diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
index b1f17e2d..7c8f376a 100644
--- a/.github/workflows/ci.yaml
+++ b/.github/workflows/ci.yaml
@@ -30,6 +30,11 @@ jobs:
       - name: Setup
         uses: eddelbuettel/github-actions/r-ci@master
 
+      - name: Setup ccache for R
+        uses: eddelbuettel/github-actions/ccache-with-R@master
+        with:
+          key: ${{ matrix.name }}-${{ matrix.openmp }}
+
       - name: Configure
         run: ./configure; cat src/Makevars
 
diff --git a/ChangeLog b/ChangeLog
index 6a00f94b..919ab55c 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,14 @@
+2026-07-09  Dirk Eddelbuettel  <edd@debian.org>
+
+	* DESCRIPTION (Version, Date): RcppArmadillo 15.4.1-1
+        * inst/NEWS.Rd: Idem
+        * configure.ac: Idem
+        * configure: Idem
+
+	* inst/include/armadillo_bits/: Armadillo 15.4.1
+
+	* .github/workflows/ci.yaml: Enable ccache in matrix
+
 2026-06-18  Dirk Eddelbuettel  <edd@debian.org>
 
 	* DESCRIPTION (Version, Date): RcppArmadillo 15.4.0-1
diff --git a/DESCRIPTION b/DESCRIPTION
index 65fdf3d5..0fc38ff3 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,8 +1,8 @@
 Package: RcppArmadillo
 Type: Package
 Title: 'Rcpp' Integration for the 'Armadillo' Templated Linear Algebra Library
-Version: 15.4.0-1
-Date: 2026-06-17
+Version: 15.4.1-1
+Date: 2026-07-09
 Authors@R: c(person("Dirk", "Eddelbuettel", role = c("aut", "cre"), email = "edd@debian.org",
                     comment = c(ORCID = "0000-0001-6419-907X")),
              person("Romain", "Francois", role = "aut",
diff --git a/configure b/configure
index d000293c..52db7a7d 100755
--- a/configure
+++ b/configure
@@ -1,6 +1,6 @@
 #! /bin/sh
 # Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.72 for RcppArmadillo 15.4.0-1.
+# Generated by GNU Autoconf 2.72 for RcppArmadillo 15.4.1-1.
 #
 # Report bugs to <edd@debian.org>.
 #
@@ -603,8 +603,8 @@ MAKEFLAGS=
 # Identity of this package.
 PACKAGE_NAME='RcppArmadillo'
 PACKAGE_TARNAME='rcpparmadillo'
-PACKAGE_VERSION='15.4.0-1'
-PACKAGE_STRING='RcppArmadillo 15.4.0-1'
+PACKAGE_VERSION='15.4.1-1'
+PACKAGE_STRING='RcppArmadillo 15.4.1-1'
 PACKAGE_BUGREPORT='edd@debian.org'
 PACKAGE_URL=''
 
@@ -1223,7 +1223,7 @@ if test "$ac_init_help" = "long"; then
   # Omit some internal or obsolete options to make the list less imposing.
   # This message is too long to be a string in the A/UX 3.1 sh.
   cat <<_ACEOF
-'configure' configures RcppArmadillo 15.4.0-1 to adapt to many kinds of systems.
+'configure' configures RcppArmadillo 15.4.1-1 to adapt to many kinds of systems.
 
 Usage: $0 [OPTION]... [VAR=VALUE]...
 
@@ -1285,7 +1285,7 @@ fi
 
 if test -n "$ac_init_help"; then
   case $ac_init_help in
-     short | recursive ) echo "Configuration of RcppArmadillo 15.4.0-1:";;
+     short | recursive ) echo "Configuration of RcppArmadillo 15.4.1-1:";;
    esac
   cat <<\_ACEOF
 
@@ -1366,7 +1366,7 @@ fi
 test -n "$ac_init_help" && exit $ac_status
 if $ac_init_version; then
   cat <<\_ACEOF
-RcppArmadillo configure 15.4.0-1
+RcppArmadillo configure 15.4.1-1
 generated by GNU Autoconf 2.72
 
 Copyright (C) 2023 Free Software Foundation, Inc.
@@ -1482,7 +1482,7 @@ cat >config.log <<_ACEOF
 This file contains any messages produced by compilers while
 running configure, to aid debugging if configure makes a mistake.
 
-It was created by RcppArmadillo $as_me 15.4.0-1, which was
+It was created by RcppArmadillo $as_me 15.4.1-1, which was
 generated by GNU Autoconf 2.72.  Invocation command line was
 
   $ $0$ac_configure_args_raw
@@ -3946,7 +3946,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 # report actual input values of CONFIG_FILES etc. instead of their
 # values after options handling.
 ac_log="
-This file was extended by RcppArmadillo $as_me 15.4.0-1, which was
+This file was extended by RcppArmadillo $as_me 15.4.1-1, which was
 generated by GNU Autoconf 2.72.  Invocation command line was
 
   CONFIG_FILES    = $CONFIG_FILES
@@ -4001,7 +4001,7 @@ ac_cs_config_escaped=`printf "%s\n" "$ac_cs_config" | sed "s/^ //; s/'/'\\\\\\\\
 cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
 ac_cs_config='$ac_cs_config_escaped'
 ac_cs_version="\\
-RcppArmadillo config.status 15.4.0-1
+RcppArmadillo config.status 15.4.1-1
 configured by $0, generated by GNU Autoconf 2.72,
   with options \\"\$ac_cs_config\\"
 
diff --git a/configure.ac b/configure.ac
index 38accb23..95836716 100644
--- a/configure.ac
+++ b/configure.ac
@@ -11,7 +11,7 @@
 AC_PREREQ([2.69])
 
