idled support?

[Henrique de Moraes Holschuh]

On Sun, 12 Feb 2006, Sven Mueller wrote:
> Is this just personal taste or is there a real reason against usage of
> idled?

In 2.1 it ain't safe, idled is not well debugged, CMU doesn't user it, and
it would crash.  After idled crashes, you lose *all* IDLE functionality. Not
nice.

*IF* in 2.2 you can run it as a babysat service, so that cyrus master will
restart it everytime it does, AND cyrus doesn't go gaga if idled dies (i.e.
it does the sensible thing to all IDLEing connections, which is to kick them
as if new data was received)... then enabling idled would be a good idea.

That's why it is in the old 2.1 TODO list (run idled as a babysat service).

> I still have noticeable lag between changing a folder in one client and
> seeing the result in the others (tried thunderbird and outlook as well
> as kmail, changing the folder - INBOX of a special test account - from
> squirrelmail). I built a test version with idled support and ran it with

And the only reason you will *see* any change in others correctly (at least
in seen state) is due to the patches we have applied on cyrus 2.1 (and I
sure hope, 2.2 :-) ).

> idled. This made Thunderbird and Outlook noticeable faster in seeing the
> change (almost instantly), but not kmail (perhaps due to the old version
> I tested). Running this for more than 24 hours now, I didn't notice any
> problem (neither directly nor in the logs). So I would like to know your
> reasoning.

It just needs to be looked in depth to see how sturdy it is. If it is
sturdy, we enable it. If it is not, we have to _fix_ it first...

If you cannot read it as a service, I don't think it is sturdy enough,
experience tells me it is bad enough to trust cyrus master to never dump
core/exit, trusting idled on top of that ain't a good idea.

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh