What to do about CONDSTORE for GMail for TB 24?

Bron Gondwana brong at fastmail.fm
Wed Sep 4 12:59:38 UTC 2013

On Wed, Sep 4, 2013, at 07:33 AM, Ehsan Akhgari wrote:
> On 2013-09-03 3:29 PM, Mike Conley wrote:
> > Hey tb-planning,
> >
> > So during our bi-weekly Thunderbird meeting today, the issue of CONDSTORE (which was brought up in a separate thread on tb-planning, but I'm splitting out to primarily discuss what to do about the issue for TB 24) was brought up.
> >
> > Here's what we know:
> >
> > 1) GMail has recently switched on support for CONDSTORE, which allows for more efficient IMAP queries.
> > 2) We've had reports of buggy behaviour with GMail - here on tb-planning, as well as in Bugzilla[1]. Specifically, people have been reporting that Thunderbird stops retrieving their GMail mail (or at the very least, doesn't get it immediately). Wayne and Roland are still getting a handle on any murmurs in the forums and GetSatisfaction.
> > 3) The problems being experienced with CONDSTORE are not well understood.
> >
> > So it's unclear how big of an issue this is - but in the worst case-scenario, a growing number of users may find that Thunderbird is not getting their GMail mail.
> >
> > With 2 weeks left to go until a major release, should we be using that as an opportunity to pack a fix / workaround into 24?
> >
> > There is a hidden pref that, if flipped, disables Thunderbird's CONDSTORE support. Irving also mentioned that we could also instruct TB to "not believe" GMail when it says it supports CONDSTORE.
> >
> > It's hard to make a call here because, as far as I can see, our data on this problem is really lacking.
> >
> > What should we do? Anything?

Do you have any network dumps of the traffic to/from gmail in which this bug manifested?

I don't know gmail's code (other than knowing people at gmail) - and I haven't worked with their condstore implementation apart from glancing quickly at it while it was in alpha and pointing out some bugs - but I do know a lot about CONDSTORE (I rewrote the Cyrus implementation to make sure it matched the RFC perfectly).

> Have we considered reaching out to Google about this?  We have a 
> communication channel with them for these kinds of issues and I will be 
> more than happy to forward technical information about the problem to 
> them and get the right people on both sides in touch with each other.

I suspect it's very unlikely that Google can turn around a fix in 2 weeks.  From the description of the problem (labels related) it sounds like fixing their CONDSTORE implementation is going to be a serious chunk of work.

Based on that and a general "inefficient is better than incorrect" preference, I would second (third, nth?) the suggestion to add a quirk workaround that disables CONDSTORE to gmail for now.



(I suspect that 

  Bron Gondwana
  brong at fastmail.fm

More information about the tb-planning mailing list