Conference call notes 20200708 - easybuilders/easybuild GitHub Wiki
(back to Conference calls)
Notes on the 151st EasyBuild conference call, Wednesday July 8th 2020 (08:00 UTC - 10:00 CEST)
Attendees
Alphabetical list of attendees (7):
- Kenneth Hoste (HPC-UGent, Belgium)
- Jörg Saßmannshausen (NIHR Biomedical Research Centre, UK)
- Simon Branford (Univ. of Birmingham, UK)
- Alexandre Strube (JSC, Germany)
- Adam Huffman (Big Data Institute, Oxford, UK)
- Mikael Öhman (Chalmers University of Technology, Sweden)
- Leo Rojas (University of Sussex, UK)
Agenda
- update on EasyBuild v4.2.2 release
- 2020b update of common toolchains
- Q&A
Update on EasyBuild v4.2.2 release
- EasyBuild v4.2.2 will be released today
- project to track issues/PRs that should be tackled for this release: https://github.com/orgs/easybuilders/projects/6
- recent changes
- framework
- always take into account builddependencies when generating template values, also when we're not iterating over builddependencies (PR #3346)
- fixes for
--try-update-deps(still experimental) (PR #3350, PR #3353) - fixed regression in check for available modules, when using ancient modules tool version (PR #3379)
- fix for use of template values in
exts_list(PR #3377) - retain order of paths in when generating prepend_path statements for module file (don't sort them alphabetically) (PR #3367)
- also put easyblocks used by extensions in
reproddirectory (PR #3375)
- easyblocks
- easyconfigs
- stop relying on Mercurial repositories in Bitbucket (see issue #10868)
- cleanup of
2020abranch: not finished yet, but made some progress- see issue #10753
- see https://github.com/easybuilders/easybuild-easyconfigs/pulls?q=is%3Apr+sort%3Aupdated-desc+is%3Aclosed+milestone%3A4.2.2
- framework
2020b update of common toolchains
- we should start looking into this...
- GCC 10.1 (wait for 10.2)
- OpenMPI 4.1.0 (RC)
- no updates for Intel yet?
Q&A
- Jörg: ICE with GCC 9.3.0 when compiling Salmon (see https://github.com/easybuilders/easybuild-easyconfigs/pull/10631)
- also hitting this with manual
makebuild - doesn't seem to be a memory issue
- check if others can reproduce it
- issue seems to be with LTO, try to disable that?
- requires patching
- could also try lower optimization level (
-O1)
- also hitting this with manual
- Adam: we should discuss creating a
CUDAcore+CUDA(similar toGCCcore+GCC)- see https://github.com/easybuilders/easybuild-easyconfigs/pull/10935
- main motivation is avoiding multiple installations of CUDA like we have now
- Mikael: "dropping" last version part of CUDA version, good idea?
- see https://github.com/easybuilders/easybuild-easyconfigs/pull/10935 (starting point for
fosscuda/2020a) - do we really want to do in-place updates of CUDA?
- CUDAcore 11.0 wrapper should be at Core level
- may get us in trouble later (fosscuda/2020b)
- do we trust NVIDIA to follow semantic versioning so in-place updates are safe?
- Jörg: providing this flexibility clashes with reproducibility...
- could still swap back to previous CUDA 11.0.x versions if needed
- see https://github.com/easybuilders/easybuild-easyconfigs/pull/10935 (starting point for
- Alexandre: similar discussion for NCCL version in TensorFlow