note data race bug

CMakeLists.txt

@@ -196,8 +196,6 @@ set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_CURRENT_SOURCE_DIR}/cmake_au
 if(APPLE)
   message(STATUS "Detected macOS")
   
-  set(ARMA_OS macos)
-  
   set(ARMA_USE_LAPACK     true)
   set(ARMA_USE_BLAS       true)
   set(ARMA_USE_ACCELERATE true)
@@ -258,8 +256,6 @@ if(APPLE)
   
 else()
   
-  set(ARMA_OS unix)
-  
   include(ARMA_FindMKL)
   include(ARMA_FindACMLMP)
   include(ARMA_FindACML)
@@ -651,5 +647,5 @@ install(FILES "${PROJECT_BINARY_DIR}/tmp/misc/armadillo.pc" DESTINATION ${CMAKE_
 if(BUILD_SMOKE_TEST)
   message(STATUS "*** configuring smoke_test")
   enable_testing()
-  add_subdirectory(smoke_test)
+  add_subdirectory(tests1)
 endif()

README.md

@@ -22,7 +22,7 @@ Copyright 2017-2020 Data61, CSIRO
 2.  [Citation Details](#2-citation-details)
 3.  [Distribution License](#3-distribution-license)
 
-4.  [Compilers and External Dependencies](#4-compilers-and-external-dependencies)
+4.  [Prerequisites](#4-prerequisites)
 
 5.  [Linux and macOS: Installation](#5-linux-and-macos-installation)
 6.  [Linux and macOS: Compiling and Linking](#6-linux-and-macos-compiling-and-linking)
@@ -35,7 +35,7 @@ Copyright 2017-2020 Data61, CSIRO
 11. [Support for C++11 / C++14 Features](#11-support-for-c11-c14-features)
 12. [Support for OpenMP](#12-support-for-openmp)
 
-13. [Documentation](#13-documentation)
+13. [Documentation of Functions and Classes](#13-documentation-of-functions-and-classes)
 14. [API Stability and Versioning](#14-api-stability-and-versioning)
 15. [Bug Reports and Frequently Asked Questions](#15-bug-reports-and-frequently-asked-questions)
 
@@ -57,8 +57,8 @@ The library provides efficient classes for vectors, matrices and cubes,
 as well as 200+ associated functions covering essential and advanced functionality
 for data processing and manipulation of matrices.
 
-Various matrix decompositions are provided through integration with LAPACK,
-or one of its high performance drop-in replacements
+Various matrix decompositions (eigen, SVD, QR, etc) are provided through
+integration with LAPACK, or one of its high performance drop-in replacements
 (eg. OpenBLAS, Intel MKL, Apple Accelerate framework, etc).
 
 A sophisticated expression evaluator (via C++ template meta-programming)
@@ -104,154 +104,163 @@ informational purposes only and do not modify the License.
 
 ---
 
-### 4: Compilers and External Dependencies
+### 4: Prerequisites
 
-A compiler that properly supports the C++98/C++03 standards is required,
-as Armadillo makes extensive use of template meta-programming.
+Armadillo 9.900 requires a C++ compiler with proper support for at least the C++98/C++03 standards.
+Upgrading to Armadillo 10.6 is recommended to make best use of C++11/C++14 features.
+
+On Linux-based systems, install the GCC C++ compiler, which is available as a pre-built package.
+The package name might be `g++` or `gcc-c++` depending on your system.
+
+On macOS systems, a C++ compiler can be obtained by first installing Xcode (version 8 or later)
+and then running the following command in a terminal window:  
+
+    xcode-select --install
+
+On Windows systems, the MinGW toolset or Visual Studio C++ 2012 (MSVC) can be used.
 
 The functionality of Armadillo is partly dependent on other libraries:
-LAPACK, BLAS (preferably OpenBLAS), ARPACK and SuperLU.
-LAPACK and BLAS are used for dense matrices,
-while ARPACK and SuperLU are used for sparse matrices.
+OpenBLAS (or standard BLAS) and LAPACK (for dense matrices),
+as well as ARPACK and SuperLU (for sparse matrices).
+Caveat: only SuperLU versions 5.2.x can be used.
+On macOS, the Accelerate framework can be used for BLAS and LAPACK functions.
 
 Armadillo can work without the above libraries, but its functionality will be reduced.
 Basic functionality will be available (eg. matrix addition and multiplication),
-but operations like eigen decomposition or matrix inversion will not be.
-Matrix multiplication (mainly for big matrices) may not be as fast.
-
-As Armadillo is a template library, we recommended that optimisation
-is enabled during compilation of programs that use Armadillo.
-For example, for GCC and Clang compilers use -O2 or -O3
+but operations such as eigen decomposition and system solvers will not be.
+Matrix multiplication may not be as fast (mainly for large matrices).
 
 ---
 
 ### 5: Linux and macOS: Installation
 
-* Step 1:
-  Ensure a C++ compiler is installed on your system.
+Armadillo can be installed in several ways: either manually or via cmake, with or without root access.
+The cmake based installation is preferred.
+The cmake tool can be downloaded from http://www.cmake.org
+or (preferably) installed using the package manager on your system;
+on macOS systems, cmake can be installed through MacPorts or Homebrew.
 
-  - On macOS systems install Xcode (version 8 or later)
-    and then type the following command in a terminal window:  
+Before installing Armadillo, first install OpenBLAS and LAPACK, and optionally ARPACK and SuperLU.
+It is also necessary to install the corresponding development files for each library.
+For example, when installing the `libopenblas` package, also install the `libopenblas-dev` package.
 
-    xcode-select --install
 
-* Step 2:
-  Ensure the CMake tool is installed on your system.
+#### 5a: Installation via CMake
 
-  - Cmake can be downloaded from http://www.cmake.org
-    or (preferably) installed using the package manager on your system.
+The cmake based installer detects which relevant libraries
+are installed on your system (eg. OpenBLAS, LAPACK, SuperLU, ARPACK, etc)
+and correspondingly modifies Armadillo's configuration.
+The installer also generates the Armadillo runtime library,
+which is a wrapper for all the detected libraries,
+and provides a thread-safe random number generator.
 
-  - On Linux-based systems, CMake can be installed using dnf, yum, apt, aptitude, ...
-
-  - On macOS systems, CMake can be installed through MacPorts or Homebrew.
-
-* Step 3:
-  Ensure that OpenBLAS (or standard BLAS and LAPACK) is installed on your system.
-  On macOS, the Accelerate framework can be used for BLAS/LAPACK.
-
-  - On macOS, optionally install OpenBLAS for better performance.
-
-  - If support for sparse matrices is required, also install ARPACK and SuperLU.
-    Caveat: only SuperLU version 5.2 can be used!
-
-  - On Linux-based systems, the following libraries are recommended
-    to be present: OpenBLAS, LAPACK, SuperLU and ARPACK.
-    It is also necessary to install the corresponding development
-    files for each library. For example, when installing the "lapack"
-    package, also install the "lapack-devel" or "lapack-dev" package.
-  
-* Step 4:
-  Run the cmake installer.
-
-  - Open a terminal window and change into the directory that was created
-    by unpacking the armadillo archive.
-  
-  - The simplest case is to run cmake using:
+Change into the directory that was created by unpacking the armadillo archive
+(eg. `cd armadillo-9.900.5`) and then run cmake using:
 
     cmake .
 
-  - NOTE: the full stop separated from "cmake" by a space is important.
-  
-  - Options to the cmake installer:
-  
-    - On Linux, to enable the detection of FlexiBLAS, 
-      use the additional ALLOW_FLEXIBLAS_LINUX option when running cmake:
+**NOTE:** the full stop (.) separated from `cmake` by a space is important.
 
-      cmake -DALLOW_FLEXIBLAS_LINUX=ON .
+On macOS, to enable the detection of OpenBLAS, 
+use the additional `ALLOW_OPENBLAS_MACOS` option when running cmake:
 
-    - On macOS, to enable the detection of OpenBLAS, 
-      use the additional ALLOW_OPENBLAS_MACOS option when running cmake:
+    cmake -DALLOW_OPENBLAS_MACOS=ON .
 
-      cmake -DALLOW_OPENBLAS_MACOS=ON .
+Depending on your installation, OpenBLAS may masquerade as standard BLAS.
+To detect standard BLAS and LAPACK, use the `ALLOW_BLAS_LAPACK_MACOS` option:
 
-      Note: depending on your installation, OpenBLAS may masquerade as standard BLAS.
-      To detect standard BLAS and LAPACK, use the ALLOW_BLAS_LAPACK_MACOS option:
+    cmake -DALLOW_BLAS_LAPACK_MACOS=ON .
 
