Since it is impossible to auto-learn the password from CRAM-MD5, this means 'vchkpw' (v5.4.26d) has a bug because it wrote the md5 hash to the database as the password.
-----Original Message----- From: Tom Collins [mailto:[EMAIL PROTECTED] Sent: Monday, April 14, 2008 10:00 PM To: firstname.lastname@example.org Subject: Re: [vchkpw] not auto-learning passwords On Apr 14, 2008, at 3:46 PM, D. Hilbig wrote: > If I do a plaintext login which will check against the hashed password > stored in the SQL table, I can login with any password. However, a > CRAM-MD5 > login (which checks against the clear password) with any password > will fail. > If both the hashed password and clear password in the SQL table are > NULL > (empty), I'd expect the behavior to be the same regardless of CRAM- > MD5 or > plaintext. IIRC, this was intentional. It's impossible to learn a password via CRAM-MD5, so we fail until we can learn a password through some other method. Password learning happens in vchkpw, but I guess it should move into vpopmail so any app calling the API can have a password "learned". And to those who recommend Dovecot, it probably doesn't do learning either, for the same reasons. I just checked courier 3.0.8, and it looks like it should update the password... I checked dovecot 1.0.10, and found this, "Thanks to Courier-IMAP for showing how the vpopmail API should be used". It doesn't appear to have code that updates the password. Unfortunately, there's no way to update libvpopmail to have it learn the password. We'll have to update the individual apps (courier and dovecot) and get the maintainers to accept the changes into the next release. -Tom !DSPAM:4804f67f120501138613384!