 ## Process this file with autoconf to produce a configure script.
-AC_INIT([RcppArmadillo],[15.4.0-1],[edd@debian.org])
+AC_INIT([RcppArmadillo],[15.4.1-1],[edd@debian.org])
 
 ## Set R_HOME, respecting an environment variable if one is set
 : ${R_HOME=$(R RHOME)}
diff --git a/inst/NEWS.Rd b/inst/NEWS.Rd
index a55e772f..6d1c6b7e 100644
--- a/inst/NEWS.Rd
+++ b/inst/NEWS.Rd
@@ -3,6 +3,16 @@
 \newcommand{\ghpr}{\href{https://github.com/RcppCore/RcppArmadillo/pull/#1}{##1}}
 \newcommand{\ghit}{\href{https://github.com/RcppCore/RcppArmadillo/issues/#1}{##1}}
 
+\section{Changes in RcppArmadillo version 15.4.1-1 [github-only] (2026-07-09)}{
+  \itemize{
+    \item Upgraded to Armadillo release 15.4.1 (Medium Roast Agave)
+    \itemize{
+      \item Fix for rare infinite recursion bug in sparse version of \code{diagmat()}
+      \item More efficient checks for aliasing
+    }
+  }
+}
+
 \section{Changes in RcppArmadillo version 15.4.0-1 (2026-06-17)}{
   \itemize{
     \item Upgraded to Armadillo release 15.4.0 (Medium Roast Agave)
diff --git a/inst/include/armadillo_bits/Mat_bones.hpp b/inst/include/armadillo_bits/Mat_bones.hpp
index a56e4f7b..e68144db 100644
--- a/inst/include/armadillo_bits/Mat_bones.hpp
+++ b/inst/include/armadillo_bits/Mat_bones.hpp
@@ -938,6 +938,12 @@ class Mat<eT>::fixed : public Mat<eT>
   arma_warn_unused arma_inline       eT* memptr();
   arma_warn_unused arma_inline const eT* memptr() const;
   
+  arma_inline       subview_row<eT> row(const uword row_num);
+  arma_inline const subview_row<eT> row(const uword row_num) const;
+  
+  arma_inline       subview_col<eT> col(const uword col_num);
+  arma_inline const subview_col<eT> col(const uword col_num) const;
+  
   arma_warn_unused arma_inline bool is_vec() const;
   
   inline const Mat<eT>& fill(const eT val);
diff --git a/inst/include/armadillo_bits/Mat_meat.hpp b/inst/include/armadillo_bits/Mat_meat.hpp
index b2e3d272..df18a054 100644
--- a/inst/include/armadillo_bits/Mat_meat.hpp
+++ b/inst/include/armadillo_bits/Mat_meat.hpp
@@ -10746,6 +10746,66 @@ Mat<eT>::fixed<fixed_n_rows, fixed_n_cols>::memptr() const
 
 
 
+template<typename eT>
+template<uword fixed_n_rows, uword fixed_n_cols>
+arma_inline
+subview_row<eT>
+Mat<eT>::fixed<fixed_n_rows, fixed_n_cols>::row(const uword row_num)
+  {
+  arma_debug_sigprint();
+  
+  arma_conform_check_bounds( row_num >= fixed_n_rows, "Mat::row(): index out of bounds" );
+  
+  return subview_row<eT>(*this, row_num);
+  }
+
+
+
+template<typename eT>
+template<uword fixed_n_rows, uword fixed_n_cols>
+arma_inline
+const subview_row<eT>
+Mat<eT>::fixed<fixed_n_rows, fixed_n_cols>::row(const uword row_num) const
+  {
+  arma_debug_sigprint();
+  
+  arma_conform_check_bounds( row_num >= fixed_n_rows, "Mat::row(): index out of bounds" );
+  
+  return subview_row<eT>(*this, row_num);
+  }
+
+
+
+template<typename eT>
+template<uword fixed_n_rows, uword fixed_n_cols>
+arma_inline
+subview_col<eT>
+Mat<eT>::fixed<fixed_n_rows, fixed_n_cols>::col(const uword col_num)
+  {
+  arma_debug_sigprint();
+  
+  arma_conform_check_bounds( col_num >= fixed_n_cols, "Mat::col(): index out of bounds" );
+  
+  return subview_col<eT>(*this, col_num);
+  }
+
+
+
+template<typename eT>
+template<uword fixed_n_rows, uword fixed_n_cols>
+arma_inline
+const subview_col<eT>
+Mat<eT>::fixed<fixed_n_rows, fixed_n_cols>::col(const uword col_num) const
+  {
+  arma_debug_sigprint();
+  
+  arma_conform_check_bounds( col_num >= fixed_n_cols, "Mat::col(): index out of bounds" );
+  
+  return subview_col<eT>(*this, col_num);
+  }
+
+
+
 template<typename eT>
 template<uword fixed_n_rows, uword fixed_n_cols>
 arma_inline
diff --git a/inst/include/armadillo_bits/SpBase_bones.hpp b/inst/include/armadillo_bits/SpBase_bones.hpp
index 66b66105..29533b58 100644
--- a/inst/include/armadillo_bits/SpBase_bones.hpp
+++ b/inst/include/armadillo_bits/SpBase_bones.hpp
@@ -52,8 +52,6 @@ struct SpBase
   {
   arma_inline const derived& get_ref() const;
   