-      cmake -DALLOW_BLAS_LAPACK_MACOS=ON .
+By default, cmake assumes that the Armadillo runtime library and the corresponding header files 
+will be installed in the default system directory (eg. in the `/usr` hierarchy in Linux-based systems).
+To install the library and headers in an alternative directory,
+use the additional option `CMAKE_INSTALL_PREFIX` in this form:
 
-    - By default, cmake assumes that the Armadillo library and the
-      corresponding header files will be installed in the default 
-      system directory (eg. in the /usr hierarchy in Linux-based systems).
-      To install the library and headers in an alternative directory,
-      use the additional option CMAKE_INSTALL_PREFIX in this form:
+    cmake . -DCMAKE_INSTALL_PREFIX:PATH=alternative_directory
 
-      cmake . -DCMAKE_INSTALL_PREFIX:PATH=alternative_directory
+If cmake needs to be re-run, it's a good idea to first delete the `CMakeCache.txt` file
+(not `CMakeLists.txt`).
 
-  - CMake will detect which relevant libraries are installed on your system
-    (eg. OpenBLAS, LAPACK, SuperLU, ARPACK, etc)
-    and will modify Armadillo's configuration correspondingly.
-    CMake will also generate the Armadillo run-time library,
-    which is a wrapper for all the detected libraries.
+**Caveat:** if Armadillo is installed in a non-system directory,
+make sure that the C++ compiler is configured to use the `lib` and `include`
+sub-directories present within this directory.
+Note that the `lib` directory might be named differently on your system.
+On recent 64 bit Debian & Ubuntu systems it is `lib/x86_64-linux-gnu`.
+On recent 64 bit Fedora & RHEL systems it is `lib64`.
 
-  - If cmake needs to re-run, it's a good idea to first delete the
-    "CMakeCache.txt" file (not "CMakeLists.txt").
-
-  - Caveat: if Armadillo is installed in a non-system directory,
-    make sure that the C++ compiler is configured to use the "lib" and "include"
-    sub-directories present within this directory.  Note that the "lib"
-    directory might be named differently on your system.
-    On recent 64 bit Debian & Ubuntu systems it is "lib/x86_64-linux-gnu".
-    On recent 64 bit Fedora & RHEL systems it is "lib64".
-
-* Step 5:
-  If you and have access to root/administrator/superuser privileges
-  (ie. able to use "sudo") and didn't use the CMAKE_INSTALL_PREFIX option,
-  type the following command:
+If you have sudo access (ie. root/administrator/superuser privileges)
+and didn't use the `CMAKE_INSTALL_PREFIX` option, run the following command:
 
     sudo make install
 
-  If you don't have root/administrator/superuser privileges,
-  make sure that you use the CMAKE_INSTALL_PREFIX option in Step 4,
-  and type the following command:
+If you don't have sudo access, make sure to use the `CMAKE_INSTALL_PREFIX` option
+and run the following command:
 
     make install
 
+
+#### 5b: Manual Installation
+
+Manual installation involves simply copying the `include/armadillo` header
+**and** the associated `include/armadillo_bits` directory to a location
+such as `/usr/include/` which is searched by your C++ compiler.
+If you don't have sudo access or don't have write access to `/usr/include/`,
+use a directory within your own home directory (eg. `/home/blah/include/`).
+
+If required, modify `include/armadillo_bits/config.hpp`
+to indicate which libraries are currently available on your system.
+Comment or uncomment the following lines:
+
+    #define ARMA_USE_LAPACK  
+    #define ARMA_USE_BLAS  
+    #define ARMA_USE_ARPACK  
+    #define ARMA_USE_SUPERLU  
+
+If support for sparse matrices is not needed, ARPACK and SuperLU are not necessary.
+
+Note that the manual installation will not generate the Armadillo runtime library,
+and hence you will need to link your programs directly with OpenBLAS, LAPACK, etc.
+
 ---
 
 ### 6: Linux and macOS: Compiling and Linking
 
-If you have installed Armadillo via the CMake installer,
-use the following command:
+If you have installed Armadillo via the cmake installer,
+use the following command to compile your programs:
 
     g++ prog.cpp -o prog -O2 -larmadillo
 
-Otherwise, if you want to use Armadillo without installation
-(ie. without the Armadillo runtime library), use the following command:
-  
-    g++ prog.cpp -o prog.cpp -O2 -I /home/blah/armadillo-7.200.3/include -DARMA_DONT_USE_WRAPPER -lopenblas
+If you have installed Armadillo manually, link with OpenBLAS and LAPACK
+instead of the Armadillo runtime library:
 
-The above command assumes that the armadillo archive was unpacked into /home/blah/  
-The command needs to be adjusted if the archive was unpacked into a different directory
-and/or for each specific version of Armadillo (ie. "7.200.3" needs to be changed).
-  
-If you don't have OpenBLAS, on Linux change -lopenblas to -lblas -llapack
-and on macOS change -lopenblas to -framework Accelerate
+    g++ prog.cpp -o prog -O2 -lopenblas -llapack
 
-See the Questions page for more info on linking:
+If you have manually installed Armadillo in a non-standard location,
+such as `/home/blah/include/`, you will need to make sure 
+that your C++ compiler searches `/home/blah/include/` 
+by explicitly specifying the directory as an argument/option. 
+For example, using the `-I` switch in GCC and Clang:
+
+    g++ prog.cpp -o prog -O2 -I /home/blah/include/ -lopenblas -llapack
+
+If you're getting linking issues (unresolved symbols),
+enable the `ARMA_DONT_USE_WRAPPER` option:
+
+    g++ prog.cpp -o prog -O2 -I /home/blah/include/ -DARMA_DONT_USE_WRAPPER -lopenblas -llapack
+
+If you don't have OpenBLAS, on Linux change `-lopenblas` to `-lblas`;
+on macOS change `-lopenblas -llapack` to `-framework Accelerate`
+
+The `examples` directory contains a short example program that uses Armadillo.
+
+We recommend that compilation is done with optimisation enabled,
+in order to make best use of the extensive template meta-programming
+techniques employed in Armadillo.
+For GCC and Clang compilers use `-O2` or `-O3` to enable optimisation.
+
+For more information on compiling and linking, see the Questions page: 
 http://arma.sourceforge.net/faq.html
 
-The "examples" directory contains a short example program that uses the Armadillo library.
-
 ---
 
 ### 7: Windows: Installation
@@ -259,43 +268,43 @@ The "examples" directory contains a short example program that uses the Armadill
 The installation is comprised of 3 steps:
 
 * Step 1:
-  Copy the entire "include" folder to a convenient location
+  Copy the entire `include` folder to a convenient location
   and tell your compiler to use that location for header files
   (in addition to the locations it uses already).
-  Alternatively, the "include" folder can be used directly.
+  Alternatively, the `include` folder can be used directly.
 
 * Step 2:
-  Modify "include/armadillo_bits/config.hpp" to indicate which
-  libraries are currently available on your system. For example,
-  if LAPACK, BLAS (or OpenBLAS), ARPACK and SuperLU present,
-  uncomment the following lines:
+  If required, modify `include/armadillo_bits/config.hpp`
+  to indicate which libraries are currently available on your system:
 
     #define ARMA_USE_LAPACK  
     #define ARMA_USE_BLAS  
     #define ARMA_USE_ARPACK  
     #define ARMA_USE_SUPERLU  
 
-  If support for sparse matrices is not required,
-  don't worry about ARPACK or SuperLU.
+  If support for sparse matrices is not needed, ARPACK or SuperLU are not necessary.
 
 * Step 3:
   Configure your compiler to link with LAPACK and BLAS
   (and optionally ARPACK and SuperLU).
+  Note that OpenBLAS can be used as a high-performance substitute
+  for both LAPACK and BLAS.
 
 ---
 
 ### 8: Windows: Compiling and Linking
 
-Within the "examples" folder, there is an MSVC project named "example1_win64"
-which can be used to compile "example1.cpp". The project needs to be compiled as a
-64 bit program: the active solution platform must be set to x64, instead of win32.
+Within the `examples` folder, the MSVC project named `example1_win64`
+can be used to compile `example1.cpp`.
+The project needs to be compiled as a 64 bit program:
+the active solution platform must be set to x64, instead of win32.
 
 The MSVC project was tested on Windows 7 (64 bit) with Visual Studio C++ 2012.
-Adaptations may need to be made for 32 bit systems, later versions of Windows
-and/or the compiler. For example, options such as ARMA_BLAS_LONG and ARMA_BLAS_UNDERSCORE,
-defined in "armadillo_bits/config.hpp", may need to be either enabled or disabled.
+Adaptations may be required for 32 bit systems, later versions of Windows and/or the compiler.
+For example, options such as `ARMA_BLAS_LONG` and `ARMA_BLAS_UNDERSCORE`,
+defined in `include/armadillo_bits/config.hpp`, may need to be either enabled or disabled.
 
-The folder "examples/lib_win64" contains baseline (unoptimised) LAPACK and BLAS
+The folder `examples/lib_win64` contains baseline (unoptimised) LAPACK and BLAS
 libraries compiled for 64 bit Windows. The compilation was done by a third party.
 USE AT YOUR OWN RISK. The compiled versions of LAPACK and BLAS were obtained from:
   http://ylzhao.blogspot.com.au/2013/10/blas-lapack-precompiled-binaries-for.html
@@ -309,7 +318,7 @@ OpenBLAS and Intel MKL are generally the fastest replacements for both BLAS and
 
 **Caveat:** 
 for any high performance scientific/engineering workloads,
-we strongly recommend using a Linux based operating system:
+we strongly recommend using a Linux-based operating system:
   * Fedora  http://fedoraproject.org/
   * Ubuntu  http://www.ubuntu.com/
   * CentOS  http://centos.org/
@@ -322,7 +331,7 @@ Armadillo can use OpenBLAS or Intel Math Kernel Library (MKL) as high-speed
 replacements for BLAS and LAPACK. In essence this involves linking with the
 replacement libraries instead of BLAS and LAPACK.
 
-Minor modifications to include/armadillo_bits/config.hpp may be required
+Minor modifications to `include/armadillo_bits/config.hpp` may be required
 to ensure Armadillo uses the same integer sizes and style of function names
 as used by the replacement libraries. Specifically, the following defines
 may need to be enabled or disabled:
@@ -335,26 +344,27 @@ may need to be enabled or disabled:
 
 See the documentation for more information on the above defines.
 
-On Linux-based systems, MKL might be installed in a non-standard location
-such as /opt which can cause problems during linking.  Before installing
-Armadillo, the system should know where the MKL libraries are located.
-For example, /opt/intel/mkl/lib/intel64/.  This can be achieved by setting
-the LD_LIBRARY_PATH environment variable, or for a more permanent solution,
-adding the directory locations to /etc/ld.so.conf.  It may also be possible
-to store a text file with the locations in the /etc/ld.so.conf.d directory.
-For example, /etc/ld.so.conf.d/mkl.conf.  If /etc/ld.so.conf is modified
-or /etc/ld.so.conf.d/mkl.conf is created, /sbin/ldconfig must be run afterwards.
+On Linux-based systems, MKL might be installed in a non-standard location such as `/opt`
+which can cause problems during linking.
+Before installing Armadillo, the system should know where the MKL libraries are located.
+For example, `/opt/intel/mkl/lib/intel64/`.
+This can be achieved by setting the `LD_LIBRARY_PATH` environment variable,
+or for a more permanent solution, adding the directory locations to `/etc/ld.so.conf`.
+It may also be possible to store a text file with the locations
+in the `/etc/ld.so.conf.d` directory. For example, `/etc/ld.so.conf.d/mkl.conf`.
+If `/etc/ld.so.conf` is modified or `/etc/ld.so.conf.d/mkl.conf` is created,
+`/sbin/ldconfig` must be run afterwards.
 
-Below is an example of /etc/ld.so.conf.d/mkl.conf
-where Intel MKL is installed in /opt/intel
+Below is an example of `/etc/ld.so.conf.d/mkl.conf`
+where Intel MKL is installed in `/opt/intel`
 
     /opt/intel/lib/intel64  
     /opt/intel/mkl/lib/intel64  
 
 If MKL is installed and it is persistently giving problems during linking,
 Support for MKL can be disabled by editing the CMakeLists.txt file,
-deleting CMakeCache.txt and re-running the CMake based installation.
-Comment out the lines containing:
+deleting CMakeCache.txt and re-running the cmake based installation.
+Comment out the line containing:
 
     INCLUDE(ARMA_FindMKL)  
     INCLUDE(ARMA_FindACMLMP)  
@@ -364,32 +374,26 @@ Comment out the lines containing:
 
 ### 10: Support for ATLAS
 
-Armadillo can use the ATLAS library for faster versions of a subset
-of LAPACK and BLAS functions. LAPACK should still be installed to
-obtain full functionality.
-
-Caveat: the minimum recommended version of ATLAS is 3.10;
-earlier versions (such as 3.6 and 3.8) can produce incorrect
-results and/or corrupt memory, leading to random crashes.
+If OpenBLAS is not available, Armadillo can use the ATLAS library for faster versions
+of a subset of LAPACK and BLAS functions.
+LAPACK should still be installed to obtain full functionality.
+The minimum recommended version of ATLAS is 3.10.
 
 ---
 
 ### 11: Support for C++11 / C++14 Features
 
-Armadillo works with compilers supporting the older C++98 and C++03 standards,
-as well as the newer C++11 and C++14 standards.
+Armadillo 9.900 will enable extra features when a C++11/C++14 compiler is detected.
+You can also force Armadillo to make use of C++11 features by defining
+ `ARMA_USE_CXX11` before `#include <armadillo>` in your code.
+Note that you may need to explicitly enable C++11 mode in your compiler.
+For example, use the `-std=c++11` or `-std=c++14` options in gcc & clang.
 
-Armadillo will enable extra features (such as move constructors)
-when a C++11/C++14 compiler is detected. You can also force Armadillo
-to make use of C++11 features by defining ARMA_USE_CXX11 before
-`#include <armadillo>` in your code.
+**Caveat:** to make best use of C++11/C++14 features, upgrading to Armadillo 10.6 is recommended.
 
-You may need to explicitly enable C++11 mode in your compiler.
-For example, use the -std=c++11 or -std=c++14 options in gcc & clang.
-
-**Caveat:** use of the C++11 "auto" keyword is not recommended with Armadillo
+**Caveat:** use of the C++11 `auto` keyword is not recommended with Armadillo
 objects and expressions. Armadillo has a template meta-programming framework
-which creates lots of short lived temporaries that are not handled by auto.
+which creates lots of short lived temporaries that are not handled by `auto`.
 
 ---
 
@@ -399,21 +403,18 @@ Armadillo can use OpenMP to automatically speed up computationally
 expensive element-wise functions such as exp(), log(), cos(), etc.
 This requires a C++11/C++14 compiler with OpenMP 3.1+ support.
 
-When using gcc or clang, use the following options to enable both
-C++11 and OpenMP:  -std=c++11 -fopenmp
-
-Caveat: when using gcc, use of -march=native in conjunction with -fopenmp
-may lead to speed regressions on recent processors.
+For GCC and Clang compilers, use the following options to enable both C++11 and OpenMP:
+`-std=c++11 -fopenmp`
 
 ---
 
-### 13: Documentation
+### 13: Documentation of Functions and Classes
 
-The documentation for Armadillo functions and classes is available at:  
+The documentation of Armadillo functions and classes is available at:  
 http://arma.sourceforge.net/docs.html
 
-The documentation is also in the "docs.html" file in this folder,
-which can be viewed with a web browser.
+The documentation is also in the `docs.html` file distributed with Armadillo.
+Use a web browser to view it.
 
 ---
 
@@ -474,7 +475,7 @@ http://arma.sourceforge.net/faq.html
 
 ### 16: MEX Interface to Octave/Matlab
 
-The "mex_interface" folder contains examples of how to interface
+The `mex_interface` folder contains examples of how to interface
 Octave/Matlab with C++ code that uses Armadillo matrices.
 
 ---
@@ -492,3 +493,7 @@ Octave/Matlab with C++ code that uses Armadillo matrices.
 
