June 28, 2026 at 4:18 AM, "Oswald Buddenhagen via isync-devel" 
<[email protected] 
mailto:[email protected]?to=%22Oswald%20Buddenhagen%20via%20isync-devel%22%20%3Cisync-devel%40lists.sourceforge.net%3E
 > wrote:

> > 
> > The server is RFC-compliant:
> > 
> i doubt it.
> 
> rfc4551 section 1 says:
> 
> > 
> > > 
> > > A client supporting CONDSTORE extension indicates its willingness to >> 
> > > receive mod-sequence updates in all untagged FETCH responses by >> 
> > > issuing:
> > >  [...]
> > >  This document uses the term "CONDSTORE-aware client" to refer to a >> 
> > > client that announces its willingness to receive mod-sequence updates >> 
> > > as described above.
> > >
> > 
> it doesn't explicitly state that the server MUST NOT send anything 
> condstore-related otherwise, but it's kinda implied.
> 
> rfc7162 updates the extension, and its section 3.1 says:
> 
> > 
> > > 
> > > A client supporting the CONDSTORE extension indicates its willingness >> 
> > > to receive mod-sequence updates in all untagged FETCH responses by >> 
> > > issuing one of the following, which are called "CONDSTORE enabling >> 
> > > commands":
> > >  [...]
> > >  Once a client issues a CONDSTORE enabling command, it has announced >> 
> > > itself as a "CONDSTORE-aware client".
> > >
> > 
> this _still_ doesn't explicitly _forbid_ unsolicited condstore-related 
> behavior, but it tries _really_, _really_ hard to imply it. one has to be 
> particularly dense to not get the hint.

Fair enough — you're right, and I withdraw that framing. mbsync never issues a 
CONDSTORE-enabling command, so under 7162 §3.1, it never becomes 
CONDSTORE-aware, and the server should not volunteer MODSEQ.

> in summary, the server is simply broken. in fact, dovecot agreed last time 
> this has happened:
> https://dovecot.org/list/dovecot-news/2014-October/000276.html

Agreed: the v2.2.14 changelog says "MODSEQ was sent in FETCH reply even if 
CONDSTORE/QRESYNC wasn't enabled." Upstream already fixed this in 2014, so 
Migadu is running something modified or regressed. The targeted MODSEQ skip 
isn't the right fix for the tree — please drop the patch.

> so please report the problem to the provider, and ask them to ensure that 
> upstream is aware of the issue if it's not due to a local modification.

Will do — I'm filing it with Migadu now, citing 7162 §3.1 and the v2.2.14 fix, 
and asking them to confirm whether it's a local patch or stock Dovecot and to 
fix/report upstream.

> i would consider a workaround that generally discards unknown data items 
> (issuing a warning about it once per store), but i'll do that only if the 
> server isn't fixed soon.

Makes sense. No urgency on my end — I'm running a locally patched mbsync in the 
meantime. If Migadu doesn't reply and you do add the general 
discard-unknown-items tolerance, I'm happy to test it against the live server.

Thanks for the quick and thorough review.

Tyler


_______________________________________________
isync-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/isync-devel

Reply via email to