-  arma_inline bool is_alias(const SpMat<elem_type>& X) const;
-  
   arma_warn_unused inline const SpOp<derived,spop_htrans>  t() const;  //!< Hermitian transpose
   arma_warn_unused inline const SpOp<derived,spop_htrans> ht() const;  //!< Hermitian transpose
   arma_warn_unused inline const SpOp<derived,spop_strans> st() const;  //!< simple transpose
diff --git a/inst/include/armadillo_bits/SpBase_meat.hpp b/inst/include/armadillo_bits/SpBase_meat.hpp
index dab1f5cb..1f4fb24f 100644
--- a/inst/include/armadillo_bits/SpBase_meat.hpp
+++ b/inst/include/armadillo_bits/SpBase_meat.hpp
@@ -31,16 +31,6 @@ SpBase<elem_type,derived>::get_ref() const
 
 
 
-template<typename elem_type, typename derived>
-arma_inline
-bool
-SpBase<elem_type,derived>::is_alias(const SpMat<elem_type>& X) const
-  {
-  return (*this).get_ref().is_alias(X);
-  }
-
-
-
 template<typename elem_type, typename derived>
 inline
 const SpOp<derived, spop_htrans>
diff --git a/inst/include/armadillo_bits/SpGlue_bones.hpp b/inst/include/armadillo_bits/SpGlue_bones.hpp
index 6d69c1cc..4495cb35 100644
--- a/inst/include/armadillo_bits/SpGlue_bones.hpp
+++ b/inst/include/armadillo_bits/SpGlue_bones.hpp
@@ -37,7 +37,8 @@ struct SpGlue : public SpBase< typename T1::elem_type, SpGlue<T1, T2, spglue_typ
   inline  SpGlue(const T1& in_A, const T2& in_B, const elem_type in_aux);
   inline ~SpGlue();
   
-  arma_inline bool is_alias(const SpMat<elem_type>& X) const;
+  template<typename eT2>
+  arma_inline bool is_alias(const SpMat<eT2>& X) const;
   
   const T1&       A;    //!< first operand;  must be derived from SpBase
   const T2&       B;    //!< second operand; must be derived from SpBase
diff --git a/inst/include/armadillo_bits/SpGlue_meat.hpp b/inst/include/armadillo_bits/SpGlue_meat.hpp
index 4fa35d6e..0f3aa8d3 100644
--- a/inst/include/armadillo_bits/SpGlue_meat.hpp
+++ b/inst/include/armadillo_bits/SpGlue_meat.hpp
@@ -54,9 +54,10 @@ SpGlue<T1,T2,spglue_type>::~SpGlue()
 
 
 template<typename T1, typename T2, typename spglue_type>
+template<typename eT2>
 arma_inline
 bool
-SpGlue<T1,T2,spglue_type>::is_alias(const SpMat<typename T1::elem_type>& X) const
+SpGlue<T1,T2,spglue_type>::is_alias(const SpMat<eT2>& X) const
   {
   return (A.is_alias(X) || B.is_alias(X));
   }
diff --git a/inst/include/armadillo_bits/SpMat_bones.hpp b/inst/include/armadillo_bits/SpMat_bones.hpp
index 05c795a9..3d48c9eb 100644
--- a/inst/include/armadillo_bits/SpMat_bones.hpp
+++ b/inst/include/armadillo_bits/SpMat_bones.hpp
@@ -649,7 +649,8 @@ class SpMat : public SpBase< eT, SpMat<eT> >
   template<typename eT2, typename T1, typename Functor> inline void init_xform_mt(const SpBase<eT2,T1>& x, const Functor& func);
   
   //! don't use this unless you're writing internal Armadillo code
-  arma_inline bool is_alias(const SpMat<eT>& X) const;
+  template<typename eT2>
+  arma_inline bool is_alias(const SpMat<eT2>& X) const;
   
   
   protected:
diff --git a/inst/include/armadillo_bits/SpMat_meat.hpp b/inst/include/armadillo_bits/SpMat_meat.hpp
index acb9eb7e..fa652898 100644
--- a/inst/include/armadillo_bits/SpMat_meat.hpp
+++ b/inst/include/armadillo_bits/SpMat_meat.hpp
@@ -6015,11 +6015,14 @@ SpMat<eT>::init_xform_mt(const SpBase<eT2,T1>& A, const Functor& func)
 
 
 template<typename eT>
+template<typename eT2>
 arma_inline
 bool
-SpMat<eT>::is_alias(const SpMat<eT>& X) const
+SpMat<eT>::is_alias(const SpMat<eT2>& X) const
   {
-  return (&X == this);
+  arma_debug_sigprint();
+  
+  return (is_same_type<eT,eT2>::yes) && (void_ptr(this) == void_ptr(&X));
   }
 
 
diff --git a/inst/include/armadillo_bits/SpOp_bones.hpp b/inst/include/armadillo_bits/SpOp_bones.hpp
index c46834c3..271b654c 100644
--- a/inst/include/armadillo_bits/SpOp_bones.hpp
+++ b/inst/include/armadillo_bits/SpOp_bones.hpp
@@ -38,7 +38,8 @@ struct SpOp : public SpBase< typename T1::elem_type, SpOp<T1, op_type> >
   inline          SpOp(const T1& in_m, const uword     in_aux_uword_a, const uword in_aux_uword_b);
   inline         ~SpOp();
   