 * RcppArmadillo: integration of Armadillo with the R system and environment  
   http://dirk.eddelbuettel.com/code/rcpp.armadillo.html
+
+* PyArmadillo: linear algebra library for Python  
+  https://pyarma.sourceforge.io
+

cmake_aux/InstallFiles/ArmadilloConfig.cmake.in

@@ -0,0 +1,16 @@
+# - Config file for the Armadillo package
+# It defines the following variables
+#  ARMADILLO_INCLUDE_DIRS - include directories for Armadillo
+#  ARMADILLO_LIBRARY_DIRS - library directories for Armadillo (normally not used!)
+#  ARMADILLO_LIBRARIES    - libraries to link against
+
+# Tell the user project where to find our headers and libraries
+set(ARMADILLO_INCLUDE_DIRS "@ARMADILLO_INCLUDE_DIRS@")
+set(ARMADILLO_LIBRARY_DIRS "@ARMADILLO_LIB_DIR@")
+
+# Our library dependencies (contains definitions for IMPORTED targets)
+include("@ARMADILLO_CMAKE_DIR@/ArmadilloLibraryDepends.cmake")
+
+# These are IMPORTED targets created by ArmadilloLibraryDepends.cmake
+set(ARMADILLO_LIBRARIES armadillo)
+

cmake_aux/InstallFiles/ArmadilloConfigVersion.cmake.in

@@ -0,0 +1,11 @@
+set(PACKAGE_VERSION "@ARMA_VERSION_MAJOR@.@ARMA_VERSION_MINOR@.@ARMA_VERSION_PATCH@")
+
+# Check whether the requested PACKAGE_FIND_VERSION is compatible
+if("${PACKAGE_VERSION}" VERSION_LESS "${PACKAGE_FIND_VERSION}")
+  set(PACKAGE_VERSION_COMPATIBLE FALSE)
+else()
+  set(PACKAGE_VERSION_COMPATIBLE TRUE)
+  if ("${PACKAGE_VERSION}" VERSION_EQUAL "${PACKAGE_FIND_VERSION}")
+    set(PACKAGE_VERSION_EXACT TRUE)
+  endif()
+endif()

docs.html

@@ -10643,7 +10643,7 @@ See also:
 <li><a href="#polyfit">polyfit()</a></li>
 <li><a href="#imag_real">real()</a></li>
 <li><a href="#solve">solve()</a></li>
-<li><a href="https://en.wikipedia.org/wiki/Zero_of_a_function">zero of a function in Wikpedia</a></li>
+<li><a href="https://en.wikipedia.org/wiki/Zero_of_a_function">zero of a function in Wikipedia</a></li>
 </ul>
 </li>
 <br>
@@ -11762,7 +11762,7 @@ fn(A) is one of:
       &nbsp;
     </td>
     <td style="vertical-align: top;">
-      natural log of the gamma function &nbsp; <font size=-1>(only for <i>float</i> and <i>double</i> elements)</font>
+      natural log of the absolute value of gamma function &nbsp; <font size=-1>(only for <i>float</i> and <i>double</i> elements)</font>
     </td>
   </tr>
   <tr>
@@ -18005,9 +18005,10 @@ Disable use of BLAS; overrides <i>ARMA_USE_BLAS</i>
       &nbsp;
     </td>
     <td style="vertical-align: top;">
-Enable use of the built-in reimplementation of ARPACK (Armadillo 7.x and later versions).
+Enable use of NEWARP (built-in alternative to ARPACK).
 This is used for the eigen decomposition of real (non-complex) sparse matrices, ie. <a href="#eigs_gen">eigs_gen()</a>, <a href="#eigs_sym">eigs_sym()</a> and <a href="#svds">svds().</a>
 Requires ARMA_USE_LAPACK to be enabled.
+If use of both NEWARP and ARPACK is enabled, NEWARP will be preferred.
     </td>
   </tr>
   <tr>
@@ -18029,7 +18030,7 @@ Requires ARMA_USE_LAPACK to be enabled.
       &nbsp;
     </td>
     <td style="vertical-align: top;">
-Disable use of the built-in reimplementation of ARPACK; overrides <i>ARMA_USE_NEWARP</i>
+Disable use of NEWARP (built-in alternative to ARPACK); overrides <i>ARMA_USE_NEWARP</i>
     </td>
   </tr>
   <tr>
@@ -18052,7 +18053,8 @@ Disable use of the built-in reimplementation of ARPACK; overrides <i>ARMA_USE_NE
     </td>
     <td style="vertical-align: top;">
 Enable use of ARPACK, or a high-speed replacement for ARPACK.
-Armadillo requires ARPACK for the eigen decomposition of complex sparse matrices, ie. <a href="#eigs_gen">eigs_gen()</a>, <a href="#eigs_sym">eigs_sym()</a> and <a href="#svds">svds()</a>
+Armadillo requires ARPACK for the eigen decomposition of complex sparse matrices, ie. <a href="#eigs_gen">eigs_gen()</a>, <a href="#eigs_sym">eigs_sym()</a> and <a href="#svds">svds()</a>.
+If use of NEWARP is disabled, ARPACK will also be used for the eigen decomposition of real sparse matrices.
     </td>
   </tr>
   <tr>

examples/README.txt

@@ -1,24 +1,24 @@
 How to compile example1.cpp
 
 
-* Linux and macOS
-  
-  If you have installed Armadillo via the CMake installer:
+** Linux and macOS ** 
+
+If you have installed Armadillo via the CMake installer:
   g++ example1.cpp -o example1 -O2 -larmadillo
   
-  Otherwise, if you want to use Armadillo without installation:
+Otherwise, if you want to use Armadillo without installation:
   g++ example1.cpp -o example1 -O2 -I /home/blah/armadillo-7.200.3/include -DARMA_DONT_USE_WRAPPER -lopenblas
   
-  The above command assumes that the armadillo archive was unpacked into /home/blah/
-  The command needs to be adjusted if the archive was unpacked into a different directory
-  and/or for each specific version of Armadillo (ie. "7.200.3" needs to be changed)
+The above command assumes that the armadillo archive was unpacked into /home/blah/
+The command needs to be adjusted if the archive was unpacked into a different directory
+and/or for each specific version of Armadillo (ie. "7.200.3" needs to be changed)
   
-  If you don't have OpenBLAS, on Linux change -lopenblas to -lblas -llapack
-  and on macOS change -lopenblas to -framework Accelerate
+If you don't have OpenBLAS, on Linux change -lopenblas to -lblas -llapack
+and on macOS change -lopenblas to -framework Accelerate
 
 
-* Windows
-  
-  Open "example1_win64.sln" or "example1_win64.vcxproj" with Visual Studio.
-  The example1_win64 project needs to be compiled as a 64 bit program.
-  Make sure the active solution platform is set to x64, instead of win32.
+** Windows **
+
+Open "example1_win64.sln" or "example1_win64.vcxproj" with Visual Studio.
+The example1_win64 project needs to be compiled as a 64 bit program.
+Make sure the active solution platform is set to x64, instead of win32.

examples/example1.cpp

@@ -14,12 +14,14 @@ main(int argc, char** argv)
   {
   cout << "Armadillo version: " << arma_version::as_string() << endl;
   
-  mat A(2,3);  // directly specify the matrix size (elements are uninitialised)
+  // construct a matrix according to given size and form of element initialisation
+  mat A(2,3,fill::zeros);
   
-  cout << "A.n_rows: " << A.n_rows << endl;  // .n_rows and .n_cols are read only
+  // .n_rows and .n_cols are read only
+  cout << "A.n_rows: " << A.n_rows << endl;
   cout << "A.n_cols: " << A.n_cols << endl;
   
-  A(1,2) = 456.0;  // directly access an element (indexing starts at 0)
+  A(1,2) = 456.0;  // access an element (indexing starts at 0)
   A.print("A:");
   
   A = 5.0;         // scalars are treated as a 1x1 matrix
@@ -27,7 +29,7 @@ main(int argc, char** argv)
   
   A.set_size(4,5); // change the size (data is not preserved)
   
-  A.fill(5.0);     // set all elements to a particular value
+  A.fill(5.0);     // set all elements to a specific value
   A.print("A:");
   
   // endr indicates "end of row"

include/armadillo_bits/BaseCube_meat.hpp

@@ -187,15 +187,15 @@ BaseCube<elem_type,derived>::is_zero(const typename get_pod_type<elem_type>::res
       const T val_real = access::tmp_real(val);
       const T val_imag = access::tmp_imag(val);
       
