-
Notifications
You must be signed in to change notification settings - Fork 2.2k
Telcon: 2023 03 22
Massimiliano Culpo edited this page Mar 22, 2023
·
6 revisions
To be held Wednesday March 22nd, 9am PT (UTC -7:00)
- Massimiliano Culpo (host)
- Dom Heinzeller
- Mark Krentel
- Jack Morrison
- Wileam Phan
- Phil Reiger
- Jawad Ahmed
This meeting will be Q&A: there are no pre-planned general Spack topics.
-
Mark How can I make it easy for
hpctoolkit
users to build with configuration suggested byhpctoolkit
maintainers?- Mark: Pointed to this PR https://github.com/spack/spack/issues/16909 where a feature is suggested to cover this use case. The core idea is to allow a package to override default values of variants etc. of its dependencies.
- Massimiliano: Expressed concerns about the proposed approach (see https://github.com/spack/spack/issues/16909#issuecomment-1476969126 and following discussion).
- Massimiliano: Proposed to investigate a solution where "suggested" configuration for a package can be stored alongside the package.py, and imported in an environment with e.g. a command. That would be an explicit action done by the user (which would thus be aware of changes in configuration), but considerably simpler than the current workflow.
-
Ahmed Is it expected that concretizing an environment with
unify:when_possible
leads to duplicate packages?- Massimiliano: That situation might happen. The only option that would error if there are duplicate in the environment is
unify:true
. See https://spack.readthedocs.io/en/latest/environments.html#spec-concretization
- Massimiliano: That situation might happen. The only option that would error if there are duplicate in the environment is
-
Ahmed What would be the best way to use python-extensions with an environment?
- Massimiliano: If possible for your use case, the best way would be to use
unify:true
and setup a view. See https://spack.readthedocs.io/en/latest/environments.html#filesystem-views
- Massimiliano: If possible for your use case, the best way would be to use
-
Wileam Pinged for a review of https://github.com/spack/spack/pull/35919 Asked what to do with the build of LLVM in that recipe.
- Massimiliano: Possibly, the cleanest way to deal with that is to modify the
llvm
package too and havegcc
depend on it when theamdgcn
variant is activated.
- Massimiliano: Possibly, the cleanest way to deal with that is to modify the
-
Massimiliano Advertized new
requires
directive coming when https://github.com/spack/spack/pull/36286 is merged- Mark: There might be cases in the future where
one_of
needs to be exposed to therequires
directive (e.g. dyninst can compile with GCC or Clang) - Massimiliano: The mechanism is already there for
packages.yaml
. If we needone_of
it would be easy to extend the directive API to expose the same functionality as in the configuration file.
- Mark: There might be cases in the future where