Marc Patermann a écrit : > Hi, > > Bastian Blank schrieb: >> On Sun, Feb 08, 2009 at 03:38:22AM -0500, Sahil Tandon wrote: >>> This works as I'd expect, but will it break anything else? >> >> Yes. It will break the complete mail handling of the client. _Never_ >> ever touch a message id. > Not all users are dumb. ;) > > Sender: I'm missing your response, didn't you get my mail? > Recipient: Can you tell me the message-id? > Sender: Yes, *looks it up in "Sent"-folder* it is > 012...@domain.tld. > Recipient: One Moment please, *searches his mailbox* no, it's not > here ... > > Just one of the cases you might break what user would expect. >
This "case" is a bit artificial. People would search for the recipient and the subject. and even if they find the message in the Sent folder, it doesn't help much. MTA logs are more helpful. The real problem is breaking threads. so more testing is needed. anyway, my opinion (at this time, as I didn't yet find a sufficient counter argument) is that the message-id should be generated by the MSA: - it is easier to guarantee unicity, because in general, MSAs know their "public" hostname - it is easier to guarantee well-formed Message-IDs. MUAs (and I am including web applications and the like) have a habit of implementing broken behaviour - it simplifies the MUA task. admins would (some day:) spend less time configure the many MUAs in their network (there are obviously fewer MSAs). note that I am not using "hide internal infos" or "ease backscatter detection" as arguments, even though these look "interesting" to customers.
