Bug#989294: unblock: insighttoolkit4/4.13.3withdata-dfsg1-4.1

[Sebastian Ramacher <sramacher@debian.org>]

On 2021-06-16 12:58:20, Andreas Beckmann wrote:
> Control: tag -1 - moreinfo
> 
> On 31/05/2021 13.54, Sebastian Ramacher wrote:
> > On 2021-05-31 13:32:33, Andreas Beckmann wrote:
> 
> > > Please unblock package insighttoolkit4
> > > 
> > > Let's add some Breaks for smoother upgrades from buster due to the hdf5
> > > library renames.
> > 
> > Let's wait a bit before unblocking this one. We need to figure out how
> > to deal with #988722. If we end up doing a transition for that, this
> > Breaks is not necessary (and potentially harmful).
> 
> With hdf5 and gdal made co-installable, I still find incomplete upgrades
> involving libinsighttoolkit4.12. The remaining non-coinstallable library is
> libnifti2/buster from src:nifticlib (The changelog in the NMU is now
> outdated (still mentioning the hdf5 renames).)
> 
> libnifti2/buster contained
> /usr/lib/libnifticdf.so.2
> /usr/lib/libniftiio.so.2
> /usr/lib/libznz.so.2
> 
> which have been split into separate packages in bullseye:
> 
> libnifticdf2: /usr/lib/x86_64-linux-gnu/libnifticdf.so.2
> libniftiio2: /usr/lib/x86_64-linux-gnu/libniftiio.so.2
> libznz3: /usr/lib/x86_64-linux-gnu/libznz.so.3 (SONAME BUMP)
> 
> Would a transitional metapackage work like for libhdf5*-103?
> 
> # apt-cache rdepends libnifti2
> libnifti2
> Reverse Depends:
>   dicomnifti
>   elastix
>   fsl-5.0-core
>   gifti-bin
>   libgiftiio0
>   libinsighttoolkit4.12
>   libmdc3
>   libmia-2.4-4
>   libnifti-dev
>   libsimpleitk1.0
>   mia-tools
>   minc-tools
>   mitools
>   nifti-bin
>   odin
>   python-nifti
> 
> installing all of them and thereafter checking for the users of libznz.so.2:
> 
> # dpkg -S $(grep -r libznz.so.2 /usr | awk '{print $3}') | cut -d: -f1 |
> sort -u
> dicomnifti
> elastix
> fsl-5.0-core
> gifti-bin
> libgiftiio0
> libmdc3
> libnifti2
> libsimpleitk1.0
> mitools
> nifti-bin
> odin
> 
> reintroducing libnifti2 as a transitional package depending on libnifticdf2,
> libniftiio2 would have to add versioned Breaks against all users of
> libznz.so.2
> OK. that would work for libinsighttoolkit4.12 but if any of the other
> packages is installed, co-installability is gone again.
> 
> We could also drop the Breaks: libnifti2 from libniftiio2 as there is no
> direct file conflict due to multiarchification. But then we would allow
> having two versions of libniftiio.so.2 installed, one linked against
> libznz.so.2 the other against libznz.so.3.
> 
> So I'd favor adding the Breaks against libinsighttoolkit4.12
> 
> An example package affected by this is libotb-apps, with the Breaks added
> the upgrade resolves to
> 
> -  The following packages have been kept back:
> -    libotb-apps
> +  The following packages will be REMOVED:
> +    libinsighttoolkit4.12 libnifti2 libotbapplicationengine-6.6-1
> +    libotbcarto-6.6-1 libotbcommandline-6.6-1 libotbcommandlineparser-6.6-1
> +    libotbcommon-6.6-1 libotbcurladapters-6.6-1 libotbedge-6.6-1
> +    libotbextendedfilename-6.6-1 libotbfuzzy-6.6-1 libotbgdaladapters-6.6-1
> +    libotbice-6.6-1 libotbimagebase-6.6-1 libotbimageio-6.6-1
> +    libotbimagemanipulation-6.6-1 libotbiobsq-6.6-1 libotbiogdal-6.6-1
> +    libotbiokml-6.6-1 libotbiolum-6.6-1 libotbiomstar-6.6-1
> libotbioonera-6.6-1
> +    libotbiorad-6.6-1 libotbiotilemap-6.6-1 libotblearningbase-6.6-1
> +    libotbmapla-6.6-1 libotbmathparser-6.6-1 libotbmetadata-6.6-1
> +    libotbmonteverdi-6.6-1 libotbmonteverdicore-6.6-1
> libotbmonteverdigui-6.6-1
> +    libotbossimadapters-6.6-1 libotbpolarimetry-6.6-1
> libotbprojection-6.6-1
> +    libotbqtwidget-6.6-1 libotbrcc8-6.6-1 libotbsampling-6.6-1
> +    libotbstatistics-6.6-1 libotbstreaming-6.6-1 libotbsupervised-6.6-1
> +    libotbtestkernel-6.6-1 libotbvectordatabase-6.6-1
> libotbvectordataio-6.6-1
> +    libotbwavelet-6.6-1


Besides libotb-apps, is this an issue for any other package? I've randomly
tried two packages from your list above (elastix and odin), but apt
found upgrade paths just fine.

libotp-apps has a reported popcon of 0. Even if that means that there
might be users that just do not have popcon installed, I'm not in favor
of adding even more Breaks for a virtually non-existant problem. And for
those that have libotp-apps installed, the issue is covered by 4.4.5 of
the release-notes.

Cheers

> 
> 
> Andreas
> 

-- 
Sebastian Ramacher