Re: [infinispan-dev] Improved Marshaller

Manik Surtani wrote: > I just checked the following in trunk. > Galder, you may want to do something similar for your JBMAR > prototype. > The changes are around InternalCacheEntry and now InternalCacheValue. > Previously, the marshaller stored an InternalCacheEntry as [key, > value, canExpire, created, lifespan, lastUsed, maxIdle], setting > either of the last 4 elements to -1 if they are unused. In a > system mainly comprising of mortal values, this is wasteful since > each of the last 2 values - which would be unused - are longs and > still written to. > So this change identifies the InternalCacheEntry and writes just > what is needed, hence the use of specific magic numbers for each of > the InternalCacheEntry implementations. Using a different magic > number for each entry type allows me to trim what is written. > Same approach used for InternalCacheValue. > Yes, even more ugly switch statements - and this is why we need > JBMAR to clean this up - if and when it can match performance and > compactness of the resulting byte array. :-) This has been done now.