-  arma_inline bool is_alias(const SpMat<elem_type>& X) const;
+  template<typename eT2>
+  arma_inline bool is_alias(const SpMat<eT2>& X) const;
   
   const T1&       m;            //!< the operand; must be derived from SpBase
         elem_type aux;          //!< auxiliary data, using the element type as used by T1
diff --git a/inst/include/armadillo_bits/SpOp_meat.hpp b/inst/include/armadillo_bits/SpOp_meat.hpp
index 86a14d43..6bbc6ff6 100644
--- a/inst/include/armadillo_bits/SpOp_meat.hpp
+++ b/inst/include/armadillo_bits/SpOp_meat.hpp
@@ -64,9 +64,10 @@ SpOp<T1, op_type>::~SpOp()
 
 
 template<typename T1, typename op_type>
+template<typename eT2>
 arma_inline
 bool
-SpOp<T1, op_type>::is_alias(const SpMat<typename T1::elem_type>& X) const
+SpOp<T1, op_type>::is_alias(const SpMat<eT2>& X) const
   {
   return m.is_alias(X);
   }
diff --git a/inst/include/armadillo_bits/SpSubview_bones.hpp b/inst/include/armadillo_bits/SpSubview_bones.hpp
index fdb36240..ce64d7bb 100644
--- a/inst/include/armadillo_bits/SpSubview_bones.hpp
+++ b/inst/include/armadillo_bits/SpSubview_bones.hpp
@@ -323,7 +323,8 @@ class SpSubview : public SpBase< eT, SpSubview<eT> >
   inline const_row_iterator end_row(const uword row_num) const;
   
   //! don't use this unless you're writing internal Armadillo code
-  arma_inline bool is_alias(const SpMat<eT>& X) const;
+  template<typename eT2>
+  arma_inline bool is_alias(const SpMat<eT2>& X) const;
   
   
   private:
diff --git a/inst/include/armadillo_bits/SpSubview_col_list_bones.hpp b/inst/include/armadillo_bits/SpSubview_col_list_bones.hpp
index 47ed069f..cb04ee6e 100644
--- a/inst/include/armadillo_bits/SpSubview_col_list_bones.hpp
+++ b/inst/include/armadillo_bits/SpSubview_col_list_bones.hpp
@@ -89,6 +89,8 @@ class SpSubview_col_list : public SpBase< eT, SpSubview_col_list<eT,T1> >
   inline static void schur_inplace(SpMat<eT>& out, const SpSubview_col_list& in);
   inline static void   div_inplace(SpMat<eT>& out, const SpSubview_col_list& in);
   
+  template<typename eT2>
+  arma_inline bool is_alias(const SpMat<eT2>& X) const;
   
   friend class SpMat<eT>;
   };
diff --git a/inst/include/armadillo_bits/SpSubview_col_list_meat.hpp b/inst/include/armadillo_bits/SpSubview_col_list_meat.hpp
index 5ca7456c..cbc81e91 100644
--- a/inst/include/armadillo_bits/SpSubview_col_list_meat.hpp
+++ b/inst/include/armadillo_bits/SpSubview_col_list_meat.hpp
@@ -714,4 +714,15 @@ SpSubview_col_list<eT,T1>::div_inplace(SpMat<eT>& out, const SpSubview_col_list&
 
 
 
+template<typename eT, typename T1>
+template<typename eT2>
+arma_inline
+bool
+SpSubview_col_list<eT,T1>::is_alias(const SpMat<eT2>& X) const
+  {
+  return m.is_alias(X);
+  }
+
+
+
 //! @}
diff --git a/inst/include/armadillo_bits/SpSubview_meat.hpp b/inst/include/armadillo_bits/SpSubview_meat.hpp
index dda6dab0..0abb4108 100644
--- a/inst/include/armadillo_bits/SpSubview_meat.hpp
+++ b/inst/include/armadillo_bits/SpSubview_meat.hpp
@@ -1786,9 +1786,10 @@ SpSubview<eT>::end_row(const uword row_num) const
 
 
 template<typename eT>
+template<typename eT2>
 arma_inline
 bool
-SpSubview<eT>::is_alias(const SpMat<eT>& X) const
+SpSubview<eT>::is_alias(const SpMat<eT2>& X) const
   {
   return m.is_alias(X);
   }
diff --git a/inst/include/armadillo_bits/arma_version.hpp b/inst/include/armadillo_bits/arma_version.hpp
index 00ea4cc2..50d053dd 100644
--- a/inst/include/armadillo_bits/arma_version.hpp
+++ b/inst/include/armadillo_bits/arma_version.hpp
@@ -23,7 +23,7 @@
 
 #define ARMA_VERSION_MAJOR 15
 #define ARMA_VERSION_MINOR 4
-#define ARMA_VERSION_PATCH 0
+#define ARMA_VERSION_PATCH 1
 #define ARMA_VERSION_NAME  "Medium Roast Agave"
 
 
diff --git a/inst/include/armadillo_bits/diskio_meat.hpp b/inst/include/armadillo_bits/diskio_meat.hpp
index 53285666..e4112ee1 100644
--- a/inst/include/armadillo_bits/diskio_meat.hpp
+++ b/inst/include/armadillo_bits/diskio_meat.hpp
@@ -1865,7 +1865,7 @@ diskio::load_csv_ascii(Mat<eT>& x, std::istream& f, std::string& err_msg, const
   
   try { x.zeros(f_n_rows, f_n_cols); } catch(...) { err_msg = "not enough memory"; return false; }
   
