Control: retitle 862343 RM: open-infrastructure-locales-c.utf-8 -- RoQA;
Provides 'locales' and 'locales-all', breaks package builds when installed,
appears unnecessary
On Thu, 11 May 2017 at 15:26:56 +0100, Iain Lane wrote:
> tl;dr: open-infrastructure-locales-c.utf-8 provides 'locales' and
> 'locales-all' but doesn't provide its functionality. experimental's
> resolver chooses this package over the real locales, and breaks package
> builds if they need its functionality. The report has a few instances of
> this breaking builds already.
There is no open-infrastructure-locales-c.utf-8 package in experimental.
The bug primarily affects experimental (because sbuild uses a different
resolver there) but if it is solved by removing
open-infrastructure-locales-c.utf-8, the removal must be from unstable.
I've retitled the bug accordingly.
Unfortunately, the resolver used for experimental considers all packages
in an or-group (Build-Depends: A|B), and all implementations of a virtual
package, to be equally valid candidates (#857557).
See also #862178 (basically a different side of #859912).
If open-infrastructure-locales-c.utf-8 ever reaches a stable suite,
it's possible that the bug would also affect backports, which uses
a different non-standard resolver (aptitude). The default (apt-based)
resolver is unsuitable for suites where it is not desirable to install
the newest possible version.
I agree with the various DDs who have expressed concerns that this
package's use of Provides is invalid. It is not a compatible
implementation of the locales and locales-all packages.
S
