[Debian-med-packaging] Bug#848758: Latest upgrade of Numpy breaks tests of other packages (Was: Bug#848758: python-skbio: FTBFS: Test failures)

[Adrian Bunk]

On Wed, Dec 21, 2016 at 03:38:36PM +0100, Ole Streicher wrote:
> Hi Adrian,
> 
> On 21.12.2016 15:29, Adrian Bunk wrote:
> > On Wed, Dec 21, 2016 at 02:05:57PM +0100, Ole Streicher wrote:
> >> ...
> >> The according functions are since then marked as "deprecated" and issue
> >> a warning. Numpy introduced this change with 1.11rc already, which was
> >> uploaded to Debian about 6 months ago. After it appears that many
> >> packages (als in Debian) have problems with this, they postponed it for
> >> another release -- which happens just now.
> >> ...
> > 
> > You are saying you knew for 6 months that many packages in Debian
> > have problems with the changes that have just been uploaded.
> > 
> > Why were no bugs filed a few months before the upload?
> 
> Most of my own programs have CI tests which immediately gave a signal. I
> forwarded this upstream (mainly as github issues) and took part in the
> diskussion between upstream (f.e. astropy) and numpy, so that finally
> the decision was made to postpone.
> 
> Since I looked only on my own packages, there was no need to file bugs;
> especially since the problem was postponed. The upstream maintainers
> that were informed took measures in between; therefore I have only one
> package (astroml) currently failing.
> 
> I had no information about any other package; therefore I couldn't file
> bug reports. I can just recommend: If a Python package comes with tests,
> use the Debian CI infrastructure for it. This helps a lot as an early
> indicator of problems. And since the problems appear on packages with
> build time test, I would like to ask back: why don't you use CI?

I am not personally involved with python-skbio, but I can clearly 
understand why Andreas does not have the spare time to manually
check such things in all his packages:
  https://qa.debian.org/developer.php?login=tille@debian.org

Looking at the CI logs for python-skbio, it also seems that breakages 
caused by pandas are the only ones that are visible there.

When tests fail in unstable there is very soon an RC bug from Lucas 
(this is the third such bug for python-skbio this year), so once the 
breakage is in unstable it anyway gets reported pretty quickly.

I can clearly understand why Andreas is unhappy to see when such a 
change is uploaded without prior warning 1.5 months after the start
of the freeze, and 1 week before an important deadline for his package.

> Best regards
> 
> Ole

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed