On 4 Oct 2017, at 14:43, Anton Gladky <gl...@debian.org> wrote: > 2017-10-04 15:13 GMT+02:00 Andreas Tille <andr...@an3as.eu>: >> Hi, >> >> to deal with #877419 I'd suggest the following approach: >> >> 1. Ignore test suite errors on those architectures that are >> known to fail (see attached patch, NOT TESTED, please review) >> 2. Drop severity of the bug from serious to important >> 3. Fix bug #877419 later but let pandas and its dependencies >> (statsmodels and lots of others) migrate to testing >> >> If you think this is a sensible strategy and you want me to upload >> pandas with this patch I would also do >> >> 4. Move pandas to Debian Science as discussed before >> >> Kind regards >> >> Andreas. > > Hi Andreas, > > one more option is to drop pandas on those archs (filing > RM bug) and fill the list of supported archs explicitly > in d/control.
Please, no, don't put a specific list of architectures in d/control. The inability to build pandas on those architectures is a bug, whereas explicit architecture lists in d/control should only be used if the package fundamentally cannot work on others. Otherwise these failures disappear on buildd.d.o and porters are unaware of them. You can still file an RM for the currently-broken architectures if you really need the new version to migrate to testing, and then they will reappear once the build starts succeeding again. Regards, James
