Bug#454498: 2.3.13 now available.

Duncan Gibb Duncan.Gibb at SiriusIT.co.uk
Mon Dec 22 17:53:06 UTC 2008


Sven Mueller wrote:

SM> Duncan is now also able to use the SVN repo

Unless anyone has an objection - or a better idea - I intend to copy the
current cyrus23/cyrus-imapd-2.3-2.3.8 as cyrus23/cyrus-imapd-2.3-2.3.13
and start syncing that with the less "progressive" aspects of the
internal Sirius 2.3.13 build.

I'll ask on this list before adding anything controversial.


SM> There need to be some upgrade scripts since BDB is historically
SM> bad at backwards compatibility and cyrus uses some BDB databases
SM> by default, which need to be upgraded if moving from 4.2 to 4.4
SM> or 4.6.

I agree and will try to work on this in due course.

My approach to BDB versions is that this should be a build-time, rather
than commit-time choice.  So where the existing debian/control has

  Build-Depends: [..], libdb4.5-dev, [..]

and debian/rules hard-codes

  DBENGINE=BerkeleyDB4.5

I propose to use

  Build-Depends: [..], libdb4.4-dev | libdb4.5-dev | libdb4.6-dev, [..]

and the slightly obtuse

  export BDB_VERSION ?= $(shell dpkg-query -l 'libdb4.[2-6]-dev' | grep
^ii | awk '{gsub("[libdev-]", "", $2); print $2}')

  DBENGINE=BerkeleyDB$(BDB_VERSION)

  configure [..] --with-bdb=db-$(BDB_VERSION) [..]


I have no idea whether this will meet official Debian packaging standards.


SM> Also Henrique (hmh) seems to distrust newer BDB versions by default.

Not all recent versions have been good.  Here at Sirius we normally
build Cyrus on the same hosts as we build OpenLDAP, so per-distro we
pick a version of BDB they both work well with.  For Etch deployments
that's still 4.4.


SM> I know at least two companies (besides my own) that have
SM> cyrus-imapd in a simple failover scenario where the data files
SM> of cyrus live on a shared DRBD device and are only available on
SM> one of the two servers at any given time (which means that the
SM> upgrade scripts should warn about missing data files, but not
SM> die IMHO

I agree but for a completely different reason: we often configure
traditional Murder front-ends with "partition-default: /dev/null" so
admins can't accidentally create local mailboxes.  This breaks the
current update scripts but it hasn't annoyed us enough yet to package a
better way of doing things.


Cheers


Duncan

-- 
Duncan Gibb, Technical Director
Sirius Corporation plc - The Open Source Experts
http://www.siriusit.co.uk/
Tel: +44 870 608 0063



More information about the Pkg-Cyrus-imapd-Debian-devel mailing list