Use CacheableSupportsMethodInterface for performance improvement in normalizers

Problem/Motivation

Back in Symfony 4.1 there was a new interface introduced. CachableSupportsMethodInterface: see this blogpost.

Applications can define lots of normalizers/denormalizers and Symfony must call the supportsNormalization() method for each of them whenever a new object is normalized/denormalized. In theory the result of supportsNormalization() can depend on multiple factors. In practice most normalizers only depend on the type and format and that information is easily cacheable.

That's the trick used to improve the Serializer performance. We've introduced a new CacheableSupportsMethodInterface for those normalizers/denormalizers that only use the type and the format in their supports*() methods:

namespace Symfony\Component\Serializer\Normalizer;

interface CacheableSupportsMethodInterface
{
    public function hasCacheableSupportsMethod(): bool;
}

We've already implemented this interface in all the built-in normalizers, so you don't have to change your code. If you have created your own normalizers, check if they can be cached in the same way and implement the interface if needed.

According to our own benchmarks, this change can make simple apps which use few normalizers up to 10% faster. Complex applications with lots of normalizers, such as API Platform apps, can be up to 40% faster.

All normalizer, at lease in JSON:API only check for support in the protected $supportedInterfaceOrClass property. This means we can speed up normalizers theoretically.

The reason we wouldn't want this, if there are normalizers that find out if they support normalization of a value based on something else than the type and format it might break. This is not the case in JSON:API and it is VERY hard to extend those normalizers.

Proposed resolution

Add and implement CacheableSupportsMethodInterface in Drupal\jsonapi\Normalizer\NormalizerBase.

Remaining tasks

Review code change and performance test.

User interface changes

None

API changes

None

Data model changes

None

Release notes snippet

No needed i think?

Performance tests

To test performance i used JsonApiDocumentTopLevelNormalizerTest. I changed the code so that every test runs its body 1000 times. I've run this 13 times with the hasCacheableSupportsMethod returning TRUE and 13 times when it returns FALSE. I've timed how long the test bodies ran for the tests.

Result overview:

td>

td>

td>