-  if(strict)  { x.fill(Datum<eT>::nan); }   // take into account that each row may have a unique number of columns
+  if(strict && is_real<eT>::yes)  { x.fill(Datum<eT>::nan); }   // take into account that each row may have a unique number of columns
   
   const bool use_mp = (arma_config::openmp) && (f_n_rows >= 2) && (f_n_cols >= 64);
   
diff --git a/inst/include/armadillo_bits/memory.hpp b/inst/include/armadillo_bits/memory.hpp
index 49f4ac36..3e1f9071 100644
--- a/inst/include/armadillo_bits/memory.hpp
+++ b/inst/include/armadillo_bits/memory.hpp
@@ -22,9 +22,8 @@
 
 struct memory
   {
-  template<typename eT> arma_malloc inline static eT* acquire(const uword n_elem);
-  
-  template<typename eT> arma_inline static void release(eT* mem);
+  template<typename eT> arma_inline static eT*  acquire(const uword n_elem);
+  template<typename eT> arma_inline static void release(      eT*   mem   );
   
   template<typename eT> arma_inline static bool      is_aligned(const eT*  mem);
   template<typename eT> arma_inline static void mark_as_aligned(      eT*& mem);
@@ -33,9 +32,10 @@ struct memory
 
 
 
+// NOTE: arma_inline is used as a partial workaround for bugs in GCC 15:
+// NOTE: false positive warnings from -Wmismatched-new-delete and -Wmaybe-uninitialized
 template<typename eT>
-arma_malloc
-inline
+arma_inline
 eT*
 memory::acquire(const uword n_elem)
   {
diff --git a/inst/include/armadillo_bits/mtSpOp_meat.hpp b/inst/include/armadillo_bits/mtSpOp_meat.hpp
index 9a851d85..dd0081ca 100644
--- a/inst/include/armadillo_bits/mtSpOp_meat.hpp
+++ b/inst/include/armadillo_bits/mtSpOp_meat.hpp
@@ -94,7 +94,7 @@ arma_inline
 bool
 mtSpOp<out_eT, T1, op_type>::is_alias(const SpMat<eT2>& X) const
   {
-  return (void_ptr(&X) == void_ptr(&m));
+  return m.is_alias(X);
   }
 
 
diff --git a/inst/include/armadillo_bits/mtSpReduceOp_bones.hpp b/inst/include/armadillo_bits/mtSpReduceOp_bones.hpp
index 60d75b7e..418bb2f8 100644
--- a/inst/include/armadillo_bits/mtSpReduceOp_bones.hpp
+++ b/inst/include/armadillo_bits/mtSpReduceOp_bones.hpp
@@ -52,6 +52,9 @@ struct mtSpReduceOp : public SpBase< out_eT, mtSpReduceOp<out_eT, T1, op_type> >
   inline          mtSpReduceOp(const T1& in_m, const uword in_aux_uword_a, const uword in_aux_uword_b);
   inline         ~mtSpReduceOp();
   
+  template<typename eT2>
+  arma_inline bool is_alias(const SpMat<eT2>& X) const;
+  
   const T1&   m;            //!< the operand; must be derived from SpBase
         uword aux_uword_a;  //!< auxiliary data, uword format
         uword aux_uword_b;  //!< auxiliary data, uword format
diff --git a/inst/include/armadillo_bits/mtSpReduceOp_meat.hpp b/inst/include/armadillo_bits/mtSpReduceOp_meat.hpp
index 531ce239..fd87b9ae 100644
--- a/inst/include/armadillo_bits/mtSpReduceOp_meat.hpp
+++ b/inst/include/armadillo_bits/mtSpReduceOp_meat.hpp
@@ -52,4 +52,15 @@ mtSpReduceOp<out_eT, T1, op_type>::~mtSpReduceOp()
 
 
 
+template<typename out_eT, typename T1, typename op_type>
+template<typename eT2>
+arma_inline
+bool
+mtSpReduceOp<out_eT, T1, op_type>::is_alias(const SpMat<eT2>& X) const
+  {
+  return m.is_alias(X);
+  }
+
+
+
 //! @}
diff --git a/inst/include/armadillo_bits/op_diagmat_bones.hpp b/inst/include/armadillo_bits/op_diagmat_bones.hpp
index 91ded0bc..e072a18c 100644
--- a/inst/include/armadillo_bits/op_diagmat_bones.hpp
+++ b/inst/include/armadillo_bits/op_diagmat_bones.hpp
@@ -27,6 +27,9 @@ struct op_diagmat
   template<typename T1>
   inline static void apply(Mat<typename T1::elem_type>& out, const Op<T1,op_diagmat>& X);
   
+  template<typename T1>
+  inline static void apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1,op_diagmat>& X);
+  
   template<typename T1>
   inline static void apply(Mat<typename T1::elem_type>& out, const Proxy<T1>& P);
   
@@ -48,6 +51,9 @@ struct op_diagmat2
   template<typename T1>
   inline static void apply(Mat<typename T1::elem_type>& out, const Op<T1,op_diagmat2>& X);
   
+  template<typename T1>
+  inline static void apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1,op_diagmat2>& X);
+  
   template<typename T1>
   inline static void apply(Mat<typename T1::elem_type>& out, const Proxy<T1>& P, const uword row_offset, const uword col_offset);
   };
diff --git a/inst/include/armadillo_bits/op_diagmat_meat.hpp b/inst/include/armadillo_bits/op_diagmat_meat.hpp
index 1dd0e58a..5f59f84b 100644
--- a/inst/include/armadillo_bits/op_diagmat_meat.hpp
+++ b/inst/include/armadillo_bits/op_diagmat_meat.hpp
@@ -104,6 +104,20 @@ op_diagmat::apply(Mat<typename T1::elem_type>& out, const Op<T1, op_diagmat>& X)
 
