[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Sundials lagging behind upstream



Upstream has at the very least moved files around, and based on the changelog that upstream posted, there are definitely ABI breaks. I did go through the files and update d/copyright, but I need to do one final pass before I push it. I plan on doing that in the next week or so.

James

On Mon, 14 May 2018, 6:27 am Dima Kogan, <dima@secretsauce.net> wrote:
Andreas Tille <andreas@an3as.eu> writes:

> I need to package the R interface to sundials[1] which contains a code
> copy of some more recent version than the Debian package.  I wonder
> whether there is some reason to stick to version 2.7.0 while upstream
> has released 3.1.1.

Hi. Lack of time is the only reason. I chip away at this periodically
and it's getting closer, but the whole this isn't yet done. Packages in
git currently build and their tests pass, but a few things still need
doing. Current list in no particular order:

- MPI tests now hit this bug:

  https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=839387

  To make these pass I use the nofakeroot script from that report. This
  is a hack, but I don't know exactly how to do this properly yet. I
  asked the question in that report, which hopefully will be answered.

- There's a new cmake option: SUNDIALS_INDEX_TYPE that can be set to
  either int32_t or int64_t. With the former, the FORTRAN interface to
  KLU refuses to build. With the latter, LAPACK support refuses to
  build. The cmake invocation complains in both cases. The git tree
  currently chooses int32_t. Somebody needs to understand this
  limitation. The previous release of sundials supposedly supported both
  LAPACK and talking to KLU with FORTRAN, and we should keep both if we
  can.

- The examples patches need rebasing. These were heavily patched to
  allow the examples to work with installed libraries (user installs
  packages, grabs examples from /usr/share/doc/... and is able to build
  and run them). Patches were needed because the makefiles were assuming
  in-tree building of examples only.

- Upstream is bumping the SONAME. Would be good to run
  abi-compliance-checker to see how broken the ABIs really are.

- There're many new sub-libraries. These are called sunlinsol_... I
  don't know what these are yet. Maybe we should be shipping new
  packages?

I'll get everything done, but if we need to get it done faster, I'll
need help. That said, feel free to pester me, that definitely gets
things moving :) I'll poke at it more today.

dima


Reply to: