Change cache serialization to use JSON

The PHP requirements for Drupal7 are PHP5.2, so json_* functions would be available.

Cache stored in the table is currently stored as binary data. I think this makes a lot of sense and is also much more compact than any string serialization. While this would be a good move if the cache was still serialized into a string, I don't think we should move from a binary storage method to any string-based storage method.

-1 if I've correctly understood all site caching to be in binary.

Even though the data is stored in a binary field in the cache table it's just a string. serialize() and unserialize() are know to be slow and are called several times to render most any page. Switching from serialize to json has the potential to make a nice performance improvement on both the PHP (the functions are faster) and SQL (data size is smaller) side.

Though JSON won't allow you to store objects that aren't stdClass. Wonder if that's a problem.

http://php.net/json_decode seems to have problems with deep nesting. So looks to a non-option to me.

http://www.phpdevblog.net/2009/11/serialize-vs-var-export-vs-json-encode...
We could have a look at var_export(). But I am not so sure about nested arrays. And besides var_export() does not handle circular references.

Overall I think we should stick to serialize().

Agreed. Also this is only for the default database caching backend.

It is swappable, for cache at least; D7 has a nice OO caching solution. You could write a custom DrupalCacheInterface implementation.

For what it's worth, I've been doing some module development (D6) and experimenting with json_encode() vs serialize() - yes, json_encode (and decode) is faster, but real-life performance gain is small (ok, if you needed to loop 1000 times it would be worth using json). There are definitely some data structures that don't encode properly, not sure why, so switching to json_encode might open a can of worms. I don't have test data to hand, but I tried for example using a customised cache.inc which does json_encode instead of serialise - the resulting change on page display times is negligible.

Further thought: The cache_router module seems to work well. Some cache "engines" such as APC allow direct storage of binary data, so serializing becomes an unnecessary overhead (more to the point, unserializing, since we suppose that cache data is write once/read many). So a suggestion would be: cater for alternative caching systems such as cache_router and let them handle serialization (or whatever encoding they might require) rather than doing it outside of the cache interface.

serialization is already done in the cache interface in D7, so custom backends don't need to do what core does at all.