 
 
+template<typename T1>
+inline
+void
+op_diagmat::apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1, op_diagmat>& X)
+  {
+  arma_debug_sigprint();
+  
+  const Proxy<T1> P(X.m);
+  
+  op_diagmat::apply(out, P);
+  }
+
+
+
 template<typename T1>
 inline
 void
@@ -701,6 +715,23 @@ op_diagmat2::apply(Mat<typename T1::elem_type>& out, const Op<T1, op_diagmat2>&
 
 
 
+template<typename T1>
+inline
+void
+op_diagmat2::apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1, op_diagmat2>& X)
+  {
+  arma_debug_sigprint();
+  
+  const uword row_offset = X.aux_uword_a;
+  const uword col_offset = X.aux_uword_b;
+  
+  const Proxy<T1> P(X.m);
+  
+  op_diagmat2::apply(out, P, row_offset, col_offset);
+  }
+
+
+
 template<typename T1>
 inline
 void
diff --git a/inst/include/armadillo_bits/op_diagvec_bones.hpp b/inst/include/armadillo_bits/op_diagvec_bones.hpp
index 64bd7cb1..68461681 100644
--- a/inst/include/armadillo_bits/op_diagvec_bones.hpp
+++ b/inst/include/armadillo_bits/op_diagvec_bones.hpp
@@ -27,6 +27,9 @@ struct op_diagvec
   template<typename T1>
   inline static void apply(Mat<typename T1::elem_type>& out, const Op<T1,op_diagvec>& X);
   
+  template<typename T1>
+  inline static void apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1,op_diagvec>& X);
+  
   template<typename T1>
   inline static void apply_proxy(Mat<typename T1::elem_type>& out, const Proxy<T1>& P);
   
@@ -45,6 +48,9 @@ struct op_diagvec2
   template<typename T1>
   inline static void apply(Mat<typename T1::elem_type>& out, const Op<T1,op_diagvec2>& X);
   
+  template<typename T1>
+  inline static void apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1,op_diagvec2>& X);
+  
   template<typename T1>
   inline static void apply_proxy(Mat<typename T1::elem_type>& out, const Proxy<T1>& P, const uword row_offset, const uword col_offset);
   };
diff --git a/inst/include/armadillo_bits/op_diagvec_meat.hpp b/inst/include/armadillo_bits/op_diagvec_meat.hpp
index 42117a3e..c4ff93b9 100644
--- a/inst/include/armadillo_bits/op_diagvec_meat.hpp
+++ b/inst/include/armadillo_bits/op_diagvec_meat.hpp
@@ -48,6 +48,20 @@ op_diagvec::apply(Mat<typename T1::elem_type>& out, const Op<T1, op_diagvec>& X)
 
 
 
+template<typename T1>
+inline
+void
+op_diagvec::apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1, op_diagvec>& X)
+  {
+  arma_debug_sigprint();
+  
+  const Proxy<T1> P(X.m);
+  
+  op_diagvec::apply_proxy(out, P);
+  }
+
+
+
 template<typename T1>
 inline
 void
@@ -491,6 +505,26 @@ op_diagvec2::apply(Mat<typename T1::elem_type>& out, const Op<T1, op_diagvec2>&
 
 
 
+template<typename T1>
+inline
+void
+op_diagvec2::apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1, op_diagvec2>& X)
+  {
+  arma_debug_sigprint();
+  
+  const uword a = X.aux_uword_a;
+  const uword b = X.aux_uword_b;
+  
+  const uword row_offset = (b >  0) ? a : 0;
+  const uword col_offset = (b == 0) ? a : 0;
+  
+  const Proxy<T1> P(X.m);
+  
+  op_diagvec2::apply_proxy(out, P, row_offset, col_offset);
+  }
+
+
+
 template<typename T1>
 inline
 void
diff --git a/inst/include/armadillo_bits/op_repelem_bones.hpp b/inst/include/armadillo_bits/op_repelem_bones.hpp
index 0f58c952..d826a262 100644
--- a/inst/include/armadillo_bits/op_repelem_bones.hpp
+++ b/inst/include/armadillo_bits/op_repelem_bones.hpp
@@ -28,6 +28,8 @@ struct op_repelem
   template<typename obj> inline static void apply_noalias(Mat<typename obj::elem_type>& out, const obj& X, const uword copies_per_row, const uword copies_per_col);
   
   template<typename T1> inline static void apply(Mat<typename T1::elem_type>& out, const Op<T1,op_repelem>& in);
+  
+  template<typename T1> inline static void apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1,op_repelem>& in);
   };
 
 
diff --git a/inst/include/armadillo_bits/op_repelem_meat.hpp b/inst/include/armadillo_bits/op_repelem_meat.hpp
index 9da9f431..5e9342a4 100644
--- a/inst/include/armadillo_bits/op_repelem_meat.hpp
+++ b/inst/include/armadillo_bits/op_repelem_meat.hpp
@@ -100,4 +100,21 @@ op_repelem::apply(Mat<typename T1::elem_type>& out, const Op<T1,op_repelem>& in)
 
 
 
