-
Notifications
You must be signed in to change notification settings - Fork 151
Conference call notes 20260729
Kenneth Hoste edited this page Jul 29, 2026
·
1 revision
(back to Conference calls)
Notes on the 298th EasyBuild conference call, Wednesday 29 July 2026 (15:00 UTC / 17:00 CEST)
List of attendees (12):
- Simon Branford (Univ. of Birmingham, UK)
- Jasper Grimm (University of York, UK)
- Alexander Grund (ZIH, Dresden, Germany)
- Kenneth Hoste (HPC-UGent)
- Georgios Kafanas (University of Luxembourg)
- Kurt Lust (University of Antwerp, LUMI)
- Hugo Meiland (Microsoft HPC)
- Vladislav Mogilev (Univ. of Birmingham, UK)
- Sam Moors (Vrije Universiteit Brussel, Belgium)
- Sébastien Moretti (SIB, Switzerland)
- Mikael Öhman (Chalmers University of Technology, Sweden)
- Jurij Pečar (EMBL, Germany)
- Jörg Saßmannshausen (Imperial College London, UK)
- overview of recent developments
- Q&A
- latest EasyBuild release: EasyBuild v5.3.1 (19 June 2026)
- next (stable) EasyBuild release: in next 2 weeks...
- ETA for EasyBuild 6.0: spring 2027
- milestones: framework
- ...
- future easyconfigs merge sprints planned:
- Mon 10 Aug 2026
- Mon 19 Oct 2026
- Mon 21 Dec 2026
- Mon 15 Feb 2027
- aiming for semi-fixed schedule every other month: 3rd Monday in even months (with exceptions)
- at some point a sprint to revise & close outdated open issues across different EasyBuild repos
(changed made in PRs marked with * are included latest EasyBuild stable release)
-
blog/docs (merged PRs)
- ...
-
framework (merged PRs)
-
easyblocks (merged PRs)
-
easyconfigs (merged PRs)
-
blog/docs (open PRs + issues)
- ...
-
framework (open PRs + issues)
-
bugs
- Don't check for
libandbinpaths in sanity check step for each extension (PR #5221)
- Don't check for
-
enhancements
- format extension names in module files with
exts_formatterfunction (PR #5186) - reduce number of module load commands used to set up build environment (WIP PR #5211)
- will result in significant speedup for easyconfigs that have a lot of extensions...
- also bwrap module installpath (PR #5214)
- format extension names in module files with
-
code cleanup
- ...
-
changes
- deprecate passing
extensionoption tosanity_check_step&sanity_check_load_modulemethods inEasyBlockclass (PR #5220)
- deprecate passing
-
CI/test suite
- ...
-
bugs
-
easyblocks (open PRs + issues)
- bug fixes/reports
- enhancements
- Add PTX code for (highest) CUDA compute capability in PyTorch & torchvision (PR #4129)
- Rework of the
JuliaPackageeasyblock (PR #4146) - enhance custom easyblock for OpenFOAM to addd support for new intel compilers (icx) (PR #4177)
- Add ignore patterns for tests on RISC-V to custom easyblock for LLVM (PR #4178)
- Update
cuda_ccflags for newer ELPA versions (PR #4186) - Add
.pyifiles for easyblocks to support LSPs (PR #4183)
-
updates
- refactor Score-P easyblock and prepare for v10.0 (PR #4133)
-
changes
- Do not pass extension to sanity_check_load_module (PR #4184)
- companion to framework PR #5220
- adapt
NvidiaBasegeneric easyblock for removal of test examples in hpcx v2.50 (PR #4174)
- Do not pass extension to sanity_check_load_module (PR #4184)
-
code cleanup
- ...
-
new easyblocks
- ...
-
CI/test suite
- Add test for correct import of other easyblocks (PR #4182)
-
easyconfigs (open PRs + issues)
-
over 770 open easyconfig PRs currently
- up from ~680 mid June'26
- easyconfigs merge sprint coming soon: Mon 10 Aug'26
-
bug fixes/reports
- ...
-
enhancements
- ...
