On Fri, Apr 27, 2001 at 12:41:47PM -0700, Ian Clarke wrote: > FCP* seems to be separating out into three layers, these are: > > [* By FCP I also include the XML-RPC interface] > > #1 Basic > Standard insert/request along with pk generation stuff > > #2 Metadata > Intercepts metadata documents and acts accordingly > > #3 Stacks > Supports "stacks", these are a generic name for documents which are > updated by appending a higher number each-time. Stacks may be public or > private, private stacks are SSKs and can only be added to by the SSK's > owner. Public stacks may be verified or unverified. Verified stacks > are tied to a private stack which a server can use to indicate that an > entry in a public stack has been read. This is useful for submission > mechanisms. Finally, stacks may include features to make them more > efficient such as combining them with DBRs. > Stack support is useful since a client can request that a server perform > a stack operation, which may take hours to complete, and terminate > before the operation has completed.
This is not enough. For practical reasons, it is necessary to do something like have separate enumerations for each day or whatever unit of time. The reason is what I call the "find the last item" problem. I've already told Brandon about this, and he has implemented this functionality in IndexClient. This problem is that when someone tries to go from the start of an enumeration, the start of the enumeration may be quite old and therefore many items imbetween the start of the enumeration and the current end of the enumeration may be missing, resulting in one thinking that the current end of the enumeration is before where it actually is. By having separate enumerations for each day, you don't have big enumerations which have to be intact, and therefore you have increased reliability.