+template<typename T1>
+inline
+void
+op_repelem::apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1,op_repelem>& in)
+  {
+  arma_debug_sigprint();
+  
+  const uword copies_per_row = in.aux_uword_a;
+  const uword copies_per_col = in.aux_uword_b;
+  
+  const quasi_unwrap<T1> U(in.m);
+  
+  op_repelem::apply_noalias(out, U.M, copies_per_row, copies_per_col);
+  }
+
+
+
 //! @}
diff --git a/inst/include/armadillo_bits/op_vectorise_bones.hpp b/inst/include/armadillo_bits/op_vectorise_bones.hpp
index eabe2eed..c446f655 100644
--- a/inst/include/armadillo_bits/op_vectorise_bones.hpp
+++ b/inst/include/armadillo_bits/op_vectorise_bones.hpp
@@ -34,8 +34,6 @@ struct op_vectorise_col
   template<typename T1> inline static void apply_direct(Mat_noalias<typename T1::elem_type>& out, const T1& expr);
   
   template<typename eT> inline static void apply_subview(Mat<eT>& out, const subview<eT>& sv);
-  
-  template<typename T1> inline static void apply_proxy(Mat<typename T1::elem_type>& out, const Proxy<T1>& P);
   };
 
 
diff --git a/inst/include/armadillo_bits/op_vectorise_meat.hpp b/inst/include/armadillo_bits/op_vectorise_meat.hpp
index 27d4e201..40cf856f 100644
--- a/inst/include/armadillo_bits/op_vectorise_meat.hpp
+++ b/inst/include/armadillo_bits/op_vectorise_meat.hpp
@@ -80,39 +80,12 @@ op_vectorise_col::apply_direct(Mat<typename T1::elem_type>& out, const T1& expr)
       }
     }
   else
-  if( (quasi_unwrap<T1>::has_orig_mem) || (is_Mat<typename Proxy<T1>::stored_type>::value) || (arma_config::openmp && Proxy<T1>::use_mp) )
     {
-    const quasi_unwrap<T1> U(expr);
+    Mat<eT> tmp = expr;
     
-    if(U.is_alias(out))
-      {
-      Mat<eT> tmp(U.M.memptr(), U.M.n_elem, 1);
-      
-      out.steal_mem(tmp);
-      }
-    else
-      {
-      out.set_size(U.M.n_elem, 1);
-      
-      arrayops::copy(out.memptr(), U.M.memptr(), U.M.n_elem);
-      }
-    }
-  else
-    {
-    const Proxy<T1> P(expr);
+    tmp.set_size(tmp.n_elem, 1);
     
-    if(P.is_alias(out))
-      {
-      Mat<eT> tmp;
-      
-      op_vectorise_col::apply_proxy(tmp, P);
-      
-      out.steal_mem(tmp);
-      }
-    else
-      {
-      op_vectorise_col::apply_proxy(out, P);
-      }
+    out.steal_mem(tmp);
     }
   }
 
@@ -133,12 +106,17 @@ op_vectorise_col::apply(Mat_noalias<typename T1::elem_type>& out, const Op<T1,op
 template<typename T1>
 inline
 void
-op_vectorise_col::apply_direct(Mat_noalias<typename T1::elem_type>& out, const T1& expr)
+op_vectorise_col::apply_direct(Mat_noalias<typename T1::elem_type>& actual_out, const T1& expr)
   {
   arma_debug_sigprint();
   
   typedef typename T1::elem_type eT;
   
+  Mat<eT>& out = actual_out;
+  
+  // no special handling for T1 = Mat, as that currently can't happen;
+  // the Mat class uses the Mat_noalias type only for delayed expressions
+  
   if(is_subview<T1>::value)
     {
     const subview<eT>& sv = reinterpret_cast< const subview<eT>& >(expr);
@@ -146,19 +124,12 @@ op_vectorise_col::apply_direct(Mat_noalias<typename T1::elem_type>& out, const T
     op_vectorise_col::apply_subview(out, sv);
     }
   else
-  if( (quasi_unwrap<T1>::has_orig_mem) || (is_Mat<typename Proxy<T1>::stored_type>::value) || (arma_config::openmp && Proxy<T1>::use_mp) )
     {
-    const quasi_unwrap<T1> U(expr);
-    
-    out.set_size(U.M.n_elem, 1);
+    Mat<eT> tmp = expr;
     
-    arrayops::copy(out.memptr(), U.M.memptr(), U.M.n_elem);
-    }
-  else
-    {
-    const Proxy<T1> P(expr);
+    tmp.set_size(tmp.n_elem, 1);
     
-    op_vectorise_col::apply_proxy(out, P);
+    out.steal_mem(tmp);
     }
   }
 
@@ -190,71 +161,6 @@ op_vectorise_col::apply_subview(Mat<eT>& out, const subview<eT>& sv)
 
 
 
