Skip to content

Telcon 2016 02 18

Jump to bottom Edit New page
Todd Gamblin edited this page Feb 18, 2016 · 16 revisions

Participants

  1. Todd Gamblin
  2. Jim Galarowicz
  3. Mike Collette
  4. Peter Scheibel
  5. Ben Boeckel
  6. Brett Viren
  7. Greg Lee
  8. David Shrader

Feature updates/use cases

  1. External packages merge (Todd)
  • Almost done, getting merged this weekend
  1. Build dependencies (Ben Boeckel)
  • Most tests are passing, Travis agreeing with local tests
  • Some issues with duplicate dependencies during concretization.
  • YAML reading works
  • Next step is to add tests for packages that actually use build dependencies.
    • See spack test-install
  1. RPM/TOSS3 packaging (Peter Scheibel)
  • Peter now following DESTDIR approach for RPM packaging after trying other, less complex things.
    • Original approach:
      • Was originally trying to use a chroot environment in mock to install directly to the install location.
      • Koji generates mock configuration and isn't too flexible about it, could potentially do this if mock was more customizable in Koji.
    • Not hard for most packages, but ones with bad build systems may not support DESTDIR
    • May need to annotate packages that do not support DESTDIR (i.e. those w/their own build system)
    • TODO for Peter: talk to Jim Foraker about pros/cons of these approaches.
      • Ben Boeckel: can just make packages that do not support DESTDIR depend on autodestdir as a build dependency.
      • Peter: DESTDIR probably works 90% of the time, staging to chroot in mock works 100%, would be nice to get to 100%.
    • Outstanding issues: what to do for Python distutils, boost, etc.
      • Ben: For distutils, --root=${DESTDIR} works for another RPM.
  • Update from Peter on Python .pth files
    • Having separate .pth files works, could simplify Python extension logic a lot.
    • Simplifies packaging Python extensions as RPM files.
  1. OS support update (Mario Melara)
  • Mario on Jury duty; more updates in 2 weeks (hopefully)
  1. Variant forwarding (FNAL folks? and/or Massimiliano/Todd)
  1. PyPI packaging of Spack
  • Keeps coming up, finally had a detailed conversation.
  • Thanks to Brett Viren (see mailing list)
  • Brett involved with HEP-SF, want to unify packaging across high energy physics
    • Use case: use pip to set up Spack for a particular physics experiment
      • Let spack set up the build environment from there.
  • Spack may be a little resistant to being a proper Python package.
    • spack pkg list currently resistant to packaging (assumes git repo)
    • Other parts of Spack rely on certain path structure that doesn't gel with pip.
      • Todd: can make some of the directory paths more configurable so we can use a pip layout.
        • see value in pip packaging for broader exposure, Python projects.
        • want to keep easy clone-and-go-ness of current Spack.
  1. OS X support and Erik Schnetter's many compiler PRs
  2. To be merged soon: 1. OS X RPATH support 1. Compiler wrapper modification 1. RPATH update
  3. Other OS X modifications 1. To be merged *after1. Mario & Greg's new OS support is in 1. should make it easy to have packages conditional on the OS, OS version, etc. 1. Seems to be needed for many of Erik's PRs
  4. Todd trying very hard to merge them -- has been working on externals. * Spack in general getting a lot of new interest * Becoming hard to keep up with PRs -- hopefully building the community can help with some of this. * Need to work on spreading the review/commit burden out more.

News

  1. New interest from FNAL, HEP Software Foundation Packaging Working Group
  • Welcome Brett Viren (BNL)
    • Affiliated with HEP Software Foundation
    • Spack brought to his attention by Jim Amundsen (FNAL)
  1. New folks from ORNL (Robert French et al.)

New packages

  1. New packages from Joe Ciurej (welcome Joe!)
  2. New stuff from EPFL (Quantum Espresso, others)
  3. Byfl packaged by Scott Pakin and Tom Scogland!

Issues

See the Pulse page for details.

  1. Ongoing issues with Intel compilers
  2. think these are to do with Spack's wrapper.
  3. Fixed bugs with spack compilers
  4. Fixed bug in spack create
  5. Fixed lingering bug in dotkit naming that was causing long filenames
  6. dotkits no longer contain variants (boost has too many!)
  7. Many unresolved conversations hopefully addressed by some of notes above.

Discussion

  • Mike C. using Spack on Trinity.
    • Most of the way through building his 50 libraries.
    • Started this morning, almost done now.
  • Followup question from David Shrader on Trinity
    • What type of packages is Mike building on Trinity?
    • ARES and dependencies (see SC15 paper)
Add a custom footer
Clone this wiki locally