On Tue, 2006-01-10 at 10:38 +0200, Timo Sirainen wrote:

On Mon, 2006-01-09 at 16:46 -0800, Kenneth Porter wrote:

...

On Tuesday, January 10, 2006 12:25 AM +0100 Fredrik Tolf fredrik@dolda2000.com wrote:

...

I'm wondering how close Dovecot can be said to be to 1.0 "Final".

...

From a developer's perspective, a better question is, "What's the roadmap?" What milestones are still unmet before 1.0 can be "released"?

Mostly I just want to get bugs fixed. People still report some annoyingly difficult to reproduce bugs from time to time..

I guess I could release 1.0beta1 soon since my latest changes don't seem to be broken. Except I see this crash now:

PAX: execution attempt in: <NULL>, 00000000-00000000 00000000 PAX: terminating task: /usr/local/libexec/dovecot/imap(imap):12023, uid/euid: 1020/1020, PC: 00000000, SP: 5da5ef9c

Maybe it's because of the IDLE changes. If someone can reproduce it, please tell me how :)

I'm probably not telling you anything new, so I might be stating the obvious, but: have you ever tried valgrind ? Highly recommended.

Mike.

On Tue, 2006-01-10 at 03:24 -0800, Marc Perkel wrote:

I don't know what other people think, if we can do a quick poll here, but comparing the current Alpha 5 to other 1.0 version of software, I think Dovecot is way more solid than most 1.0 versions. In fact, I'd give it a 1.3 if it were me. (Hope this makes sense to everyone.)

In fact, if 1.0 ever comes out, it will certainly be the most stable 1.0 piece of software that I have ever dealt with :-)

Udo Rader

-- bestsolution.at EDV Systemhaus GmbH http://www.bestsolution.at

On 11/01/2006 12:24 a.m., Marc Perkel wrote:

Timo Sirainen wrote:

...

On Mon, 2006-01-09 at 16:46 -0800, Kenneth Porter wrote:

...

On Tuesday, January 10, 2006 12:25 AM +0100 Fredrik Tolf fredrik@dolda2000.com wrote:

...

I'm wondering how close Dovecot can be said to be to 1.0 "Final". From a developer's perspective, a better question is, "What's the roadmap?" What milestones are still unmet before 1.0 can be "released"?

Mostly I just want to get bugs fixed. People still report some annoyingly difficult to reproduce bugs from time to time..

Timo,

For what it's worth, you are a perfectionist. Far more than most other programmers. Most 1,0 versions of software aren't early as solid as your current Alpha version. I'm not suggesting that you lower your standards to everyone elses level but no one expects a 1.0 version to be perfect. In comparison, think about how buggy the current versions of Firefox and Thunderbird are. By your standatds they wouldn't even be up to Alpha even today.

I'm mostly backing Timo on this...

My opinion is that "beta" rather than "alpha" is the right way to describe the software right now, it's certainly not ready for release, but it is certainly usable. My definition of 'beta' is that of software which is feature complete but known to have bugs which affect most if not all people, and I think dovecot fits this description perfectly at the moment, rather than 'alpha'. I personally have a small list of unresolved problems eg epoll with SSL fails miserably after about 5 mins with some weird errors, dovecot-lda is not building for me due to a cvs commit the other day, and in the last 48 hours I have had 61 instances of "(imap) killed with signal 11" in my logs for which I have no idea why or what causes this. Although things work fairly well...but for three users, that's a lot of signal 11s ;-)

I think Timo is right in holding back a bit because there are too many visible bugs and asserts being reported.. Perhaps more users beating on it when it's labelled "beta" might yield some clues as to what is going on and be able to reproduce the bugs before it's declared as "stable". Because once it's labelled as "stable", many people will start judging the code and Timo, and first impressions do last, including how many errors are showing in the logs.

I'll not comment much on Thunderbird, I use it and mostly like it, but quite frankly there are some shocker bugs open in the stable releases especially if you're doing things a bit outside the norm. In other words, I'd hope the release quality of nearly any piece of software was better than TB ;)

reuben

On Tue, 2006-01-10 at 14:27 +0200, Timo Sirainen wrote:

BTW. Postfix debugging howto tells how to run Postfix processes in gdb:

  debugger_command =
      PATH=/bin:/usr/bin:/usr/local/bin; export PATH; (echo cont; echo
      where; sleep 8640000) | gdb $daemon_directory/$process_name
      $process_id 2>&1
      >$config_directory/$process_name.$process_id.log & sleep 5

Maybe the same thing could be implemented for Dovecot..

