On Sun, 22 Jan 2006, Timo Sirainen wrote:
Is this planned for a future version or would the benefits be too small to justify that?
I don't think it would benefit much. It'd be one less hash table lookup, but those are fast anyway. And it would require more code..
Is dovecot really hashing plain passwd files and checking if they have changed before access to re-read and re-hash them?
I'd thought these were just read linearly on demand - and then it gets O(n) rather than O(1).
Another idea: can userdb prefetch {} fall back to retrying the passdb entries to see if they come up with useful information?
I'm not exactly sure what you mean, but if the passdb didn't return userdb information, I don't think there's any point to configure userdb to be prefetch..
The passwd file has all necessary information. It is actually automatically derived from the userdb file I'm feeding to maildrop with a short Perl script.
-- Matthias Andree