-template<typename T1>
-inline
-void
-op_vectorise_col::apply_proxy(Mat<typename T1::elem_type>& out, const Proxy<T1>& P)
-  {
-  arma_debug_sigprint();
-  
-  typedef typename T1::elem_type eT;
-  
-  const uword N = P.get_n_elem();
-  
-  out.set_size(N, 1);
-  
-  if(N == 0)  { return; }
-  
-  eT* outmem = out.memptr();
-  
-  if(Proxy<T1>::use_at == false)
-    {
-    // TODO: add handling of aligned access ?
-    
-    typename Proxy<T1>::ea_type A = P.get_ea();
-    
-    uword i,j;
-    
-    for(i=0, j=1; j < N; i+=2, j+=2)
-      {
-      const eT tmp_i = A[i];
-      const eT tmp_j = A[j];
-      
-      outmem[i] = tmp_i;
-      outmem[j] = tmp_j;
-      }
-    
-    if(i < N)
-      {
-      outmem[i] = A[i];
-      }
-    }
-  else
-    {
-    const uword n_rows = P.get_n_rows();
-    const uword n_cols = P.get_n_cols();
-    
-    if(n_rows == 1)
-      {
-      for(uword i=0; i < n_cols; ++i)
-        {
-        outmem[i] = P.at(0,i);
-        }
-      }
-    else
-      {
-      for(uword col=0; col < n_cols; ++col)
-      for(uword row=0; row < n_rows; ++row)
-        {
-        *outmem = P.at(row,col);
-        outmem++;
-        }
-      }
-    }
-  }
-
-
-
 template<typename T1>
 inline
 void
diff --git a/inst/include/armadillo_bits/spdiagview_bones.hpp b/inst/include/armadillo_bits/spdiagview_bones.hpp
index 9f4b24be..269965ef 100644
--- a/inst/include/armadillo_bits/spdiagview_bones.hpp
+++ b/inst/include/armadillo_bits/spdiagview_bones.hpp
@@ -107,6 +107,9 @@ class spdiagview : public SpBase< eT, spdiagview<eT> >
   inline static void extract(SpMat<eT>& out, const spdiagview& in);
   inline static void extract(  Mat<eT>& out, const spdiagview& in);
   
+  template<typename eT2>
+  arma_inline bool is_alias(const SpMat<eT2>& X) const;
+  
   
   friend class SpMat<eT>;
   };
diff --git a/inst/include/armadillo_bits/spdiagview_meat.hpp b/inst/include/armadillo_bits/spdiagview_meat.hpp
index 480dc256..d7a36841 100644
--- a/inst/include/armadillo_bits/spdiagview_meat.hpp
+++ b/inst/include/armadillo_bits/spdiagview_meat.hpp
@@ -1076,4 +1076,15 @@ spdiagview<eT>::randn()
 
 
 
+template<typename eT>
+template<typename eT2>
+arma_inline
+bool
+spdiagview<eT>::is_alias(const SpMat<eT2>& X) const
+  {
+  return m.is_alias(X);
+  }
+
+
+
 //! @}
diff --git a/inst/include/armadillo_bits/spop_diagmat_bones.hpp b/inst/include/armadillo_bits/spop_diagmat_bones.hpp
index f19a1cda..fb61b223 100644
--- a/inst/include/armadillo_bits/spop_diagmat_bones.hpp
+++ b/inst/include/armadillo_bits/spop_diagmat_bones.hpp
@@ -26,6 +26,9 @@ struct spop_diagmat
   template<typename T1>
   inline static void apply(SpMat<typename T1::elem_type>& out, const SpOp<T1, spop_diagmat>& in);
   
+  template<typename T1>
+  inline static void apply(SpMat_noalias<typename T1::elem_type>& out, const SpOp<T1, spop_diagmat>& in);
+  
   template<typename T1>
   inline static void apply_noalias(SpMat<typename T1::elem_type>& out, const SpBase<typename T1::elem_type, T1>& expr);
   
@@ -40,7 +43,6 @@ struct spop_diagmat
   
   template<typename T1, typename T2>
   inline static void apply_noalias(SpMat<typename T1::elem_type>& out, const SpGlue<T1,T2,spglue_times>& expr);
-  
   };
 
 
@@ -51,6 +53,9 @@ struct spop_diagmat2
   template<typename T1>
   inline static void apply(SpMat<typename T1::elem_type>& out, const SpOp<T1, spop_diagmat2>& in);
   
+  template<typename T1>
+  inline static void apply(SpMat_noalias<typename T1::elem_type>& out, const SpOp<T1, spop_diagmat2>& in);
+  
   template<typename eT>
   inline static void apply_noalias(SpMat<eT>& out, const SpMat<eT>& X, const uword row_offset, const uword col_offset);
   };
diff --git a/inst/include/armadillo_bits/spop_diagmat_meat.hpp b/inst/include/armadillo_bits/spop_diagmat_meat.hpp
index 19503b8c..314e2c89 100644
--- a/inst/include/armadillo_bits/spop_diagmat_meat.hpp
+++ b/inst/include/armadillo_bits/spop_diagmat_meat.hpp
@@ -46,6 +46,18 @@ spop_diagmat::apply(SpMat<typename T1::elem_type>& out, const SpOp<T1, spop_diag
 
 
 
+template<typename T1>
+inline
+void
+spop_diagmat::apply(SpMat_noalias<typename T1::elem_type>& out, const SpOp<T1, spop_diagmat>& in)
+  {
+  arma_debug_sigprint();
+  
+  spop_diagmat::apply_noalias(out, in.m);
+  }
+
+
+
 template<typename T1>
 inline
 void
@@ -374,6 +386,23 @@ spop_diagmat2::apply(SpMat<typename T1::elem_type>& out, const SpOp<T1, spop_dia
 
 
 
+template<typename T1>
+inline
+void
+spop_diagmat2::apply(SpMat_noalias<typename T1::elem_type>& out, const SpOp<T1, spop_diagmat2>& in)
+  {
+  arma_debug_sigprint();
+  
+  const uword row_offset = in.aux_uword_a;
+  const uword col_offset = in.aux_uword_b;
+  
+  const unwrap_spmat<T1> U(in.m);
+  
+  spop_diagmat2::apply_noalias(out, U.M, row_offset, col_offset);
+  }
+
+
+
 template<typename eT>
 inline
 void