Cherry-pick upstream Spack package updates - #82
Conversation
* py-numba: add 0.66.0, py-llvmlite: add 0.48.0 * Add myself as maintainer
NAG's linker forwards GNU linker flags (like --version-script) to GCC without the required -Wl, prefix, causing the link step to fail. Add a patch to scipy's meson.build that excludes nagfor from the version-script code path, matching the existing clang-cl exclusion. Applies to @1.17: %nag.
* py-torch: Fix for Cmake and environment variables * Add link * Committed unneeded file. Oops * Narrow patch * Move to backport patch
Rust currently overwrites CARGO_HOME for dependent builds with a package-stage directory. This prevents users from using a prepopulated Cargo cache when CARGO_NET_OFFLINE=true, since Cargo instead searches the empty stage-local cache. Honor an explicitly configured CARGO_HOME only for offline builds. Retain Spack's isolated per-package Cargo cache for the normal online case.
Co-authored-by: Alec Scott <hi@alecbcs.com>
Co-authored-by: Matthew Mehrtens <12023414+mcmehrtens@users.noreply.github.com>
|
Whoops. Gemini made a PR. I was just thinking of pushing a branch. I've drafted this until I can test! |
* py-urllib3: deprecate versions with CVEs Signed-off-by: Adam J. Stewart <ajstewart426@gmail.com> * ruff Signed-off-by: Adam J. Stewart <ajstewart426@gmail.com> * Bump spack commit to allow multiple setuptools-scm versions --------- Signed-off-by: Adam J. Stewart <ajstewart426@gmail.com> Co-authored-by: Alec Scott <hi@alecbcs.com> Co-authored-by: Alec Scott <alec@llnl.gov>
…'t have setup.py)
…l-oneapi-mkl path fix
…fix intel-oneapi-mkl spec lookup
|
@rickgrubin-noaa I've added your changes from #70 into here. I think. |
|
Question for PR #70 - the PR still says WIP. Are these changes tested and ready to go in? @rickgrubin-noaa |
Co-authored-by: Dom Heinzeller <dom.heinzeller@icloud.com>
Okay. As a test I decided to try: in one of my envs. And it failed on the install of @rickgrubin-noaa can you test to make sure it works for you. |
Wow, not sure how I screwed up that one. Works fine with the sha256 hash confirmed here. |
Well if it's like the manual hashes I've done before, it's because I downloaded a zip file version or .xz or something. That's always a fun day's worth of "Why did the hash change?!" 😄 |
|
@climbfuji et al. I'm doing my final sweep of this now. I'm currently adding FMS 2026.01.01 for my GEOS library builds. If that works, I think I might be done with the package updates. of course, lots of the FMS 2026.01.01 stuff then goes into spack-stack in the yamls there. But I want to make sure it builds for me and I can use it with GEOS. |
|
Uh oh. I'm having issues building 2026.01.01 and my fear is we might need a similar CMake patch for disabling the unit tests that @rem1776 had for 2025.04 and 2026.01. @AlexanderHrabski-NOAA were you able to build 2026.01.01 in your testing? |
@mathomp4 I recall building 2026.01.01 during my testing of the linking patch, but that testing occurred before the CMake patch for disabling cmake testing. During that session, I had my AI code assistant fix a couple different cmake issues in a couple different FMS versions so that I could test my patch with a single MRE. But I was so focused on verifying my own patch that I don't recall if anything was broken in |
@AlexanderHrabski-NOAA Okay. GPT whipped up a patch for 2026.01.01 pretty simply. I'm now working my way to the build. Of course, it's like package 190 out of 200 in this build so it'll be a bit. But if it works, I'll update this and make a PR to spack-packages. Thankfully, I think 2026.02 when it's released has the fix in code. :) |
|
Okay. I made a PR for spack-packages here: and pushed the same change here. I'm almost there! Testing found a couple more small issues I'm working out but I'm so close! |
5 participants
This PR cherry-picks upstream
spack/spack-packagespackage updates, essential fork fixes, and new geospatial packages (pyresamplestack) intospack-stack-dev:Upstream PR Cherry-Picks & Updates:
mimallocvariant)py-awscrt@0.35.0,awscli-v2@2.35.16, autocomplete patch)%gcc@15:)-std=gnu17when%gcc@15:)py-numba@0.66.0,py-llvmlite@0.48.0)CARGO_HOMEfor offline dependent builds)New Packages (
pyresamplestack from PR #70):Fork-Specific Fixes:
nag_linux_threads.patch,nag_hl_fortran_mpi_module.patch).use_stdc_no_threads.patchto@:1.8since newer SciPy versions use Meson instead ofsetup.py.INTEL_MKL_DIRpath when usingintel-oneapi-mkl.gettextandncursesinconfigure_argsbefore accessingprefix/libs.Note: PRs spack#5627 (
bufr), spack#5628 (ecflow), spack#5537 (py-xarray), and spack#5631 (uwtools) were already merged intospack-stack-devin previous PRs.Please merge (do not squash merge) so that individual commit histories are preserved.