Telcon: 2016 09 08

Participants

1. Todd Gamblin (LLNL)
2. Thomas Merrick (TAMU-Corpus Christi)
3. Peter Scheibel (LLNL)
4. Patrick Gartung (FNAL)
5. Mike Collette (LLNL)
6. Mario Melara (NERSC)
7. Greg Lee (LLNL)
8. Greg Becker (LLNL)

News

1. GitHub now allows better collaborating on forks

• Maintainers can push edits to PR branches
• Should be useful for working with contributors

1. SC16 tutorial slides are submitted

• Slides will go on USB stick
• Still working on tutorial materials online

1. Thanks again to contributors for:

• extensive documentation updates
• a contributor guide
• module documentation updates

1. Reworking of LAPACK/BLAS linking makes using these packages more robust

• Based on work at EPFL

Bugs

1. Hash stability (#1409) is now merged
2. Further stability fixes to spack reindex are in

• Still tracking down a ref count issue with spack reindex

1. Added a spack debug command to facilitate collecting diagnostics from users
   • spack debug create-db-tarball will create a tarball with Spack's DB and spec.yaml files with which maintainers can check spack reindex. Has been useful for reproducing issues.
2. Still working on some outstanding bugs from last time
3. Fetching issues resolved

Discussion

1. Patrick: Jim Amundsen at fermi still toying with ideas for SpackDev
2. Patrick adding CERN's CMS software package to Spack

• not CMake-based so not the easiest integration
• relies on building an RPM first, then installs the RPM, has mechanisms for finding that
• Build system is called Scram, used on early CERN experiments
• Fermi build and newer CERN builds using CMake -- easier to integrate

1. Tom Merrick: Generated module files should load runtime dependencies, but they're not in some cases
2. Mario:

• having a meeting at NERSC to discuss Spack integration for Cori Phase 1 and 2
• Ran a spackathon with NERSC consultants recently
• Consultants want ability to install packages outside Spack install directory
  • Move opt to /usr/common/software
  • Move module files to a custom location

1. Mike Collette wants a release before porting his code to new version

• We'll try to push out 0.10 in the next couple weeks.
• 1.0 is probably close, but saving 1.0 label for when we've got more package testing going.