dovecot - Jun 2004 - 1.0-test16 - nearing usable state

http://dovecot.org/test/

Here you go, maildir syncing problems finally fixed (I hope). Well,
except there's this "new-dir-only syncing" optimization which
I'm not
really sure if it works as it should.

The bug really was in index code as I was beginning to suspect. The good
thing is that while trying to figure it out over the last several weeks
I thought of many other potential problems (race conditions) which got
fixed. Those would have been much harder to find than this one :)

Finally I thought I'd stop looking at the code and just begin dumping
tons of debugging data to see where exactly it goes wrong. After few
hours of debugging I figured out it was because message was being
removed from index but not from maildir. 5 hours of sleep later I woke
up, thought a while, figured out where it must be, got up, half an hour
more debugging and found it. :) Code merging two UID range buffers into
one was buggy, inserting one line of code fixed it.

So .. Maybe I'll finally start using this myself now. The next important
problems seem to be some random error messages about index locking
timeouts and some asserts in some situations.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL:
<http://dovecot.org/pipermail/dovecot/attachments/20040616/0523c331/attachment-0001.bin>

Timo Sirainen wrote:
> http://dovecot.org/test/
> 
> Here you go, maildir syncing problems finally fixed (I hope). Well,
> except there's this "new-dir-only syncing" optimization which
I'm not
> really sure if it works as it should.
Good news Timo! Good to hear that you're making progress tracking down 
these problems heading for a 1.0 release. :)

Andrew

On Wed, Jun 16, 2004 at 04:03:01AM +0300, Timo Sirainen
wrote:
> http://dovecot.org/test/
> 
> Here you go, maildir syncing problems finally fixed (I hope). Well,
> except there's this "new-dir-only syncing" optimization which
I'm not
> really sure if it works as it should.
> 
> So .. Maybe I'll finally start using this myself now. The next
important
> problems seem to be some random error messages about index locking
> timeouts and some asserts in some situations.
Yahoo!