Please add the bounce encapsulation patch for qmail, located at http://www.finnie.org/software/qmail-bounce-encap/qmail-bounce-encap-20040120.patch
The latest patch is backwards compatible with "normal" qmail; you must enable it via /var/qmail/control/usebounceencap in order to alter qmail's functionality, so there should be no adverse affects to including this patch.
Short intro: Several MTAs encapsulate the original message in a MIME
message/rfc822 part when sending a bounce. qmail, however, does not.
This makes it difficult to extract the original message on graphical MUAs
(Evolution and Outlook for example let you drag and drop encapsulated
messages into other folders as if they were regular messages (because they
are!)). This patch adds a message/rfc822 encapsulation to the bounce
messages that qmail sends out.
Steps to Reproduce:
reading your ChangeLog there for the message unique identifier, look at using time + process id (just like maildir filenames). you should be able to borrow the code from the delievry code.
qmail-send does not fork itself out for each message processed, so the pid is going to always remain the same. And since the time is not hires, it's possible that both the bounce and the bounced bounce can be processed in the same second, which really screws up the MIME encapsulation. (Unlikely, but before I added the stdlib.h/rand() functionality, I would see it occur every couple weeks.)
In response to the previous comments, RFC1918 tagging functionality was change, and has been running beautifully for the last few months.
New version at http://www.finnie.org/software/qmail-bounce-encap/qmail-bounce-encap-20040210.patch
Heh, I don't think anybody noticed, but that previous comment should say "RFC822", not "RFC1918" :)
Added to qmail-1.03-r16. Could you test it, please?
Sorry for the delay, seems to work fine. Thanks!
Closing the bug. Thanks for testing!