Telcon: 2022 06 29
Peter Scheibel edited this page Jun 29, 2022
·
17 revisions
- Peter Scheibel (host)
- Massimiliano Culpo
- Richarda Butler
- Matt Belhorn (OLCF)
- Greg Becker
- Brian Van Essen
- Todd Gamblin
- Phil Sakievich
- Brian:
unify: false
- why is that the default?- All created environments have
unify: false
explicitly added (withspack create
) - As a side effect, if you
-
true
would make more sense - Next Spack release will change away from
unify: false
- Unify-when-possible might be the safest default in the sense that it avoids concretization errors
- But... it doesn't necessarily produce what users might expect (i.e. if they aren't aware, then they might get different installs of things)
- All created environments have
- The annoyances of
~
configuration came up- Also strategies for combining configurations for user experience
- Phil: CI runs with the config they provide to users
- Brian: externals are a pain that often influence config: can sites define a Cray-manifest analog (i.e. describing already-installed DAGs that are available)
- Peter: at the moment creating a manifest file is time-consuming so IMO would be primarily useful for "slow-changing" packages
- Best-effort co-concretization PR prevents concretizing specs on the command line
- https://github.com/spack/spack/pull/28941
-
spack spec
works, but notspack solve
- This specific behavior can be reverted without causing problems elsewhere: Todd will address that
- Massimiliano: 0.18 had a performance regression that will be fixed in 0.18.1 and also if you pull the latest
develop
branch - Brian: it would be nice if labs could share their individual Spack forks
- Then if someone from one lab had a problem they could check if another lab had already addressed it
- Phil: generally can make use of package inheritance to minimize differences
- Peter: this works unless you need to remove constraints
- There are efforts to address this
- Peter: this works unless you need to remove constraints
- (Massimiliano) Continue from https://github.com/spack/spack/wiki/Telcon%3A-2022-06-08: multiple build systems used by one package
- (Matt) if we build an environment, and one package fails, then module files are not generated
- Want
spack module refresh
but only to generate module files that haven't been generated yet
- Want
- Updating compiler wrapper to strip
-Werror
(optionally): https://github.com/spack/spack/pull/30284 - (Wileam) Adding fpm as a build system
- Issue on
fpm
repo: https://github.com/fortran-lang/fpm/issues/79
- Issue on
- Default action for concretization in environments: https://github.com/spack/spack/pull/29942#issuecomment-1097287194
- Continue from https://github.com/spack/spack/wiki/Telcon%3A-2022-04-06:
warning()
directive - (Wileam) Can't bootstrap
llvm@13
and@14
if gold is enabled (which is the default behavior) https://github.com/spack/spack/issues/29350- Pinpointed issue to external package (
binutils
carried by OS) based on comment - PR drafted: https://github.com/spack/spack/pull/30954
- Can externals detection be expanded from executables only to arbitrary files (e.g. header or
.a
/.so
library files)?
- Pinpointed issue to external package (
- How to reorganize the docs?
- Continue from https://github.com/spack/spack/wiki/Telcon%3A-2022-04-13:
not
in spec syntax