-
(noteworthy) new software
- ...
- software updates
- ...
-
changes
- ...
-
CI/test suite
- improve analyzer for tagbot to avoid trouble with long diffs (PR #26544)
-
over 770 open easyconfig PRs currently
-
PythonPackagevsPythonBundlediscussion (cfr. Slack thread started by Alexander)- came up in discussion with Davide recently
- "decision" that single Python packages should use
PythonBundle(with a single extension)- Alexander is not a fan of this, style issue with being forced to use
exts_list - also can only use a single source file for extension, which is hard to fix in framework without breaking something
- Alexander is not a fan of this, style issue with being forced to use
- some Python package gets included as extension in a bundle because it's required, but not actually related to the Python package that requires it (example:
meson-python)- stuff like
meson-pythonshould be in a separate easyconfig - but we can only "detect" this post-hoc
- stuff like
- not actually decided (not enforced by CI)
- Sam proposed this, mostly so the Python packages pop up in list of extension (via
modulecommand) - use of
PythonBundlestill makes sense for large things likePython-bundle-PyPIorPyTorch(many related Python pkgs) - hard to come up with a clear policy that we can enforce through CI?
- new check for
pip listcan't be fully controlled viaPythonPackage(it can be for extensions)- new
python_pkg_name(orextension_name) easyconfig parameter could be introduced?-
extension_namecould imply a "ghost" extension (single-itemexts_list) - current common
modulenameis very confusing, should be fixed (moduleshould be reserved for environment modules)
-
- new
- CI should not allow single-item
exts_list+PythonBundle- and to not allow Python pkgs being installed as extension that are only a build dependency (like
meson-python)- could use metadata in
pyproject.tomlthat specifies whether it's a runtime dependency or not...?
- could use metadata in
- and to not allow Python pkgs being installed as extension that are only a build dependency (like
- Hugo is working on RISC-V and AMD ROCm related stuff
- how to get (quicker) feedback to PRs?
- raise question in Slack, talk to people directly who are known to be interested (e.g. Caspar for ROCm, Bob for RISC-V)
- status update on new bot for testing contributions
- see it in action @ https://github.com/easybuilders/easybuild-easyconfigs/pull/26309#issuecomment-4750580470
- new bot is event-based (based on EESSI bot), so faster reponse to request for testing
- idea is to eventually test in multiple environments (same as currently in
jsc-zen3, EESSI, ...) - not fully on par yet with old bot (for example, no way to specify extra options to be passed to
eb)
- (Mikael)
pkg-configis blocked to pick up OS dependencies, but we still rely on OS pkgs sometimes-
pkgconf --libs libinputis broken, becauselibudev.pcin system is not found...- currently trying to get rid of those OS deps, like using
libudev-zeroinstead - we could make this opt-in somehow, with some kind of allowlist?
- currently trying to get rid of those OS deps, like using
- Meson always uses
pkgconfwhen looking for stuff, not easy to patch out
-
- (Jörg) problem with
gWidgets2tcltkR package, which requires OS package providing fonts to be installed (xvfb-runis not enough) - question w.r.t. PETSc
- Firedrake uses two PETSc versions per year
- ABI is changing between PETSc versions?
- but that shouldn't directly dictate which PETSc version EasyBuild uses?
- would statically linking to a PETSc (build) dependency make sense?
- there's some interest from Firedrake in EESSI
- question was raised why we don't use pre-built wheels for
FEniCS-UFL- generally speaking, we install from source, because that gives us more control
- Python version being used, performance optimizations for non-Python code, etc.
- in the case of
FEniCS-UFLspecifically, we could use the available (genericpy3) wheel (from PyPI), but there's no significant downside for us in not doing so
- generally speaking, we install from source, because that gives us more control
- Firedrake uses two PETSc versions per year
- Jurij: interest in latest R bundles (Bioconductor)
- merged earlier today, see PR #26416
- when would this be available in EESSI?
- will get installed in new EESSI version (2026.06), compat layer for that is in place, but installing stuff in software layer is yet to be started (is currently being prepared)