> one of two behaviors. Having two possibilities isn't much different > from having three because the client still doesn't know which it will
Right, we've had this discussion before, and many times. What you're missing is that the reference name argument isn't of much (or any) use to a GUI client, and is there to allow a *user* to type a mailbox name. If user "jones" is using a system where her mailboxes are in a path called "/u/jones/mail/", she could type "inbox" or "projects/imap" to get "/u/jones/mail/inbox" and "/u/jones/mail/projects/imap", respectively. Or she could type "/u/smith/mail/projects/imap" to get another user's (properly shared) mailbox. Or perhaps "~/archives/inbox2002" to get "/u/jones/archives/inbox2002". A GUI client, or, indeed, any client generating the mailbox names internally, shouldn't use the reference name at all, but should just do things like LIST "" "/u/jones/mail/projects/*" rather than doing LIST "/u/jones/mail/" "projects/*" and relying on what the server might do with it. The latter command is useful if the client has allowed the user to enter something, and the user has entered "projects/*" (or "~/archives/%" or whatever). One might still argue that, well, now the *user* still doesn't know how the server will put them together, but, really, the user will learn how to get what she wants with the client/server setup she uses, and the client developer doesn't have to try to second-guess anything. Barry -- Barry Leiba, Internet Messaging Technology ([EMAIL PROTECTED]) http://www.research.ibm.com/people/l/leiba