Attached gdbhelper.c will do pretty much the same for Dovecot. It writes /tmp/gdbhelper.* files when a process crashes (or rather, writes them constantly and deletes them if the process didn't crash). If you have trouble getting backtraces, this should help.

Compile:

gcc gdbhelper.c -o gdbhelper -Wall

And move it to eg. /usr/local/bin/

Change dovecot.conf:

.. protocol imap { mail_executable = /usr/local/bin/gdbhelper /usr/local/libexec/dovecot/imap ..

The same can be done for pop3 and dovecot-auth if needed.

Reuben Farrelly reuben-dovecot@reub.net wrote Wed, 11 Jan 2006 00:59:18 +1300:

| dovecot-lda is not building for me due to a cvs commit the other | day

What would that be? If it's related to com_err.h, try regenerating your Auto* stuff (sh autogen.sh).

-- Linus

On Wed, 2006-01-11 at 01:55 +1300, Reuben Farrelly wrote:

On 11/01/2006 1:43 a.m., Linus Nordberg wrote:

...

Reuben Farrelly reuben-dovecot@reub.net wrote Wed, 11 Jan 2006 00:59:18 +1300:

| dovecot-lda is not building for me due to a cvs commit the other | day

What would that be? If it's related to com_err.h, try regenerating your Auto* stuff (sh autogen.sh).

It is. Already wiped, re-checked out the tree and rerun autogen.sh in that order:

if gcc -DHAVE_CONFIG_H -I. -I. -I../.. -I/usr/src/dovecot/dovecot -I/usr/src/dovecot/dovecot/src/lib -I../../src -std=gnu99 -g -O2 -Wall -W -Wmissing-prototypes -Wmissing-declarations -Wpointer-arith -Wchar-subscripts -Wformat=2 -Wbad-function-cast -I/usr/kerberos/include -MT addr.o -MD -MP -MF ".deps/addr.Tpo" -c -o addr.o addr.c;
then mv -f ".deps/addr.Tpo" ".deps/addr.Po"; else rm -f ".deps/addr.Tpo"; exit 1; fi In file included from sieve_interface.h:37, from script.h:31, from addr.y:34: sieve_err.h:10:21: error: com_err.h: No such file or directory

Ah. This is because lda-config.h never gets included, so the com_err.h location check doesn't work. The included is for Dovecot's config.h. Hmm..

Quoting Marc Perkel marc@perkel.com:

For what it's worth, you are a perfectionist. Far more than most other

Nothing wrong with that.

programmers. Most 1,0 versions of software aren't early as solid as your current Alpha version. I'm not suggesting that you lower your standards to everyone elses level but no one expects a 1.0 version to be perfect.

No, but we do expect it to work (not crash, etc) except in very rare cases.

And - I can tell you that the Alpha label is definitely hurting you and my recommendation as a person with marketing experience is do one or two beta versions, call it done and then 1.0 and then start fixing it. If you do that and you get a lot more people on board then other people will help you track down your bugs.

I am one of the view who have not been able to use dovecot because it isn't stable enough, and as such I'm one of the few who stands behind the alpha naming of it.

I don't know what other people think, if we can do a quick poll here, but comparing the current Alpha 5 to other 1.0 version of software, I think Dovecot is way more solid than most 1.0 versions. In fact, I'd give it a 1.3 if it were me. (Hope this makes sense to everyone.)

Count me as one who is still eagerly awaiting a stable 1.0 release, and who has not had any luck so far with the alpha version up through alpha3 (I've not tried anything since then).

Now, I've noticed a lot of people who say "I use it with hundreds of users" but I want to use it with thousands, not hundreds, and that does make a difference. I also want to use it in a non-heterogeneous system, which is no doubt 90% of my problem...

When timo thinks it is beta quality, I'll try it out, and hopefully it will work for me. Until then, I will wait and suffer with another solution...

-- Eric Rostetter The Department of Physics The University of Texas at Austin

Go Longhorns!

Quoting Kenneth Porter shiva@sewingwitch.com:

On Tuesday, January 10, 2006 10:07 AM -0600 Eric Rostetter rostetter@mail.utexas.edu wrote:

...

I am one of the view who have not been able to use dovecot because it isn't stable enough, and as such I'm one of the few who stands behind the alpha naming of it.

How stable is "enough"?

Less than one segfault per day would be nice.

What server are you comparing it to? And if

Just about anything else?

another server is more stable, what would motivate us to switch back to Dovecot once it achieves comparable stability?

Speed. It is much faster than the others. Plus it is more actively developed than some older ones, and many claim it is more secure, etc.

I switched from UW-IMAP based on the Fedora switch, and Dovecot seems much faster than the UW code. (I'm using sendmail/procmail with mbox on the delivery side, and UW's own mbx format when I was using UW-IMAP.)

