[HaoZeke]

To see that 2026.06 has no CPU build of GROMACS, on any host with software.eessi.io mounted:

source /cvmfs/software.eessi.io/versions/2026.06/init/bash
module avail GROMACS
export EESSI_USER_INSTALL=$(mktemp -d)
module load EESSI-extend/2026.06-easybuild
git clone https://github.com/EESSI/software-layer.git && cd software-layer
git apply <<'EOF'
--- a/easystacks/software.eessi.io/2026.06/eessi-2026.06-eb-5.4.1-2026.1.yml
+++ b/easystacks/software.eessi.io/2026.06/eessi-2026.06-eb-5.4.1-2026.1.yml
@@ -15,3 +15,7 @@ easyconfigs:
       options:
         # see https://github.com/easybuilders/easybuild-easyconfigs/pull/27186
         from-commit: a25f8578df309795c8c9734e0a435686001248a0
+  - GROMACS-2026.4-foss-2026.1.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/27122
+        from-commit: aac42b62d957f2a521309bdb14d01de1b3431946
EOF
eb --easystack easystacks/software.eessi.io/2026.06/eessi-2026.06-eb-5.4.1-2026.1.yml --robot --dry-run

module avail GROMACS finds nothing.
The dry run lists GROMACS/2026.4-foss-2026.1 as [ ] and all 94 of its dependencies as [x] (Java/25.0.5 comes from the GATK entry already on main):

 * [ ] .../easybuild/easyconfigs/j/Java/Java-25.0.5.eb (module: Java/25.0.5)
 * [ ] .../files_commit_aac42b62d957f2a521309bdb14d01de1b3431946/GROMACS-2026.4-foss-2026.1.eb (module: GROMACS/2026.4-foss-2026.1)

This PR adds the CPU build, GROMACS 2026.4 on foss/2026.1 (LGPL-2.1, https://www.gromacs.org).
The recipe is not in a release yet, so the entry pins from-commit to the head of the easyconfigs PR:

• {bio}[foss/2026.1] GROMACS v2026.4 easybuilders/easybuild-easyconfigs#27122

This stays a draft until that PR is merged, and then the from-commit pin goes.
#1658 adds only the CUDA 13.3.0 build of 2026.3 under accel/nvidia, and no other open PR touches GROMACS.
#1656 already deployed mpi4py and networkx, so the recipe needs nothing new besides GROMACS itself.
GROMACS 2026.4 is the latest upstream release.

One double-precision test is flaky, because PositionRestraintCommTest.PositionRestraintsTwoCOMs sets no ld-seed.
On the double MPI build, about 5 in 1000 runs of that test fail.
If one target fails there, a rebuild of that target should pass.
A one-line test patch that fixes the seed is ready for PR 27122; once it lands, I'll move from-commit to it.

AI disclosure, per the EasyBuild AI policy: Cursor, routing between Claude Opus 5.5, GPT-5.6 Sol and Grok 4.7, was used to find the minimal set of missing recipes, check it against open PRs and repeat the flaky test.
The recipe has its own disclosure on 27122.
The dry run above is mine, I reviewed the easystack change and the results myself, and I leave bot: build and bot:deploy to maintainers.