This is not the discussion I was thinking of. Should I just create a new one?

The one I was talking about was the fact that we create a org.hibernate.cache.spi.CacheKey and then each caching impl wraps that in its own concept of a "cache key". Essentially 2 "cache key" instantiations per cache entry. What we had discussed was possibly just dropping our CacheKey (its methods are used in pretty questionable ways anyway) and simply use the cache provider.

Today, we do this in psuedo code:

CacheKey cacheKey = session.generateCacheKey( ... ); // which calls our CacheKey ctor
someCacheRegion.add( cacheKey, someData ); // the region typically wraps our CacheKey

we could instead have:

Object cacheKey = regionFactory.generateCacheKey( ... ); // the SPI calls expect Object already...
someCacheRegion.add( cacheKey, someData ); // It's already the cache providers own key type, no need to wrap! FTW!