With the setup 2.885 release candidate, when selecting 'sync'
(dist-upgrade) mode, you may find you have some packages installed which
setup wants to downgrade.
Mostly, these are packages where a version was removed (or relabelled as
test), and hasn't (yet) been superseded by a higher version. Currently,
you are just supposed to know that you need to downgrade these manually.
I have written a script  which analyses historical setup.ini data
from the Cygwin Time Machine (thanks!), looking for anomalies with
version numbers which cause this behaviour.
(caveat: CTM only updates once per day, so problems which existed for
less than a day e.g. the recent hiccup with libgc1 , may not be in
I've implemented a feature in 2.885 which causes setup to selectively
dist-upgrade package versions listed as 'replace-versions:' in
setup.ini, so problem versions can be replaced automatically, which
seems appropriate for all of these.
[*] These versions aren't ordered correctly by the version comparison we
[†] A standalone xwinclip with this version existed briefly (for x86_64
only), before xwinclip was absorbed into the xorg-server source
[‡] The upstream version numbering scheme changed
[§] no idea
> I've implemented a feature in 2.885 which causes setup to selectively
> dist-upgrade package versions listed as 'replace-versions:' in
> setup.ini, so problem versions can be replaced automatically, which
> seems appropriate for all of these.
On 24/02/2018 15:39, Jon Turney wrote:
> I've added replace-versions: annotations in override.hint for the below:
> It's increasingly pointless to compensate for mistakes the longer ago
> they were made, so I don't think I'll bother doing anything about even
> older mistakes which only exist in x86...
I've updated the script which generates this report so it now
understands about replace-versions: lines
On 26/02/2018 21:38, Tony Kelman wrote:
>>> [f] https://cygwin.com/ml/cygwin/2015-03/msg00384.html >>
>> socat 2.0.0-b7-1 was relabelled as test
> I'm seeing an attempted update to socat 2.0.0-b8-1 (which doesn't even
> exist on the mirror I'm using, AFAICT) so that may need the same treatment?
Slightly different, but this version is missing the test: label, which I
added. (it's also missing an install package, for some reason :))