sundials_2.7.0+dfsg-1~1_amd64.changes REJECTED

Dima Kogan dima at secretsauce.net
Tue Apr 11 18:56:25 UTC 2017


Andreas Tille <andreas at an3as.eu> writes:

> Please let me know if any help might be needed. New queue is currently
> very short - so this might be a good time for uploading.

Hi. Before trying to upload again, I'd like to actually get this into a
state when I myself know that it's semi-working. My benchmark is that I
should be able to

- install the packages
- build the examples from the packages
- run the examples I just built

I worked on it a bit last week, and it's in much better shape now, but
still not upload-worthy, I think. I talked to James Tocknell about
progress to coordinate with him, but I don't think he had the cycles to
do anything here.

The current state of the tree in master:

- All the various nvec types split into separate packages. All the
  different PARALLEL types split into separate packages

- A single libsundials-dev does everything. This probably is wrong, but
  is good enough to test. Let's get everything else right, then split
  this

- debian/control has bogus descriptions and dependencies

- I'm not building examples WHEN BUILDING THE PACKAGES, but instead I'm
  shipping them in libsundials-doc (should instead go into the -dev
  package of whatever is being demonstrated). The user should be able to

    1. cd /usr/share/doc/whatever/examples
    2. mkdir build
    3. cd build
    4. cmake ..
    5. make
    6. run example

  This required some patching to the example cmake files (I've only
  touched arkode for now), but this now works at least for
  arkode/C_serial

  Once we're happy with how this is done, we can apply these patches to
  all the examples

- The pkgconfig business for PETSc wasn't working. Rather pkg-config
  itself was working, but cmake was being stupid and doing something
  wrong with the results. It took me way too long to debug it, and I
  never did figure it out. But we have Build-depends, so we don't need
  to find anything: we already know where everything is. Thus the
  current makefiles simply take the output of pkg-config and don't test
  it. If somebody wants to fix that, go for it.


In any case, the packages are building are working somewhat. I'll pick
this up again when I have the cycles, but if somebody wants to push on
it before then, PLEASE do so!

dima



More information about the debian-science-maintainers mailing list