-      if(std::abs(val_real) > tol)  { return false; }
-      if(std::abs(val_imag) > tol)  { return false; }
+      if(eop_aux::arma_abs(val_real) > tol)  { return false; }
+      if(eop_aux::arma_abs(val_imag) > tol)  { return false; }
       }
     }
   else  // not complex
     {
     for(uword i=0; i < n_elem; ++i)
       {
-      if(std::abs(Pea[i]) > tol)  { return false; }
+      if(eop_aux::arma_abs(Pea[i]) > tol)  { return false; }
       }
     }
   

include/armadillo_bits/Base_bones.hpp

@@ -22,7 +22,7 @@
 template<typename elem_type, typename derived>
 struct Base_extra_yes
   {
-  arma_inline const Op<derived,op_inv> i() const;   //!< matrix inverse
+  inline arma_warn_unused const Op<derived,op_inv> i() const;   //!< matrix inverse
   
   arma_deprecated inline const Op<derived,op_inv> i(const bool ) const;   //!< kept only for compatibility with old user code
   arma_deprecated inline const Op<derived,op_inv> i(const char*) const;   //!< kept only for compatibility with old user code
@@ -154,8 +154,8 @@ struct Base
   inline arma_warn_unused bool has_inf()    const;
   inline arma_warn_unused bool has_nan()    const;
   
-  arma_inline const Op<derived,op_vectorise_col> as_col() const;
-  arma_inline const Op<derived,op_vectorise_row> as_row() const;
+  inline arma_warn_unused const Op<derived,op_vectorise_col> as_col() const;
+  inline arma_warn_unused const Op<derived,op_vectorise_row> as_row() const;
   };
 
 

include/armadillo_bits/Base_meat.hpp

@@ -411,15 +411,15 @@ Base<elem_type,derived>::is_zero(const typename get_pod_type<elem_type>::result
       const T val_real = access::tmp_real(val);
       const T val_imag = access::tmp_imag(val);
       
-      if(std::abs(val_real) > tol)  { return false; }
-      if(std::abs(val_imag) > tol)  { return false; }
+      if(eop_aux::arma_abs(val_real) > tol)  { return false; }
+      if(eop_aux::arma_abs(val_imag) > tol)  { return false; }
       }
     }
   else  // not complex
     {
     for(uword i=0; i<n_elem; ++i)
       {
-      if(std::abs(Pea[i]) > tol)  { return false; }
+      if(eop_aux::arma_abs(Pea[i]) > tol)  { return false; }
       }
     }
   
@@ -723,7 +723,8 @@ Base<elem_type,derived>::has_nan() const
 
 
 template<typename elem_type, typename derived>
-arma_inline
+inline
+arma_warn_unused
 const Op<derived,op_vectorise_col>
 Base<elem_type, derived>::as_col() const
   {
@@ -733,7 +734,8 @@ Base<elem_type, derived>::as_col() const
 
 
 template<typename elem_type, typename derived>
-arma_inline
+inline
+arma_warn_unused
 const Op<derived,op_vectorise_row>
 Base<elem_type, derived>::as_row() const
   {
@@ -746,7 +748,8 @@ Base<elem_type, derived>::as_row() const
 // extra functions defined in Base_extra_yes
 
 template<typename elem_type, typename derived>
-arma_inline
+inline
+arma_warn_unused
 const Op<derived,op_inv>
 Base_extra_yes<elem_type, derived>::i() const
   {

include/armadillo_bits/Cube_meat.hpp

@@ -1034,6 +1034,8 @@ Cube<eT>::slice(const uword in_slice)
   
   arma_debug_check( (in_slice >= n_slices), "Cube::slice(): index out of bounds" );
   
+  // BUG: data race when accessed from multiple threads; causes memory leakage
+  
   if(mat_ptrs[in_slice] == NULL)
     {
     const eT* ptr = (n_elem_slice > 0) ? slice_memptr(in_slice) : NULL;
@@ -1056,6 +1058,8 @@ Cube<eT>::slice(const uword in_slice) const
   
   arma_debug_check( (in_slice >= n_slices), "Cube::slice(): index out of bounds" );
   
+  // BUG: data race when accessed from multiple threads; causes memory leakage
+  
   if(mat_ptrs[in_slice] == NULL)
     {
     const eT* ptr = (n_elem_slice > 0) ? slice_memptr(in_slice) : NULL;

include/armadillo_bits/Mat_meat.hpp

@@ -2671,12 +2671,12 @@ Mat<eT>::operator/=(const SpBase<eT, T1>& m)
   {
   arma_extra_debug_sigprint();
   
+  // NOTE: use of this function is not advised; it is implemented only for completeness 
+  
   const SpProxy<T1> p(m.get_ref());
   
   arma_debug_assert_same_size(n_rows, n_cols, p.get_n_rows(), p.get_n_cols(), "element-wise division");
   
-  // If you use this method, you are probably stupid or misguided, but for completeness it is implemented.
-  // Unfortunately the best way to do this is loop over every element.
   for(uword c = 0; c < n_cols; ++c)
   for(uword r = 0; r < n_rows; ++r)
     {
@@ -6902,7 +6902,7 @@ Mat<eT>::fill(const eT val)
 
 
 
-//! fill the matrix with the specified value
+//! fill the matrix with the specified pattern
 template<typename eT>
 template<typename fill_type>
 inline

include/armadillo_bits/SpBase_bones.hpp

@@ -100,9 +100,8 @@ struct SpBase
   inline arma_warn_unused bool has_inf()    const;
   inline arma_warn_unused bool has_nan()    const;
   
-  inline const SpOp<derived,spop_vectorise_col> as_col() const;
-  inline const SpOp<derived,spop_vectorise_row> as_row() const;
-  
+  inline arma_warn_unused const SpOp<derived,spop_vectorise_col> as_col() const;
+  inline arma_warn_unused const SpOp<derived,spop_vectorise_row> as_row() const;
   };
 
 

include/armadillo_bits/SpBase_meat.hpp

@@ -435,8 +435,8 @@ SpBase<elem_type,derived>::is_zero(const typename get_pod_type<elem_type>::resul
       const T val_real = access::tmp_real(val);
       const T val_imag = access::tmp_imag(val);
       
-      if(std::abs(val_real) > tol)  { return false; }
-      if(std::abs(val_imag) > tol)  { return false; }
+      if(eop_aux::arma_abs(val_real) > tol)  { return false; }
+      if(eop_aux::arma_abs(val_imag) > tol)  { return false; }
       
       ++it;
       }
@@ -445,7 +445,7 @@ SpBase<elem_type,derived>::is_zero(const typename get_pod_type<elem_type>::resul
     {
     while(it != it_end)
       {
-      if(std::abs(*it) > tol)  { return false; }
+      if(eop_aux::arma_abs(*it) > tol)  { return false; }
       
       ++it;
       }
@@ -714,6 +714,7 @@ SpBase<elem_type,derived>::has_nan() const
 
 template<typename elem_type, typename derived>
 inline
+arma_warn_unused
 const SpOp<derived,spop_vectorise_col>
 SpBase<elem_type, derived>::as_col() const
   {
@@ -724,6 +725,7 @@ SpBase<elem_type, derived>::as_col() const
 
 template<typename elem_type, typename derived>
 inline
+arma_warn_unused
 const SpOp<derived,spop_vectorise_row>
 SpBase<elem_type, derived>::as_row() const
   {

include/armadillo_bits/SpMat_meat.hpp

@@ -782,16 +782,14 @@ SpMat<eT>::operator/=(const SpMat<eT>& x)
   {
   arma_extra_debug_sigprint();
   
+  // NOTE: use of this function is not advised; it is implemented only for completeness
+  
   arma_debug_assert_same_size(n_rows, n_cols, x.n_rows, x.n_cols, "element-wise division");
   
-  // If you use this method, you are probably stupid or misguided,
-  // but for compatibility with Mat, we have implemented it anyway.
   for(uword c = 0; c < n_cols; ++c)
+  for(uword r = 0; r < n_rows; ++r)
     {
-    for(uword r = 0; r < n_rows; ++r)
-      {
-      at(r, c) /= x.at(r, c);
-      }
+    at(r, c) /= x.at(r, c);
     }
   
   return *this;

include/armadillo_bits/SpSubview_meat.hpp

@@ -475,7 +475,6 @@ SpSubview<eT>::operator%=(const SpBase<eT, T1>& x)
 
 
 
-//! If you are using this function, you are probably misguided.
 template<typename eT>
 template<typename T1>
 inline
@@ -484,6 +483,8 @@ SpSubview<eT>::operator/=(const SpBase<eT, T1>& x)
   {
   arma_extra_debug_sigprint();
   
+  // NOTE: use of this function is not advised; it is implemented only for completeness
+  
   SpProxy<T1> p(x.get_ref());
   
   arma_debug_assert_same_size(n_rows, n_cols, p.get_n_rows(), p.get_n_cols(), "element-wise division");
@@ -1017,7 +1018,7 @@ inline
 bool
 SpSubview<eT>::check_overlap(const SpSubview<eT>& x) const
   {
-  const subview<eT>& t = *this;
+  const SpSubview<eT>& t = *this;
 
   if(&t.m != &x.m)
     {

include/armadillo_bits/arma_cmath.hpp

@@ -358,7 +358,7 @@ arma_inline
 typename arma_real_only<eT>::result
 arma_sign(const eT x)
   {
-  return (x > eT(0)) ? eT(+1) : ( (x < eT(0)) ? eT(-1) : eT(0) );
+  return (x > eT(0)) ? eT(+1) : ( (x < eT(0)) ? eT(-1) : ((x == eT(0)) ? eT(0) : x) );
   }
 
 

include/armadillo_bits/arma_rng.hpp

@@ -342,7 +342,7 @@ struct arma_rng::randn
     }
   
   
-  arma_inline
+  inline
   static
   void
   dual_val(eT& out1, eT& out2)
@@ -457,6 +457,27 @@ struct arma_rng::randn< std::complex<T> >
     }
   
   
+  inline
+  static
+  void
+  dual_val(std::complex<T>& out1, std::complex<T>& out2)
+    {
+    #if defined(_MSC_VER)
+      T a;
+      T b;
+    #else
+      T a(0);
+      T b(0);
+    #endif
+    
+    arma_rng::randn<T>::dual_val(a,b);
+    out1 = std::complex<T>(a,b);
+    
+    arma_rng::randn<T>::dual_val(a,b);
+    out2 = std::complex<T>(a,b);
+    }
+  
+  
   inline
   static
   void

include/armadillo_bits/arma_version.hpp

@@ -21,7 +21,7 @@
 
 #define ARMA_VERSION_MAJOR 9
 #define ARMA_VERSION_MINOR 900
-#define ARMA_VERSION_PATCH 4
+#define ARMA_VERSION_PATCH 6
 #define ARMA_VERSION_NAME  "Nocturnal Misbehaviour"
 
 

include/armadillo_bits/arrayops_meat.hpp

@@ -135,7 +135,7 @@ arrayops::clean(eT* mem, const uword n_elem, const eT abs_limit, const typename
     {
     eT& val = mem[i];
     
-    val = (std::abs(val) <= abs_limit) ? eT(0) : val;
+    val = (eop_aux::arma_abs(val) <= abs_limit) ? eT(0) : val;
     }
   }
 
@@ -1028,7 +1028,7 @@ arrayops::is_zero(const eT* mem, const uword n_elem, const eT abs_limit, const t
     {
     for(uword i=0; i<n_elem; ++i)
       {
-      if(std::abs(mem[i]) > abs_limit)  { return false; }
+      if(eop_aux::arma_abs(mem[i]) > abs_limit)  { return false; }
       }
     }
   

include/armadillo_bits/config.hpp

@@ -307,6 +307,10 @@
   #undef ARMA_CRIPPLED_LAPACK
 #endif
 
+#if defined(ARMA_WARN_LEVEL)
+  #pragma message ("WARNING: the ARMA_WARN_LEVEL option is not supported in this version of Armadillo")
+#endif
+
 
 // if Armadillo was installed on this system via CMake and ARMA_USE_WRAPPER is not defined,
 // ARMA_AUX_LIBS lists the libraries required by Armadillo on this system, and

include/armadillo_bits/config.hpp.cmake

@@ -307,6 +307,10 @@
   #undef ARMA_CRIPPLED_LAPACK
 #endif
 
+#if defined(ARMA_WARN_LEVEL)
+  #pragma message ("WARNING: the ARMA_WARN_LEVEL option is not supported in this version of Armadillo")
+#endif
+
 
 // if Armadillo was installed on this system via CMake and ARMA_USE_WRAPPER is not defined,
 // ARMA_AUX_LIBS lists the libraries required by Armadillo on this system, and

include/armadillo_bits/constants.hpp

@@ -113,7 +113,7 @@ namespace priv
 
 
 //! various constants.
-//! Physical constants taken from NIST 2014 CODATA values, and some from WolframAlpha (values provided as of 2009-06-23)
+//! Physical constants taken from NIST 2018 CODATA values, and some from WolframAlpha (values provided as of 2009-06-23)
 //! http://physics.nist.gov/cuu/Constants
 //! http://www.wolframalpha.com
 //! See also http://en.wikipedia.org/wiki/Physical_constant

include/armadillo_bits/diagmat_proxy.hpp

@@ -90,7 +90,7 @@ class diagmat_proxy_default
     }
   
   
-  arma_inline bool is_alias(const Mat<elem_type>&) const { return false; }
+  inline bool is_alias(const Mat<elem_type>& X) const { return P.is_alias(X); }
   
   const Proxy<T1> P;
   const bool      P_is_vec;

include/armadillo_bits/diskio_meat.hpp

@@ -4377,7 +4377,7 @@ diskio::load_ppm_binary(Cube<eT>& x, std::istream& f, std::string& err_msg)
     f >> f_maxval;
     f.get();
     
-    if( (f_maxval > 0) || (f_maxval <= 65535) )
+    if( (f_maxval > 0) && (f_maxval <= 65535) )
       {
       x.set_size(f_n_rows, f_n_cols, 3);
       
@@ -4566,7 +4566,7 @@ diskio::load_ppm_binary(field<T1>& x, std::istream& f, std::string& err_msg)
     f >> f_maxval;
     f.get();
     
-    if( (f_maxval > 0) || (f_maxval <= 65535) )
+    if( (f_maxval > 0) && (f_maxval <= 65535) )
       {
       x.set_size(3);
       Mat<eT>& R = x(0);

include/armadillo_bits/eop_aux.hpp

@@ -198,7 +198,7 @@ class eop_aux
   
   
   template<typename T1, typename T2> arma_inline static typename   arma_integral_only<T1>::result pow (const T1 base, const T2 exponent) { return T1( std::pow( double(base), double(exponent) ) ); }
-  template<typename T1, typename T2> arma_inline static typename arma_real_or_cx_only<T1>::result pow (const T1 base, const T2 exponent) { return std::pow(base, exponent);                         }
+  template<typename T1, typename T2> arma_inline static typename arma_real_or_cx_only<T1>::result pow (const T1 base, const T2 exponent) { return T1( std::pow(        base,         exponent  ) ); }
   
   
   template<typename eT>

include/armadillo_bits/fn_log_det.hpp

@@ -44,7 +44,7 @@ log_det
     out_val  = eT(Datum<T>::nan);
     out_sign = T(0);
     
-    arma_warn("log_det(): failed to find determinant");
+    arma_debug_warn("log_det(): failed to find determinant");
     }
   }
 

include/armadillo_bits/fn_mvnrnd.hpp

@@ -56,7 +56,6 @@ mvnrnd(const Base<typename T1::elem_type, T1>& M, const Base<typename T1::elem_t
 
 
 template<typename T1, typename T2>
-arma_warn_unused
 inline
 typename
 enable_if2
@@ -82,7 +81,6 @@ mvnrnd(Mat<typename T1::elem_type>& out, const Base<typename T1::elem_type, T1>&
 
 
 template<typename T1, typename T2>
-arma_warn_unused
 inline
 typename
 enable_if2

include/armadillo_bits/fn_svd.hpp

@@ -157,7 +157,7 @@ svd_econ
     U.soft_reset();
     S.soft_reset();
     V.soft_reset();
-    arma_debug_warn("svd(): decomposition failed");
+    arma_debug_warn("svd_econ(): decomposition failed");
     }
   
   return status;

include/armadillo_bits/fn_wishrnd.hpp

@@ -82,7 +82,6 @@ wishrnd(Mat<typename T1::elem_type>& W, const Base<typename T1::elem_type, T1>&
 
 
 template<typename T1, typename T2>
-arma_warn_unused
 inline
 typename
 enable_if2
@@ -167,7 +166,6 @@ iwishrnd(Mat<typename T1::elem_type>& W, const Base<typename T1::elem_type, T1>&
 
 
 template<typename T1, typename T2>
-arma_warn_unused
 inline
 typename
 enable_if2

include/armadillo_bits/glue_max_meat.hpp

@@ -58,7 +58,7 @@ glue_max::apply(Mat<eT>& out, const Proxy<T1>& PA, const Proxy<T2>& PB)
   const uword n_rows = PA.get_n_rows();
   const uword n_cols = PA.get_n_cols();
   
