>
>No! In GC-like allocation mode I meant to use malloc to allocate
>memory in big chunks (> 8K) and use Last_Allocated_Byte counter for
>each chunk in palloc() to "allocate" memory. pfree will do nothing.
>GC-destroyer will just free a few chunks - without any scans.
>Many GC-storages will be available simultaneously (GC_Storage_Identifier
>will be returned by StartGCAllocation() call and used by EndGCAllocation()
>to free memory in given storage). GC-allocations will be made in current
memory
>context (in term of postgres) ==> code using special memory contexts
>(relation cache etc) will not be affected at all (switching to another
>context will stop GC-allocation untill first context restored)
>as well elog(ERROR) clean up feature.
>
This seems like an effective strategy too me. It also provides a solution
to the 8 byte alignment problem.
With regards from Maurice.