home *** CD-ROM | disk | FTP | other *** search
- NNTP's model for keeping per-user seen state has two problems:
-
- * It needs each message to have a unique-within-group identifier which
- is preserved across sessions. For performance reasons, it is a good
- idea for these identifiers to be strictly ordered. IMAP would either
- have to add such a construct, or declare one of the existing
- constructs as having this property. My preference would be for
- INTERNALDATE to have this property.
-
- * Storing this type of state information on the client is contrary to
- the goal of nomadic access. Users cannot access the server from
- different clients without moving the state information between the
- clients.
-
- In some sense, we have to paint ourselves into one corner or another.
- We have to pick either the client or the server as being responsible
- for keeping the per-user seen state. We can (and should) make it
- possible to do it the other way, but one model should be the rule and
- the other the exception. Otherwise, different clients will handle
- this different ways and confusion will reign.
-
- I suppose a good tack a client implementation could be "try to keep
- /SEEN state on the server. If this isn't possible, keep the state
- yourself." Such a client could detect whether FETCHing a message
- causes the /SEEN flag to be set.
-
- > I'm hoping IMSP can be used to bring the per-user
- > state file to the client from wherever it might be stored.
-
- An IMSP client can store almost anything it wants in an IMSP option.
- The difficulty is getting different clients to use the same
- conventions for storing the same information.
-
- Similarly, an IMAP server implementation could use some external
- distributed database to store the per-user state. I plan to use this
- approach if/when we get around to implementing replicated bboards.
-
- --
- _.John G. Myers Internet: jgm+@CMU.EDU
- LoseNet: ...!seismo!ihnp4!wiscvm.wisc.edu!give!up
-
-
-
-
-