-  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise maximum");
+  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise max()");
   
   out.set_size(n_rows, n_cols);
   
@@ -102,7 +102,7 @@ glue_max::apply(Mat< std::complex<T> >& out, const Proxy<T1>& PA, const Proxy<T2
   const uword n_rows = PA.get_n_rows();
   const uword n_cols = PA.get_n_cols();
   
-  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise maximum");
+  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise max()");
   
   out.set_size(n_rows, n_cols);
   
@@ -179,7 +179,7 @@ glue_max::apply(Cube<eT>& out, const ProxyCube<T1>& PA, const ProxyCube<T2>& PB)
   const uword n_cols   = PA.get_n_cols();
   const uword n_slices = PA.get_n_slices();
   
-  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise maximum");
+  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise max()");
   
   out.set_size(n_rows, n_cols, n_slices);
   
@@ -225,7 +225,7 @@ glue_max::apply(Cube< std::complex<T> >& out, const ProxyCube<T1>& PA, const Pro
   const uword n_cols   = PA.get_n_cols();
   const uword n_slices = PA.get_n_slices();
   
-  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise maximum");
+  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise max()");
   
   out.set_size(n_rows, n_cols, n_slices);
   

include/armadillo_bits/glue_min_meat.hpp

@@ -58,7 +58,7 @@ glue_min::apply(Mat<eT>& out, const Proxy<T1>& PA, const Proxy<T2>& PB)
   const uword n_rows = PA.get_n_rows();
   const uword n_cols = PA.get_n_cols();
   
-  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise minimum");
+  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise min()");
   
   out.set_size(n_rows, n_cols);
   
@@ -102,7 +102,7 @@ glue_min::apply(Mat< std::complex<T> >& out, const Proxy<T1>& PA, const Proxy<T2
   const uword n_rows = PA.get_n_rows();
   const uword n_cols = PA.get_n_cols();
   
-  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise minimum");
+  arma_debug_assert_same_size(n_rows, n_cols, PB.get_n_rows(), PB.get_n_cols(), "element-wise min()");
   
   out.set_size(n_rows, n_cols);
   
@@ -179,7 +179,7 @@ glue_min::apply(Cube<eT>& out, const ProxyCube<T1>& PA, const ProxyCube<T2>& PB)
   const uword n_cols   = PA.get_n_cols();
   const uword n_slices = PA.get_n_slices();
   
-  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise minimum");
+  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise min()");
   
   out.set_size(n_rows, n_cols, n_slices);
   
@@ -225,7 +225,7 @@ glue_min::apply(Cube< std::complex<T> >& out, const ProxyCube<T1>& PA, const Pro
   const uword n_cols   = PA.get_n_cols();
   const uword n_slices = PA.get_n_slices();
   
-  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise minimum");
+  arma_debug_assert_same_size(n_rows, n_cols, n_slices, PB.get_n_rows(), PB.get_n_cols(), PB.get_n_slices(), "element-wise min()");
   
   out.set_size(n_rows, n_cols, n_slices);
   

include/armadillo_bits/gmm_diag_meat.hpp

@@ -2157,6 +2157,10 @@ gmm_diag<eT>::km_iterate(const Mat<eT>& X, const uword max_iter, const bool verb
       }
     #else
       {
+      acc_hefts.zeros();
+      acc_means.zeros();
+      last_indx.zeros();
+      
       uword* acc_hefts_mem = acc_hefts.memptr();
       uword* last_indx_mem = last_indx.memptr();
       

include/armadillo_bits/gmm_full_meat.hpp

@@ -2186,6 +2186,10 @@ gmm_full<eT>::km_iterate(const Mat<eT>& X, const uword max_iter, const bool verb
       }
     #else
       {
+      acc_hefts.zeros();
+      acc_means.zeros();
+      last_indx.zeros();
+      
       uword* acc_hefts_mem = acc_hefts.memptr();
       uword* last_indx_mem = last_indx.memptr();
       

include/armadillo_bits/op_cond_meat.hpp

@@ -58,11 +58,13 @@ op_cond::rcond(const Base<typename T1::elem_type, T1>& X)
     {
     const strip_trimat<T1> S(X.get_ref());
     
-    arma_debug_check( (S.M.is_square() == false), "rcond(): matrix must be square sized" );
+    const quasi_unwrap<typename strip_trimat<T1>::stored_type> U(S.M);
+    
+    arma_debug_check( (U.M.is_square() == false), "rcond(): matrix must be square sized" );
     
     const uword layout = (S.do_triu) ? uword(0) : uword(1);
     
-    return auxlib::rcond_trimat(S.M, layout);
+    return auxlib::rcond_trimat(U.M, layout);
     }
   
   Mat<eT> A = X.get_ref();

include/armadillo_bits/op_hist_meat.hpp

@@ -58,6 +58,12 @@ op_hist::apply_noalias(Mat<uword>& out, const Mat<eT>& A, const uword n_bins, co
     if(max_val < val_i) { max_val = val_i; }
     }
   
+  if(min_val == max_val)
+    {
+    min_val -= (n_bins/2);
+    max_val += (n_bins/2);
+    }
+  
   if(arma_isfinite(min_val) == false) { min_val = priv::most_neg<eT>(); }
   if(arma_isfinite(max_val) == false) { max_val = priv::most_pos<eT>(); }
   
@@ -66,7 +72,7 @@ op_hist::apply_noalias(Mat<uword>& out, const Mat<eT>& A, const uword n_bins, co
   
   for(uword ii=0; ii < n_bins; ++ii)
     {
-    c_mem[ii] = (0.5 + ii) / double(n_bins);   // TODO: may need to be modified for integer matrices
+    c_mem[ii] = (0.5 + ii) / double(n_bins);
     }
   
   c = ((max_val - min_val) * c) + min_val;
@@ -89,17 +95,26 @@ op_hist::apply(Mat<uword>& out, const mtOp<uword, T1, op_hist>& X)
   
   const uword dim = (T1::is_xvec) ? uword(U.M.is_rowvec() ? 1 : 0) : uword((T1::is_row) ? 1 : 0);
   
-  if(U.is_alias(out))
+  if(is_non_integral<typename T1::elem_type>::value)
     {
-    Mat<uword> tmp;
-    
-    op_hist::apply_noalias(tmp, U.M, n_bins, dim);
-    
-    out.steal_mem(tmp);
+    if(U.is_alias(out))
+      {
+      Mat<uword> tmp;
+      
+      op_hist::apply_noalias(tmp, U.M, n_bins, dim);
+      
+      out.steal_mem(tmp);
+      }
+    else
+      {
+      op_hist::apply_noalias(out, U.M, n_bins, dim);
+      }
     }
   else
     {
-    op_hist::apply_noalias(out, U.M, n_bins, dim);
+    Mat<double> converted = conv_to< Mat<double> >::from(U.M);
+    
+    op_hist::apply_noalias(out, converted, n_bins, dim);
     }
   }
 

include/armadillo_bits/op_index_max_meat.hpp

@@ -98,7 +98,7 @@ op_index_max::apply_noalias(Mat<uword>& out, const Mat<eT>& X, const uword dim)
       
       for(uword row=0; row < X_n_rows; ++row)
         {
-        tmp_mem[row] = std::abs(col_mem[row]);
+        tmp_mem[row] = eop_aux::arma_abs(col_mem[row]);
         }
       }
     else
@@ -113,7 +113,7 @@ op_index_max::apply_noalias(Mat<uword>& out, const Mat<eT>& X, const uword dim)
       for(uword row=0; row < X_n_rows; ++row)
         {
         T& max_val = tmp_mem[row];
-        T  col_val = (is_cx<eT>::yes) ? T(std::abs(col_mem[row])) : T(access::tmp_real(col_mem[row]));
+        T  col_val = (is_cx<eT>::yes) ? T(eop_aux::arma_abs(col_mem[row])) : T(access::tmp_real(col_mem[row]));
         
         if(max_val < col_val)
           {

include/armadillo_bits/op_index_min_meat.hpp

@@ -98,7 +98,7 @@ op_index_min::apply_noalias(Mat<uword>& out, const Mat<eT>& X, const uword dim)
       
       for(uword row=0; row < X_n_rows; ++row)
         {
-        tmp_mem[row] = std::abs(col_mem[row]);
+        tmp_mem[row] = eop_aux::arma_abs(col_mem[row]);
         }
       }
     else
@@ -113,7 +113,7 @@ op_index_min::apply_noalias(Mat<uword>& out, const Mat<eT>& X, const uword dim)
       for(uword row=0; row < X_n_rows; ++row)
         {
         T& min_val = tmp_mem[row];
-        T  col_val = (is_cx<eT>::yes) ? T(std::abs(col_mem[row])) : T(access::tmp_real(col_mem[row]));
+        T  col_val = (is_cx<eT>::yes) ? T(eop_aux::arma_abs(col_mem[row])) : T(access::tmp_real(col_mem[row]));
         
         if(min_val > col_val)
           {

include/armadillo_bits/op_median_bones.hpp

@@ -43,11 +43,11 @@ class op_median
   {
   public:
   
