Telcon: 2017 01 26

Participants

1. Todd Gamblin (LLNL)
2. others TBD

News

1. NASA had a Spack tutorial yesterday (thanks Elizabeth!)
2. Spack v0.10 was released on the 17th.

• Lots of bugfixes
• New platform/os/target architecture support (has been in develop for a while, but was not in 0.9.1)

1. Spack is now on twitter @spackpm
2. Spack environments working group met yesterday, with some participants from the lsat telcon

• Focuses on how to make the environment support in spack consistent (modules and otherwise)
• How can we make it easier to deploy a development environment with Spack?
  • Build a large code, work on 4 packages, build again, etc.
• Others welcome to add to the feedback there
• Seems to be some consensus on what is needed to overhaul spack view, spack setup, and other commands.
• We'll have another meeting and try to come up with potential interfaces.

1. Plan for v1.0 continues to focus on:
2. Testing!
3. Better concretization algorithm
4. Kiel Friedt has made a prototype spack test-suite command for running combinatorial tests

• Takes a YAML file as input describing tests to run (sort of like Travis CI build matrix)
• Posts results to spack.io/cdash
• Hoping to polish this for general use soon.
• Let us know if you'd like to run a dashboard where you can post your test results.

1. Switched from coveralls to codecov for coverage

• Has a nice Chrome plugin so you can see test coverage inline on GitHub

Feature highlights

1. As usual, check the pulse page to see what's going on
2. More detailed docs on using MPI in the packaging guide #2838
3. Updated packaging docs cover AutotoolsPackage, CMakePackage, MakefilePackage classes #2780

• Note that these new packages let you run build phases separately for packages that support it:
  • spack configure
  • spack build
  • spack install
  • etc.

1. Fixes to the spec parser. You can now refer to specs by hash properly, e.g. python/1c3fba
2. Better support for xl_r compilers from IBM

• xl_r is a separate compiler because it's more than just XL with different flags
• Runtime libs are different for xl_r than xl, particularly for Fortran

1. Thanks to Adam Stewart and Elizabeth for many OCD fixes that help Spack be more consistent

• uniformly lowercase command help
• Convert everything to AutotoolsPackage
• Improved spack create to auto-detect build systems and use appropriate package
• Uniformly lowercase package names
  • now r, not R
  • dashes in package names instead of underscores.

Discussion

• Tom Merrick: How do I pass flags to MPI compilers?
  • using cflags='-O3 -etc' in packages.yaml, but not seeing the flags in the build log
  • Todd: compiler flags passed that way are injected within Spack's compiler wrappers, so you wouldn't see them in the build log.
    • You CAN see them by running spack -d install <package. This will give you two log files in the current working directory, one with the input arguments to the compiler wrapper and one with the output arguments passed to the actual compiler. Diffing these files with some diff program that can do sub-line diffs will show you exactly what got added by Spack.
    • Todd uses opendiff on his mac to do this.