serializer construction (custom handlers, event subscribers, object constructor etcpp) this is a pain in the ass.

[bassrock]

👍

[bassrock]

@schmittjoh this helps especially when using this with Symfony so that you can reuse the injected serializer in different parts of the same code run.

[bassrock]

I created a new branch that mimics this but on the latest code base with a few additions. #453

[sfavot]

👍

[goetas]

Do not know if this is still an issue, but #453 (comment) proposes an alternative implementation for this feature

[goetas]

Nested serialization calls currently require big changes, for a relatively small benefit. Probably getting a new serializer instance can partially fix the issue.

Not planning to accept the feature for now.

[xavierlacot]

So this makes the serializer stateful and unusable as a service, with all the performance and memory consumption downsides.

This is a blocker for "real life" production usage, as explained in the previous comments. This has proven to be blocking in at least 3 or 4 of our client projects, and should be clearly stated in the documentation: "jms-serializer is stateful and doesn't care of performance issues".

[patie]

👍