Fix MSVC portability: popen, missing include, OpenMP index, UTF-8

- rmsd.cpp: popen/pclose are POSIX-only; guard with _popen/_pclose on
  Windows (matching the existing pattern in orcainterface.cpp). Also
  replace the alternative operator token "and" with "&&" (MSVC needs
  /permissive- or <ciso646> for those; GCC/Clang accept them natively).
- trajectory_writer.cpp: uses std::ostringstream/stringstream without
  including <sstream> -- only worked because GCC's libstdc++ pulls it
  in transitively via other headers; MSVC's STL doesn't.
- d4param_generator.cpp: three #pragma omp for loops used size_t
  counters; MSVC's OpenMP (2.0 semantics) requires signed loop indices,
  unlike GCC/Clang's OpenMP implementations.
- CMakeLists.txt: add /utf-8 for MSVC. The codebase uses non-ASCII
  UTF-8 characters in a few identifiers/comments (e.g. "double
  Δqraw[6];" in xtb_gradient.cpp); without /utf-8, MSVC falls back to
  the system codepage and misreads the multi-byte UTF-8 sequence as
  unrelated characters (observed: "error C3872: 'U+201d' is not
  allowed in an identifier" for what is actually Δ).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: conradhuebler

CMakeLists.txt

@@ -7,6 +7,12 @@ project(curcuma)
 # set this project-wide instead of patching every translation unit.
 if(MSVC)
     add_compile_definitions(_USE_MATH_DEFINES)
+    # Claude Generated 2026 - the codebase uses non-ASCII UTF-8 characters in a few
+    # identifiers/comments (e.g. "double Δqraw[6];" in xtb_gradient.cpp). Without
+    # /utf-8, MSVC falls back to the system codepage to interpret the source file,
+    # misreading the multi-byte UTF-8 sequence as unrelated characters (observed:
+    # "error C3872: 'U+201d' is not allowed in an identifier" for what is actually Δ).
+    add_compile_options(/utf-8)
 endif()
 
 option (USE_XTB

src/capabilities/rmsd.cpp

@@ -1659,7 +1659,11 @@ bool RMSDDriver::MolAlignLib()
     std::string command = m_molalign + m_molalignarg + " " + ref_tmp + "   " + tar_tmp + " 2>&1";
     if (m_verbosity >= 3)
         CurcumaLogger::info_fmt("MolAlign command: {}", command);
+#ifdef _WIN32
+    FileOpen = _popen(command.c_str(), "r");
+#else
     FileOpen = popen(command.c_str(), "r");
+#endif
     bool ok = true;
     bool rndm = false;
     bool error = false;
@@ -1674,10 +1678,14 @@ bool RMSDDriver::MolAlignLib()
             CurcumaLogger::debug(2, fmt::format("MolAlign output: {}", std::string(line).substr(0, std::string(line).find('\n'))));
 #endif
     }
+#ifdef _WIN32
+    _pclose(FileOpen);
+#else
     pclose(FileOpen);
+#endif
 
     std::string aligned_tmp = outputPath("aligned.xyz");
-    if (std::filesystem::exists(aligned_tmp) and !rndm) {
+    if (std::filesystem::exists(aligned_tmp) && !rndm) {
         CurcumaLogger::citation("molalign");
         FileIterator file(aligned_tmp, true);
         m_reference_centered = file.Next();

src/core/energy_calculators/dispersion/d4param_generator.cpp

@@ -434,9 +434,11 @@ void D4ParameterGenerator::GenerateParameters(const std::vector<int>& atoms, con
 
         // NOTE: Cannot use collapse(2) with triangular loops (j = i + 1)
         // Parallelize outer loop only, still provides good speedup
+        // Claude Generated 2026 - MSVC's OpenMP (2.0 semantics) requires a signed
+        // loop index; GCC/Clang accept size_t fine, so this was never caught before.
         #pragma omp for schedule(dynamic, 10)
-        for (size_t i = 0; i < m_atoms.size(); ++i) {
-            for (size_t j = i + 1; j < m_atoms.size(); ++j) {
+        for (int i = 0; i < static_cast<int>(m_atoms.size()); ++i) {
+            for (int j = i + 1; j < static_cast<int>(m_atoms.size()); ++j) {
             int atom_i = m_atoms[i];
             int atom_j = m_atoms[j];
 
@@ -747,8 +749,10 @@ void D4ParameterGenerator::GenerateParameters(const std::vector<int>& atoms, con
         {
             std::vector<json> local_triples;
 
+            // Claude Generated 2026 - MSVC's OpenMP (2.0 semantics) requires a signed
+            // loop index; GCC/Clang accept size_t fine, so this was never caught before.
             #pragma omp for schedule(dynamic, 10)
-            for (size_t idx = 0; idx < triplet_vec.size(); ++idx) {
+            for (int idx = 0; idx < static_cast<int>(triplet_vec.size()); ++idx) {
                 int i = std::get<0>(triplet_vec[idx]);
                 int j = std::get<1>(triplet_vec[idx]);
                 int k = std::get<2>(triplet_vec[idx]);
@@ -2018,9 +2022,11 @@ std::vector<GFNFFDispersion> D4ParameterGenerator::GenerateDispersionPairsNative
     {
         std::vector<GFNFFDispersion> local_pairs;
 
+        // Claude Generated 2026 - MSVC's OpenMP (2.0 semantics) requires a signed
+        // loop index; GCC/Clang accept size_t fine, so this was never caught before.
         #pragma omp for schedule(dynamic, 10)
-        for (size_t i = 0; i < m_atoms.size(); ++i) {
-            for (size_t j = i + 1; j < m_atoms.size(); ++j) {
+        for (int i = 0; i < static_cast<int>(m_atoms.size()); ++i) {
+            for (int j = i + 1; j < static_cast<int>(m_atoms.size()); ++j) {
                 double r2 = (geometry_bohr.row(i) - geometry_bohr.row(j)).squaredNorm();
                 if (r2 > disp_cutoff_sq) continue;
 

src/tools/trajectory_writer.cpp

@@ -7,6 +7,7 @@
 #include "trajectory_writer.h"
 #include "../capabilities/trajectory_statistics.h"
 #include <iomanip>
+#include <sstream>
 #include <algorithm>
 #include "../core/curcuma_logger.h"