On 25 Feb 2003, Jonas Pasche wrote: >Hi Andy, >> qmail and Courier-IMAP. It certainly broke Courier-IMAP. >IMHO, it broke qmail, too. From the maildir definition: >"Old-fashioned delivery identifiers use the following formats: >* n, where n is the process ID, and where this process has been forked >to make one delivery. Unfortunately, some foolish operating systems >repeat process IDs quickly, breaking the standard time+pid combination." >Replace "some foolish operating systems" with (at least) OpenBSD, and >you've got it. Anyway, qmail itself still uses ><unixtime>.<pid>.<hostname> as the identifier when writing no mails, >thus it gets as mousetrapped as Courier IMAP itself, doesn't it?
Read on: http://marc.theaimsgroup.com/?l=qmail&m=104251026829856&w=2 >However, that's not your part of the story. Here's my idea: >If moving a message from one Maildir to another Maildir, and if you have >to create the message in "new" instead of "cur" there, doesn't it behave >like a completely now delivery to that Maildir? Yes - you've got it! ;) >If it does - aren't you free to change not only the info part from the >filename, but also the uniq part? If it's a new delivery, you're the >delivery agent, and it's your duty to find a good identifier. Yes! >That way, you'd be able to introduce "more unique" identifiers, such as >microseconds (but who guarantees OpenBSD wouldn't repeat the same PID >within a single microsecond?) or the unix_sequencenumber() mentioned on >the Maildir definition, which would eliminate the need to wait a second >before moving from "new" to "cur". No! ;) The randomness in the "unique" section only increases the probability that another Maildir client creates the same name. You must move only at-least-one second old messages to ensure that all messages in cur/ have unique bases. Andy -- Andreas Aardal Hanssen | http://www.andreas.hanssen.name/gpg Author of Binc IMAP | Nil desperandum
