forked from spack/spack
-
Notifications
You must be signed in to change notification settings - Fork 0
Telcon: 2015 05 28
Todd Gamblin edited this page May 28, 2015
·
7 revisions
- Todd Gamblin (LLNL)
- Greg Lee (LLNL)
- Chris Chambreau (LLNL)
- Jim Galarowicz (Krell)
- Matt Legendre (LLNL)
- Greg Lee deploying latest Python at LLNL using Spack
- Matt L. and Todd are still thinking about better ways to manage/expose packages to users:
-
Many different scenarios where Spack needs a concept of a "package set":
- Set of packages to build ("build profile", cf. hashdist)
- Set of packages to use to concretize newly installed packages (concretization "context")
- Set of packages for an application environment (single prefix with packages used by app developers)
- Set of packages merged into a single prefix (i.e. generalize current Python support)?
- Set of packages in modules? Users load single module for all of them?
- Should a set of packages allow multiple of some dependencies?
- Single environment could have two exe's that depend on different versions of a dep.
- Should be able to coexist in a single environment unless the dep must ALSO be visible in the environment.
-
Not clear how the user should express all of the above scenarios
-
May want to postpone work on views until after 0.9, to get 0.9 out the door.
- Conclusion: Should probably only do first two bullets here for 0.9.
- Todd is almost done with output redirection
- Spack will, by default, not display build output, and it will be logged
- Output will be logged in a file in the build directory (
spack-build-output.out
) - This will be copied to the
$prefix/.spack
directory if the build is successful.
- Output will be logged in a file in the build directory (
- You'll be able to run
spack -v
to get the build output.
- Bugfix: #34 fixed (regression introduced by #30)
- Spack will now fail fast when dependences fail to build
- Issue only affected develop
- Could cause incorrect builds the way it was written -- package could build before a dep.
- Considering adding a similar
--best-effort
option to install, where DAG order isn't violated.
- Other items?
- Saravan Pantham (UC Davis) sent a pull request with new FastMATH packages
- TBD