Install graph_longrange from PyPI

mace/modules/extensions.py

@@ -690,8 +690,8 @@ def __init__(
     ):
         if not GRAPH_LONGRANGE_AVAILABLE:
             raise ImportError(
-                "Cannot import 'graph_longrange'. Please install graph_electrostatics "
-                "from https://github.com/WillBaldwin0/graph_electrostatics."
+                "Cannot import 'graph_longrange'. Install it with "
+                "`pip install mace-torch[polar]`."
             )
         try:
             hidden_irreps: o3.Irreps = kwargs["hidden_irreps"]
@@ -991,8 +991,8 @@ def forward(
     ) -> Dict[str, Optional[torch.Tensor]]:
         if not GRAPH_LONGRANGE_AVAILABLE:
             raise ImportError(
-                "Cannot import 'graph_longrange'. Please install graph_electrostatics "
-                "from https://github.com/WillBaldwin0/graph_electrostatics."
+                "Cannot import 'graph_longrange'. Install it with "
+                "`pip install mace-torch[polar]`."
             )
         ctx = prepare_graph(
             data,

pyproject.toml

@@ -24,7 +24,7 @@ markers = [
     "network: downloads from the internet (foundation models); opt-in via MACE_CI_ALLOW_NETWORK=1",
     "cueq: requires cuEquivariance (its GPU kernels are NVIDIA-only)",
     "oeq: requires OpenEquivariance (JIT kernels for CUDA or HIP)",
-    "polar: requires graph_electrostatics (import name graph_longrange)",
+    "polar: requires graph_longrange",
     "les: requires the LES library",
     "magnetic: requires sphericart-torch (import name sphericart)",
     "torchsim: requires torch-sim-atomistic (python >= 3.12)",

requirements/polar.txt

@@ -1,4 +1,5 @@
-# External dependency of the polar extension (import name: graph_longrange).
-# Pinned by commit: an unpinned git dependency can break CI on its own.
-# To bump: update the SHA and run the polar jobs (ci-extensions.yaml).
-git+https://github.com/WillBaldwin0/graph_electrostatics.git@6a86de5e3ed35fd86a55fc046aa085fe48a72764
+# External dependency of the polar extension, from PyPI.
+# Pinned exactly so a new release cannot break CI on its own; setup.cfg's
+# `polar` extra carries the range users get.
+# To bump: update the version here and run the polar jobs (ci-extensions.yaml).
+graph_longrange==0.4.4

setup.cfg

@@ -77,6 +77,7 @@ dev =
     pytest-xdist
     pylint==4.0.6
 schedulefree = schedulefree
+polar = graph_longrange>=0.4.4,<0.5
 magnetic =
     sphericart-torch>=2.0
     torch-geometric

tests/golden/feature_inventory.md

@@ -831,7 +831,7 @@ only the bag spells.
 | `calc.export.mace_omol` | `mace_omol` | `mace/calculators/__init__.py:12` | KEEP — a recent, large, published multi-head model; converts with heads intact | `tests/foundations/test_foundations.py::test_mace_omol_elements_subset_reproduces_energy_forces` |
 | `calc.export.mace_anicc` | `mace_anicc` | `mace/calculators/__init__.py:12` | DROP — a 2023 organic-chemistry model superseded by MACE-OFF, and the only loader with a divergent signature (`model_path` instead of `model`): an API exception for an obsolete artifact. Its tracked checkpoint `mace/calculators/foundations_models/ani500k_large_CC.model` goes with it; the release notes say "use MACE-OFF" | — |
 
-## 10. Optional-dependency extras (12)
+## 10. Optional-dependency extras (13)
 
 From `setup.cfg` `[options.extras_require]`. Two facts here feed design decisions rather than
 packaging: `magnetic` pins **`sphericart-torch`**, a shipped dependency on a non-e3nn
@@ -846,6 +846,7 @@ copy — so the tree depends on both at once (§19).
 | `extra.schedulefree` | `[schedulefree]` | `setup.cfg` | KEEP | `tests/extensions/schedulefree` |
 | `extra.torchsim` | `[torchsim]` | `setup.cfg` | KEEP — a first-class deployment path, not a secondary integration; the coupling to torch-sim's still-moving API becomes MACE's problem, so the version is pinned in `requirements/` | `tests/extensions/torchsim` |
 | `extra.magnetic` | `[magnetic]` | `setup.cfg` | KEEP — `sphericart-torch` + `torch-geometric` | `tests/extensions/magnetic` |
+| `extra.polar` | `[polar]` | `setup.cfg` | KEEP | `tests/extensions/polar` |
 | `extra.cueq` | `[cueq]` | `setup.cfg` | KEEP — the backend extra naming is worth revisiting | `tests/golden/test_backend_parity_golden.py::test_the_audits_verdict_tracks_whether_the_fused_ops_are_installed` |
 | `extra.cueq-cuda-12` | `[cueq-cuda-12]` | `setup.cfg` | KEEP — idem; the ops major must match `torch.version.cuda`, not the newest available, and a CUDA 11 host wants this one | `tests/unit/test_ci_gates.py::test_the_job_reads_the_extras_out_of_setup_cfg` (the `cueq-wheel-extras` PR job resolves each one; this keeps setup.cfg and that job in step) |
 | `extra.cueq-cuda-13` | `[cueq-cuda-13]` | `setup.cfg` | KEEP — idem; cu13 ops start at cuequivariance 0.7.0 | `tests/unit/test_ci_gates.py::test_the_job_reads_the_extras_out_of_setup_cfg` (the `cueq-wheel-extras` PR job resolves each one; this keeps setup.cfg and that job in step) |
@@ -1223,7 +1224,7 @@ file on every run.
 | 7. Registries | 21 | registry key |
 | 8. Loss classes | 10 | class name |
 | 9. Calculator constructor + exports | 27 + 9 | parameter name, `__all__` |
-| 10. Optional extras | 12 | extra name |
+| 10. Optional extras | 13 | extra name |
 | 11. Model output keys | 43 | dict key |
 | 12. Calculator + eval output keys | 31 + 13 | results key, written key |
 | 13. Environment variables | 9 + 3 | variable name |