[Dovecot] SORT using cached filesystem modtimes

[Rick Flower]

Emmett Lazich wrote:

> Timo or anyone,
>
> This may be a bug or maybe a design decision. Either way I am curious 
> to hear comments.
>
> Why do the dovecot-0.99.14 server's .imap.index files contain a cached 
> copy of each message's date and time as supplied by the filesystem, as 
> opposed to the date and time found in each message header?
>
> When testing various imap clients (eg. tbird-1.0.2, evolution-2.0, 
> outlook2003, squirrelmail-1.4.4) I was intrigued by the behaviour of 
> message date sorting. I toggled on/off the use of server side SORTing 
> in some imap clients, then started forcing rebuilds of .map.index 
> files, and eventually came to the above conclusion.
>
> For normal dovecot server side processing, this condition (if I am 
> correct) may not be a problem, but if someone was to deposit a bunch 
> of maildir files firstly into an imap account (bypassing dovecot), and 
> if this person (eg. me) did not preserve the filesystem/inode file 
> attributes. Then dovecot appears to cache the "wrong" data. What 
> happens then is that some imap clients can display wrong date+times in 
> the message lists, and some imap clients find that server side SORTing 
> does not work.
>
> Keen to hear thoughts/opinions.

Emmet,

You hit the nail on the head for me.. I'm using the latest 1.0-test67 
(IIRC) build and also saw this with all of the other older versions as 
well.. I find it rather odd as well -- mostly for my SPAM directory 
which is where dspam dumps my spam.. About 10-20% of the time, much of 
the spam has a date (according to the Thunderbird email client) of 
12/31/69 at 4:00pm.. If I find one of the offending files though, I 
can't find where it got the 1969 date..  I also occasionally get this 
behavior in my inbox as well..

If you've got any ideas on how to ditch this behavior without disabling 
indexes, I'd be interested to hear how to do it..  I've got some rather 
large Maildir folders (for mailing lists) with >10,000 messages in them 
and would prefer to keep the indexes if possible..

-- Rick