From 1a8893fc457e265ec86e3f4abc7547dc2c0d3aca Mon Sep 17 00:00:00 2001
From: Rudolf Weeber <weeber@icp.uni-stuttgart.de>
Date: Sat, 13 Jun 2026 17:11:11 +0200
Subject: [PATCH] magnetostatics: guard DipolarDirectSum gpu=True without CUDA
 (bug-sweep #31)

DipolarDirectSum::do_construct forwarded the gpu flag straight into the
core actor (m_is_gpu=true) with no feature check, unlike the analogous
CoulombP3M path. On a CUDA-off build this constructed and activated
silently; the first force/energy call then either hit assert(not
m_is_gpu) (debug) or silently computed on CPU while is_gpu() reported
True (release) -- a capability misreport.

Mirror the CoulombP3M convention: at the script-interface seam, if
gpu=True, add 'CUDA' to required_features and call
CodeInfo::check_features inside parallel_try_catch, so the unsupported
request surfaces as a clean RuntimeError ("Missing features CUDA")
before the core actor is created.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 .../magnetostatics/DipolarDirectSum.hpp             | 13 ++++++++++---
 testsuite/python/dipolar_interface.py               |  6 ++++++
 2 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/src/script_interface/magnetostatics/DipolarDirectSum.hpp b/src/script_interface/magnetostatics/DipolarDirectSum.hpp
index 2a2c409880..df37087d11 100644
--- a/src/script_interface/magnetostatics/DipolarDirectSum.hpp
+++ b/src/script_interface/magnetostatics/DipolarDirectSum.hpp
@@ -27,10 +27,12 @@
 
 #include "core/magnetostatics/dipolar_direct_sum.hpp"
 
+#include "script_interface/code_info/CodeInfo.hpp"
 #include "script_interface/get_value.hpp"
 
 #include <memory>
 #include <string>
+#include <vector>
 
 namespace ScriptInterface {
 namespace Dipoles {
@@ -47,11 +49,16 @@ class DipolarDirectSum : public Actor<DipolarDirectSum, ::DipolarDirectSum> {
   }
 
   void do_construct(VariantMap const &params) override {
-    context()->parallel_try_catch([this, &params]() {
+    auto const gpu = get_value_or(params, "gpu", false);
+    context()->parallel_try_catch([this, &params, gpu]() {
+      if (gpu) {
+        std::vector<std::string> required_features;
+        required_features.emplace_back("CUDA");
+        CodeInfo::check_features(required_features);
+      }
       m_actor = std::make_shared<CoreActorClass>(
           get_value<double>(params, "prefactor"),
-          get_value<int>(params, "n_replicas"),
-          get_value_or(params, "gpu", false));
+          get_value<int>(params, "n_replicas"), gpu);
     });
   }
 };
diff --git a/testsuite/python/dipolar_interface.py b/testsuite/python/dipolar_interface.py
index 22cc14e905..8e83f168a0 100644
--- a/testsuite/python/dipolar_interface.py
+++ b/testsuite/python/dipolar_interface.py
@@ -90,6 +90,12 @@ def test_exceptions_non_p3m(self):
                 MDLC(gap_size=2., maxPWerror=0.1, actor=ddsr_gpu)
             with self.assertRaisesRegex(ValueError, "Parameter 'n_replicas' must be >= 0"):
                 DDSR(prefactor=1., n_replicas=-2, gpu=True)
+        else:
+            # requesting the GPU implementation without CUDA support compiled
+            # in must raise a clear error at construction time, rather than
+            # silently constructing a CPU actor that reports is_gpu() == True
+            with self.assertRaisesRegex(RuntimeError, "Missing features CUDA"):
+                DDSR(prefactor=1., gpu=True)
         with self.assertRaisesRegex(RuntimeError, "Parameter 'actor' is missing"):
             MDLC(gap_size=2., maxPWerror=0.1)
         with self.assertRaisesRegex(RuntimeError, "Parameter 'n_replica' is not a valid parameter"):