See, you already knew the answer then.

The one issue I've seen with Dovecot (still using 0.99) is the occasional corruption of one Thunderbird user's Trash folder with the insertion of a few K of nul's at the top.

Glad it works for you.

I'd guess that Timo's designation of 1.0 as "alpha" is what keeps Fedora from updating to it.

And you'd be wrong...

-- Eric Rostetter The Department of Physics The University of Texas at Austin

Go Longhorns!

On Jan 11, 2006, at 8:18 PM, Kenneth Porter wrote:

I didn't know if there was a reason to switch away from other
servers, such as Courier or Cyrus.

Well two things convinced us:

Dovecot is fast and stable. The developer answers requests/questions/pleas, sometimes within 10
minutes.

Plus there is an active list where people actually answer questions.
I can't say that for courier or cyrus. The lists for both of those
servers are full of people with bad answers and bad attitudes! Who
wants that?

Roger Weeks

On Wed, 2006-01-11 at 21:52 -0800, Roger Weeks wrote:

Plus there is an active list where people actually answer questions.

Indeed. Thanks everyone for all the answers to my mail!

I've been using 1.0-alpha5 about since I asked, and it has been working fairly well for me since then. There are just the occasional problem of Dovecot taking extremely long (>1 min) to do some operation every once and then. When I've tried stracing it during that time, it has once been doing something with the index, and the other times it seems to be in the ordinary idle loop, even though evolution says that it's attempting to do something (I don't think it's an Evo problem, since I never experienced it with UW-IMAP).

At the end, though, it is *much* nicer than using UW-IMAP, so I have no real reason to complain. My reason to use Dovecot is that it seems to be just about the *only* IMAP server that is capable of both Maildir storage and GSS-API authentication. When your mboxes start to grow to sizes of ~50 MB (and are accessed over NFS), Maildir is a blessing, to say the least.

Fredrik Tolf

Quoting Timo Sirainen tss@iki.fi:

What problems did you have? I'll be releasting 1.0beta soon, and there's not all that many fixes since alpha5.

As I said in a previous post, I've not tried it since alpha3, so don't wait on me... Alpha5 might work for all I know...

-- Eric Rostetter The Department of Physics The University of Texas at Austin

Go Longhorns!

On 11/01/2006 1:56 a.m., Marc Perkel wrote:

For what it's worth I just went to the Fedora Core 5 test release download page and Fedora still has Dovecot 0.99.14 in it's list of files. I'm guessing they are afraid of the Alpha label. I'd really like to see Fedora 5 have Dovecot 1.0. You might want to look into this.

http://download.fedora.redhat.com/pub/fedora/linux/core/development/x86_64/F...

It's not the 'alpha' label that is the problem, as far as I can tell, it's more a case of someone actually maintaining the package. There was a discussion about this on the Fedora Devel mailing list a few weeks ago whereby a small number of us were petitioning dovecot to be upgraded, but it seems to have fallen on deaf ears.

The main bugzilla entry for this is:

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=170960

Note in comment #1 where the maintainer says "I'm being transitioned to other responsibilities and package ownership of dovecot is transferring at the same time. The new (yet to be determined) package maintainer for dovecot will need to pick up the ball on this issue."

However that 'someone else' has yet to pick up any ball, so the package has remained unchanged since 0.99.

One of the users has kindly updated the rpm spec file and patches for dovecot, but so far no response nor update from Redhat has been forthcoming.

If other people feel inclined to do something then please add supportive comments to https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=170960 - maybe someone at Redhat will hear if enough people ask ;-)

reuben

On 01/10/2006 12:10:50 PM, Marc Perkel wrote:

Reuben Farrelly wrote:

...

On 11/01/2006 1:56 a.m., Marc Perkel wrote:

...

For what it's worth I just went to the Fedora Core 5 test release download page and Fedora still has Dovecot 0.99.14 in it's list of files. I'm guessing they are afraid of the Alpha label. I'd really like to see Fedora 5 have Dovecot 1.0. You might want to look into this.

http://download.fedora.redhat.com/pub/fedora/linux/core/development/x86_64/F...

It's not the 'alpha' label that is the problem, as far as I can tell, it's more a case of someone actually maintaining the package. There was a discussion about this on the Fedora Devel mailing list a few weeks ago whereby a small number of us were petitioning dovecot to be upgraded, but it seems to have fallen on deaf ears.

I don't think it would fall on deaf ears if version 1.0 were declared. I think this is an example where the "alpha" label or even "beta 1" would be a barrier. if Timo declared 1.0 then it wouldn't be an issue.

See also https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=170960

Regards, Willem Riede.