Telcon: 2018 03 08
scheibelp edited this page Mar 8, 2018
·
23 revisions
- Peter Scheibel (Host)
- Patrick Gartung
- Daniel Topa
- Paul Ferrell (LANL)
- Jennifer Green (LANL)
- Ben Boeckel
- Greg Becker
- Reviewing PRs in Spack: https://github.com/spack/spack/wiki/Reviewing-PRs-for-Spack
- Proposal: #7409 Update-in-place functionality
- E.g. adding security updates to a dependency
- Do you have use cases related to this?
- What do views not handle in this?
- Greg:
- If you update spack and the hashing mechanism changes, then all packages are rebuilt (rare)
- If you update spack and a package adds a variant, then if you
spack install
it or a parent again spack will install new instances - Otherwise, updating spack doesn't force rebuilding any packages (and generally the same user spec will produce the same spec DAG and package hash)
- Paul: openssl should be an external (in general packages.yaml is long)
- Paul: TOSS2 machines identify themselves as TOSS2 vs. just linux
- Greg:
- Should OS be part of the config scope?
- Paul: if I build something against a configuration of mvapich2, does Spack record which external mvapich was built against?
- Recent PRs of interest
- #7412 Concretizing architectures: assigning an arch to the root spec was not propagating [Greg]
- When is 11.3 coming out
- TBD
- LLVM: what build type to use?
- LLVM uses Release
- Many Linux distributions build most packages with RelWithDebInfo and ship debug symbols separately
- Ben: package hash PR 7193
- Ben: https://github.com/spack/spack/pull/7411
- Patric: lmod issue: #7430
- Solved #7299 (Frequent errors on Travis build tests). Travis now sets up a mirror for sources before build tests and caches it in Travis.
- #7314 Better error reporting and faster relocation for binary packages [Patrick]
- #7369 Minor bug fix: makes it possible to leave f77/fc entries undefined in compilers.yaml