Fine with me. On Apr 1, 2013, at 5:41 PM, Robert Sparks <rjsparks@xxxxxxxxxxx> wrote: > On 3/28/13 1:17 PM, SM wrote: >> Hi Eric, >> At 05:13 28-03-2013, Burger Eric wrote: >>> Rather than guessing all of the bad things that could happen, I would offer it would be better to say what we mean, like: >>> The IMAP interface MUST NOT provide any IMAP facilities that modify the underlying message and message metadata, such as mailbox, flags, marking for deletion, etc. If the client is authenticated and authorized, the IMAP interface MUST provide per-user marking of the underlying message as read or flagged. >> >> The IMAP interface is a front-end to the read-only mailboxes (archive). It's easier to do what is mentioned above. > I'm taking more or less that approach in my working copy (I'll be posting -06 soon). >> >>> Something to ponder: >>> I use the IMAP interface once, mark a bunch of things as read, and then decide never to use the IMAP interface ever again. How long does the server need to keep my (per-user) marking metadata? E.g., besides CPU and I/O issues, there is a potentially unbounded storage problem as well. It is unbounded because in IMAP I can assign any kind of label (marking) to a message, even ones I make up. >>> >>> One thought for an approach to a solution: >>> 1. per-user markings expire after X time units (six months?) >>> 2. per-user markings may take up at most X storage units (512KB?) >> >> I would go for both. > Instead, I propose that we make it possible to notice an abuser and turn off access (this is what -06 will contain). > > I don't believe we could come to a consensus on an automatic expiry of state - there are use cases I can think of where any short > expiration (like 6-months) would be infuriating. > > If keeping this state for normal use turns out to be too expensive for us, then we will have learned something, and can start talking about future IMAP work in general to help systems mitigate that expense. >> >>> Per-user metadata can be incredibly useful - I might label things by project, work group, draft, mumble, or foo. I would not want to limit the labels to red or green. However, we need some predictable limit as well. >> >> Yes. >> >> Regards, >> -sm >