Telcon: 2022 12 07
Held Wednesday December 7th, 9am PT
- Massimiliano Culpo (host)
- Mark Krentel
- Phil Regier
- Ivan Razumov
- Dom Heinzeller
- Phil Sakievich
- Robert Underwood
(This will be just Q&A - there are no pre-planned general Spack discussion items)
-
Mark There are a few widely used projects which, on a few platforms, fail to build because of time-stamp related issues. How should we tackle this recurring problem?
- Mark: What usually happens is that some
*.texi
file is understood by the configuration as to be regenerated, and the build fail when looking fortexinfo
, since it is not listed as a build dependency. - Related PRs:
- Mark: These packages are release tarball and they shouldn't need any rebuild of docs
- Mark: One solution is to add a build dependency on
makeinfo
, but this brings a lot of dependencies into fundamental packages that don't really need them - Massimiliano: Can we either patch the configuration mechanism, or trick it that all is fine with respect to
makeinfo
? - Mark: It's worth investigating that. Issue is possibly reproducible on Summit.
- Mark: What usually happens is that some
-
Ivan It's difficult to get a sense of what would change when a huge environment is reconcretized after some small modifications
- Ivan: Why don't we add a feature to limit the solver to use only certain packages prescribed by the user ?
- Massimiliano: I think a better solution would be to have some command that lets users inspect differences on speculative concretizations, with respect to a previous state
- Massimiliano: Harmen did some work to print spec differences in a readable way on screen
-
Ivan Sometimes I want to use a commit sha to indicate a version for a local tarball I made from that commit, but Spack understands that as a
GitVersion
instead of a normalVersion
.- Massimiliano: I think Spack understands that as
GitVersion
only if they are 40 chars. If you limit the version tag to something less it will be understood as a normalVersion
.
- Massimiliano: I think Spack understands that as
-
Ivan Why can't we use a version non existing in the
package.py
when we adopt the@<git-ref>=<version>
syntax on the rhs ?- Massimiliano: I think the rhs is used to determine how to build that
git-ref
, so it must be known to Spack - Ivan: The use case is a custom build-system that needs to use different versions for different
git-ref
- Greg (offline): We should be able to use an unknown version, as long as it compares properly to the other versions
- Massimiliano: I think the rhs is used to determine how to build that
-
Dom I noted a few issues with buildcaches on macOS, and submitted a PR to fix them if anybody could review it
- Dom: Issue is https://github.com/spack/spack/issues/34374
- Dom: PR to fix it is https://github.com/spack/spack/issues/34375
-
Dom Proposed to change
--source
to true by default- Phil S., Robert, Mark: Many people might disagree with the additional space required for installation
- Dom: The issue is with CI, and with some additional work that is required to upstream changes from our fork that is tested with
--source
- Dom: An alternative solution that works for us is to make that configurable from file.
-
Robert Is there a way to get information from Spack public mirrors?
- Robert: see issue https://github.com/spack/spack/issues/34233
-
Robert Can we update the images we use in pipelines (CentOS in particular) to get more recent compilers?
- Massimiliano: I think we can. We should probably get in touch with Luke for the e4s pipelines.