@@ -1,20 +1,11 @@
-# eOn full default product (metatomic + xTB + Cap'n Proto serve + rgpot).
-# Feature surface matches conda-forge eon-feedstock.
-
 easyblock = 'MesonNinja'
 
 name = 'eOn'
 version = '2.16.0'
 
 homepage = 'https://theochemuieon.readthedocs.io/'
-description = """
-eOn long-timescale atomistic simulation: adaptive kinetic Monte Carlo (AKMC),
-nudged elastic band (NEB), and saddle-point search (min-mode, dimer).
-
-This is the full default product: C++/Fortran eonclient with metatomic, xTB,
-Cap'n Proto serve, and rgpot, plus the Python AKMC package (import eon). Feature
-surface matches the conda-forge eon feedstock for HPC module stacks.
-"""
+description = """eOn contains algorithms used to model the evolution of
+atomic-scale systems over long time scales."""
 
 toolchain = {'name': 'foss', 'version': '2026.1'}
 toolchainopts = {'pic': True, 'openmp': True}
@@ -77,19 +68,17 @@ dependencies = [
     ('Highway', '1.4.0'),
     ('inih', '62'),
     ('quill', '11.1.0', '', ('GCCcore', '15.2.0')),
-    # no gfbf-2026.1 recipe on develop yet
     ('xtb', '6.7.1', '', ('gfbf', '2024a')),
     ('CapnProto', '1.4.0'),
-    # no foss-2026.1 PyTorch on develop; foss-2024a uses Python 3.12
+    # PyTorch 2.9.1 uses Python 3.12.
     ('PyTorch', '2.9.1', '', ('foss', '2024a')),
     ('metatensor', '0.2.2', '', ('GCCcore', '15.2.0')),
     ('metatensor-torch', '0.10.0'),
     ('metatomic-torch', '0.1.15'),
 ]
 
-# Subshell: cargo cinstall must not leave CWD in the source tree for Meson.
-# Prefer builddep Rust on PATH (edition2024). PyTorch paths use python3.12
-# (foss-2024a layout); do not use %(pyshortver)s for EBROOTPYTORCH.
+# Stage readcon-core before Meson configures eOn. The subshell restores the
+# source directory, and the PyTorch paths use its Python 3.12 layout.
 preconfigopts = 'export PATH=$EBROOTRUST/bin:$PATH && '
 preconfigopts += 'unset RUSTFLAGS CARGO_ENCODED_RUSTFLAGS && '
 preconfigopts += 'export RUSTC_WRAPPER= CARGO_BUILD_RUSTC_WRAPPER= && '
@@ -123,8 +112,7 @@ configopts += "-Dnative_arch=false -Dfast_math=false -Duse_mkl=false"
 postinstallcmds = [
     'mkdir -p %(installdir)s/lib %(installdir)s/include %(installdir)s/lib/pkgconfig',
     'cp -a %(builddir)s/readcon-stage/lib/libreadcon_core* %(installdir)s/lib/',
-    # cargo-built cdylib bypasses the RPATH compiler wrappers; give it a real
-    # RPATH so rpath-mode sanity checks pass.
+    # cargo-c does not use the EasyBuild RPATH compiler wrappers.
     "patchelf --force-rpath --set-rpath '$ORIGIN' %(installdir)s/lib/libreadcon_core.so.0.13.1",
     'cp -a %(builddir)s/readcon-stage/include/readcon-core.h %(installdir)s/include/',
     ("sed 's|^prefix=.*|prefix=%(installdir)s|' %(builddir)s/readcon-stage/lib/pkgconfig/readcon-core.pc > "
 
@@ -1,12 +1,11 @@
-# TorchScript C++ bindings for metatensor (libmetatensor_torch).
 easyblock = 'CMakeNinja'
 
 name = 'metatensor-torch'
 version = '0.10.0'
 
 homepage = 'https://docs.metatensor.org/'
-description = """TorchScript C++ bindings for metatensor (libmetatensor_torch).
-Required by metatomic-torch and eOn metatomic potentials."""
+description = """Metatensor-torch provides TorchScript bindings for
+metatensor."""
 
 toolchain = {'name': 'foss', 'version': '2026.1'}
 toolchainopts = {'pic': True, 'openmp': True}
@@ -23,7 +22,7 @@ builddependencies = [
 
 dependencies = [
     ('Python', '3.14.2'),
-    # foss-2024a PyTorch layout is python3.12 (not this recipe's Python)
+    # PyTorch 2.9.1 uses Python 3.12.
     ('PyTorch', '2.9.1', '', ('foss', '2024a')),
     ('metatensor', '0.2.2', '', ('GCCcore', '15.2.0')),
 ]
 
@@ -1,12 +1,11 @@
-# metatensor-core C library (CXX release tarball).
 easyblock = 'CMakeMake'
 
 name = 'metatensor'
 version = '0.2.2'
 
 homepage = 'https://docs.metatensor.org/'
-description = """Metatensor core C library: sparse storage for atomistic
-machine learning. Built from the metatensor-core CXX release tarball."""
+description = """Metatensor stores and manipulates sparse data for atomistic
+machine learning."""
 
 toolchain = {'name': 'GCCcore', 'version': '15.2.0'}
 toolchainopts = {'pic': True}
 
@@ -1,12 +1,11 @@
-# Metatomic Torch C++ library (libmetatomic_torch).
 easyblock = 'CMakeNinja'
 
 name = 'metatomic-torch'
 version = '0.1.15'
 
 homepage = 'https://docs.metatensor.org/metatomic/'
-description = """Metatomic Torch C++ library for running metatomic models via
-TorchScript. Requires metatensor-torch and PyTorch."""
+description = """Metatomic provides interfaces for atomistic machine-learning
+models."""
 
 toolchain = {'name': 'foss', 'version': '2026.1'}
 toolchainopts = {'pic': True, 'openmp': True}
@@ -23,7 +22,7 @@ builddependencies = [
 
 dependencies = [
     ('Python', '3.14.2'),
-    # foss-2024a PyTorch layout is python3.12 (not this recipe's Python)
+    # PyTorch 2.9.1 uses Python 3.12.
     ('PyTorch', '2.9.1', '', ('foss', '2024a')),
     ('metatensor', '0.2.2', '', ('GCCcore', '15.2.0')),
     ('metatensor-torch', '0.10.0'),
 
@@ -4,11 +4,7 @@ name = 'quill'
 version = '11.1.0'
 
 homepage = 'https://github.com/odygrd/quill'
-description = """Quill is an asynchronous low-latency C++ logging library
-with compile-time format-string checks, minimal application-thread
-overhead, and structured logging support.
-
-Required by eOn's client (eonclient) for logging."""
+description = """Quill is an asynchronous low-latency C++ logging library."""
 
 toolchain = {'name': 'GCCcore', 'version': '15.2.0'}
 toolchainopts = {'pic': True}