Build CI

[CodeGat]

Closes #30

Background

This is a simple matrix implementation of build-ci using access3 and non-access3 manifests.

The PR

• Add access3, no-access3 manifests
• Add matrix-compatible build-ci

[anton-seaice]

@CodeGat

I think the OpenMPI build is incompatible:

/tmp/root/spack-stage/spack-stage-access-cice-git.e0499721dbd5ee653
            dfd8b0facd1d67a9772be39_CICE6.6.0-3-git.5-ngbf4g4afikghnxe7nx5bckag
            lz52k3f/spack-src/cicecore/cicedyn/infrastructure/comm/mpi/ice_exit
            .F90(18): error #7013: This module file was not generated by any re
            lease of this compiler.   [MPI]
     197          use mpi   ! MPI Fortran module
     198    ----------^

This is the concretisation of the failed build:

==> [2025-07-29-04:59:54.724499] Concretized access-cice@git.e0499721dbd5ee653dfd8b0facd1d67a9772be39=CICE6.6.0-3-git.5%oneapi@2025.2.0~access3 io_type=Binary
 -   ngbf4g4  access-cice@git.e0499721dbd5ee653dfd8b0facd1d67a9772be39=CICE6.6.0-3-git.5%oneapi@2025.2.0~access3~ipo~openmp build_system=cmake build_type=Release generator=make io_type=Binary arch=linux-rocky8-x86_64_v3
[^]  bbw5qer      ^cmake@3.24.4%gcc@13.2.0~doc+ncurses+ownlibs build_system=generic build_type=Release arch=linux-rocky8-x86_64
[^]  flywnq4          ^curl@8.7.1%gcc@13.2.0~gssapi~ldap~libidn2~librtmp~libssh~libssh2+nghttp2 build_system=autotools libs=shared,static tls=openssl arch=linux-rocky8-x86_64
[^]  isewm7e              ^nghttp2@1.57.0%gcc@13.2.0 build_system=autotools arch=linux-rocky8-x86_64
[^]  pmlwvz3                  ^diffutils@3.10%gcc@8.5.0 build_system=autotools arch=linux-rocky8-x86_64
[^]  y3ijosh              ^openssl@3.3.0%gcc@13.2.0~docs+shared build_system=generic certs=mozilla arch=linux-rocky8-x86_64
[^]  yde2kfm                  ^ca-certificates-mozilla@2023-05-30%gcc@13.2.0 build_system=generic arch=linux-rocky8-x86_64
[^]  l7iuzcv          ^gcc-runtime@13.2.0%gcc@13.2.0 build_system=generic arch=linux-rocky8-x86_64
[^]  poubuuq          ^ncurses@6.5%gcc@8.5.0~symlinks+termlib abi=none build_system=autotools patches=7a351bc arch=linux-rocky8-x86_64
[^]  bucswa2          ^zlib-ng@2.1.6%gcc@8.5.0+compat+new_strategies+opt+pic+shared build_system=autotools arch=linux-rocky8-x86_64
[e]  uuaslxo      ^glibc@2.28%gcc@13.2.0 build_system=autotools arch=linux-rocky8-x86_64
[^]  nql6rht      ^gmake@4.4.1%gcc@8.5.0~guile build_system=generic arch=linux-rocky8-x86_64
[^]  xvjgomg          ^gcc-runtime@8.5.0%gcc@8.5.0 build_system=generic arch=linux-rocky8-x86_64
[e]  6bfinv7          ^glibc@2.28%gcc@8.5.0 build_system=autotools arch=linux-rocky8-x86_64
 -   mftp7n7      ^intel-oneapi-runtime@2025.2.0%oneapi@2025.2.0 build_system=generic arch=linux-rocky8-x86_64_v3
[^]  mglfhd4      ^openmpi@5.0.5%gcc@13.2.0~atomics~cuda~gpfs~internal-hwloc~internal-libevent~internal-pmix~java~legacylaunchers~lustre~memchecker~openshmem~orterunprefix~romio+rsh~static+vt+wrapper-rpath build_system=autotools fabrics=none romio-filesystem=none schedulers=none arch=linux-rocky8-x86_64

I guess set ^openmpi%oneapi ?

[CodeGat]

Sorry, which was that for @anton-seaice? access or no-access? Or both?

[anton-seaice]

This is for no-access, access will fail until we update the linking in the access3-share cmake

[CodeGat]

Might add the compiler definitions to a data file once this is done. Also once this completes successfully, it should push a lot of these packages to the binary cache, so this should be as slow as it gets

[CodeGat]

Err, at least the %gcc one worked, @anton-seaice 😬 Would you know what those errors mean?

[anton-seaice]

We need to add the mpi linking to the access3-share package - that's why the access3 ones are failing

Why the standalone builds are failing is a mystery! That code is a bit of c code mixed in with the fortran, maybe that's related somehow?

[CodeGat]

https://github.com/ACCESS-NRI/CICE/actions/runs/16612429581/job/46998060356?pr=31#step:13:1682

😌 this is what we wanna see

[anton-seaice]

Ok - ill park this until i update access3-share! Thanks for the progress so far @CodeGat :)

[CodeGat]

No worries @anton-seaice, give me a shout if you need anything

[aidanheerdegen]

It would be good to use template variables for compiler versions to make it easier to keep them all sync. My preference would be to use the ones defined in the build-ci repo so they're consistent across the org.

Same for the others below. May as well use the target variable too.

[CodeGat]

Done in 672529f

[aidanheerdegen]

@anton-seaice I suggest splitting out the bits that are currently crashing, merge the stuff that is working, and represents the current state, and make an issue and PR to fix the failing CI build.

[CodeGat]

I've excluded the two failing manifests from the matrix build for now, see 0caddea @aidanheerdegen, @anton-seaice

[aidanheerdegen]

I have a question about the CMake build system changes.

Ok to merge if those are required, otherwise we should drop them for this PR.

[anton-seaice]

I have a question about the CMake build system changes.

Ok to merge if those are required, otherwise we should drop them for this PR.

Adding REQUIRED COMPONENTS Fortran and target_link_libraries(cicelib PUBLIC MPI::MPI_Fortran) is not require on Gadi, but is required on the github runners for anything to build. We are confident this is fine to add.

Moving target_compile_definitions(cicelib PRIVATE FORTRANUNDERSCORE ncdf) outside of the if statement is possibly becuase we never build with io_type = Binary on gadi ?

[anton-seaice]

Happy to merge @CodeGat - ill resolve the remaining issues with the access3-share workflow soon !

[CodeGat]

Cool. I was wondering, did you want to pin the version of access3-share rather than the branch reference main so it doesn't reuse a potentially old version from the build-cache where you expect it to be new?

[anton-seaice]

main (as a spack version of the access3-share package) is only something i was messing with in a branch of spack-packages - i don't think it would end up in main of spack-packages :)

[CodeGat]

Thanks for your help with this one, @anton-seaice