-
Notifications
You must be signed in to change notification settings - Fork 144
Conference call notes 20160831
Kenneth Hoste edited this page Aug 31, 2016
·
3 revisions
(back to Conference calls)
Notes on the 58th EasyBuild conference call, Wednesday August 31st 2016 (5pm - 6pm CET)
Alphabetical list of attendees (5):
- Damian Alvarez (JSC, Germany)
- Pablo Escobar (sciCORE/UniBas, Switzerland)
- Adam Huffman (Francis Crick Institute, UK)
- Kenneth Hoste (HPC-UGent)
- Alan O'Cais (JSC, Germany)
- discussion of proposal for deprecating old/inactive toolchains
- X11 stack as a single module
- Python without Tk/X11 dep
- Q&A
- discussion of proposal for deprecating old/inactive toolchains
- which toolchains/versions + motivation
- procedure: archiving easyconfigs + clear warning when archived easyconfigs are being used
- Pablo: archived easyconfigs should still show up in output of
eb --search
(in a separate block?)
- via 'components': https://github.com/hpcugent/easybuild-easyconfigs/pull/3440
- via
jhbuild
: https://github.com/hpcugent/easybuild-easyconfigs/pull/3463 - pros & cons for both approaches?
-
jhbuild
takes a single datestamp as 'version', and pulls in straights from repos rather than using releases -
jhbuild
approach results in a bigger set of packages being installed (packages can be filtered out) -
jhbuild
approach may be easier to maintain since deps are taken of -
jhbuild
approach also includes Mesa & more... - Alan: open issues with
jhbuild
approach:-
jhbuild
approach requires two 'sources', but these a not versioned (e.g.xorg.modules
) - problem with
Mako
Python modules required for Mesa not being found at build time, even though it's found at configure time...- can be worked around by including Python with Mako baked in as (build?) dep
-
- Damian: dedicated (generic) easyblock for
jhbuild
approach- can also be used for BigDFT
- current proposal:
- go through with bundle-of-components approach for now
- evaluate how hard that is to maintain in the future
- also support
jhbuild
approach as drop-in alternative - switch to
jhbuild
approach if bundle-of-components is too much of a hassle w.r.t. maintaining it
- proposal by Ward to drop Tk from list of dependencies for Python by default
- since it requires the whole X11 stack underneath...
- cfr. other projects:
- Homebrew: https://github.com/Homebrew/legacy-homebrew/issues/14989
- Fedora:
tkinter
is split off in a separate package
- how much impact would this have, due to lack of
Tkinter
support in Python?- is providing a separate easyconfig for Python with Tk sufficient, or do we need something better?
- would just require adding a separate dependency on
Tkinter
next to Python
- Damian: how to contribute back easyconfigs that were created with
--minimal-toolchains
enabled?- currently, tests would fail to resolve dependencies because of lack of
--minimal-toolchains
- enabling
--minimal-toolchains
in tests already is worth considering - plan is to enable
--minimal-toolchains
by default in EasyBuild v3.0.0 - ... but people that disable it may suffer because dependency resolution may not work anymore when only trying to resolve dependencies with the 'parent' toolchain
- Alan: big fat warning if robot fails to resolve deps when
--minimal-toolchains
is disabled, with pointer to documentation?
- currently, tests would fail to resolve dependencies because of lack of