-  template<typename T1>
-  inline static void apply(Mat<typename T1::elem_type>& out, const Op<T1,op_median>& in);
+  template<typename eT, typename T1>
+  inline static void apply(Mat<eT>& out, const Op<T1,op_median>& in, const typename arma_not_cx<eT>::result* junk = 0);
   
-  template<typename T, typename T1>
-  inline static void apply(Mat< std::complex<T> >& out, const Op<T1,op_median>& in);
+  template<typename eT, typename T1>
+  inline static void apply(Mat<eT>& out, const Op<T1,op_median>& in, const typename arma_cx_only<eT>::result* junk = 0);
   
   //
   //

include/armadillo_bits/op_median_meat.hpp

@@ -23,14 +23,15 @@
 //! For each row or for each column, find the median value.
 //! The result is stored in a dense matrix that has either one column or one row.
 //! The dimension, for which the medians are found, is set via the median() function.
-template<typename T1>
+template<typename eT, typename T1>
 inline
 void
-op_median::apply(Mat<typename T1::elem_type>& out, const Op<T1,op_median>& in)
+op_median::apply(Mat<eT>& out, const Op<T1,op_median>& in, const typename arma_not_cx<eT>::result* junk)
   {
   arma_extra_debug_sigprint();
+  arma_ignore(junk);
   
-  typedef typename T1::elem_type eT;
+  // typedef typename T1::elem_type eT;
   
   const uword dim = in.aux_uword_a;
   arma_debug_check( (dim > 1), "median(): parameter 'dim' must be 0 or 1" );
@@ -41,7 +42,7 @@ op_median::apply(Mat<typename T1::elem_type>& out, const Op<T1,op_median>& in)
   
   const bool is_alias = P.is_alias(out);
   
-  if( (is_Mat<P_stored_type>::value == true) || is_alias )
+  if(is_Mat<P_stored_type>::value || is_alias)
     {
     const unwrap_check<P_stored_type> tmp(P.Q, is_alias);
     
@@ -134,14 +135,16 @@ op_median::apply(Mat<typename T1::elem_type>& out, const Op<T1,op_median>& in)
 
 
 //! Implementation for complex numbers
-template<typename T, typename T1>
+template<typename eT, typename T1>
 inline
 void
-op_median::apply(Mat< std::complex<T> >& out, const Op<T1,op_median>& in)
+op_median::apply(Mat<eT>& out, const Op<T1,op_median>& in, const typename arma_cx_only<eT>::result* junk)
   {
   arma_extra_debug_sigprint();
+  arma_ignore(junk);
   
-  typedef typename std::complex<T> eT;
+  // typedef typename std::complex<T> eT;
+  typedef typename get_pod_type<eT>::result T;
   
   arma_type_check(( is_same_type<eT, typename T1::elem_type>::no ));
   
@@ -242,7 +245,7 @@ op_median::median_vec
   
   std::vector<eT> tmp_vec(n_elem);
   
-  if(is_Mat<P_stored_type>::value == true)
+  if(is_Mat<P_stored_type>::value)
     {
     const unwrap<P_stored_type> tmp(P.Q);
     

include/armadillo_bits/op_pinv_meat.hpp

@@ -119,7 +119,7 @@ op_pinv::apply_direct(Mat<typename T1::elem_type>& out, const Base<typename T1::
       {
       const T val = s_mem[i];
       
-      if(val >= tol)  {  s2_mem[count2] = T(1) / val;  ++count2; }
+      if(val >= tol)  { s2_mem[count2] = (val > T(0)) ? T(T(1) / val) : T(0); ++count2; }
       }
     
     

include/armadillo_bits/op_princomp_meat.hpp

@@ -264,6 +264,7 @@ op_princomp::direct_princomp
   arma_extra_debug_sigprint();
   
   typedef typename T1::elem_type eT;
+  typedef typename T1::pod_type   T;
   
   const unwrap<T1>    Y( X.get_ref() );
   const Mat<eT>& in = Y.M;
@@ -274,7 +275,7 @@ op_princomp::direct_princomp
     
     // singular value decomposition
     Mat<eT> U;
-    Col<eT> s;
+    Col< T> s;
     
     const bool svd_ok = (in.n_rows >= in.n_cols) ? svd_econ(U, s, coeff_out, tmp) : svd(U, s, coeff_out, tmp);
     

include/armadillo_bits/sp_auxlib_meat.hpp

@@ -351,7 +351,7 @@ sp_auxlib::eigs_gen_newarp(Col< std::complex<T> >& eigval, Mat< std::complex<T>
     
     const newarp::SparseGenMatProd<T> op(X);
     
-    arma_debug_check( (op.n_rows != op.n_cols), "eigs_sym(): given matrix must be square sized" );
+    arma_debug_check( (op.n_rows != op.n_cols), "eigs_gen(): given matrix must be square sized" );
     
     arma_debug_check( (n_eigvals + 1 >= op.n_rows), "eigs_gen(): n_eigvals + 1 must be less than the number of rows in the matrix" );
     
@@ -1146,7 +1146,7 @@ sp_auxlib::spsolve_refine(Mat<typename T1::elem_type>& X, typename T1::pod_type&
     
     // We have to actually create the object which stores the data.
     // This gets cleaned by destroy_supermatrix().
-    // We have to use SuperLU's stupid memory allocation routines since they are
+    // We have to use SuperLU's problematic memory allocation routines since they are
     // not guaranteed to be new and delete.  See the comments in def_superlu.hpp
     superlu::NCformat* nc = (superlu::NCformat*)superlu::malloc(sizeof(superlu::NCformat));
     

include/armadillo_bits/spglue_merge_meat.hpp

@@ -473,6 +473,8 @@ spglue_merge::diagview_merge(SpMat<eT>& out, const SpMat<eT>& A, const SpMat<eT>
   {
   arma_extra_debug_sigprint();
   
+  // NOTE: assuming that B has non-zero elements only on the main diagonal
+  
   out.reserve(A.n_rows, A.n_cols, A.n_nonzero + B.n_nonzero); // worst case scenario
   
   typename SpMat<eT>::const_iterator x_it  = A.begin();
@@ -485,7 +487,7 @@ spglue_merge::diagview_merge(SpMat<eT>& out, const SpMat<eT>& A, const SpMat<eT>
   
   while( (x_it != x_end) || (y_it != y_end) )
     {
-    eT out_val;
+    eT out_val = eT(0);
     
     const uword x_it_col = x_it.col();
     const uword x_it_row = x_it.row();
@@ -508,28 +510,29 @@ spglue_merge::diagview_merge(SpMat<eT>& out, const SpMat<eT>& A, const SpMat<eT>
       {
       if((x_it_col < y_it_col) || ((x_it_col == y_it_col) && (x_it_row < y_it_row))) // if y is closer to the end
         {
-        out_val = (*x_it);
+        if(x_it_col != x_it_row)  { out_val = (*x_it); }  // don't take values from the main diagonal of A
         
         ++x_it;
         }
       else
         {
-        out_val = (*y_it);
+        if(y_it_col == y_it_row)  { out_val = (*y_it); use_y_loc = true; }  // take values only from the main diagonal of B
         
         ++y_it;
-        
-        use_y_loc = true;
         }
       }
     
-    access::rw(out.values[count]) = out_val;
-    
-    const uword out_row = (use_y_loc == false) ? x_it_row : y_it_row;
-    const uword out_col = (use_y_loc == false) ? x_it_col : y_it_col;
-    
-    access::rw(out.row_indices[count]) = out_row;
-    access::rw(out.col_ptrs[out_col + 1])++;
-    ++count;
+    if(out_val != eT(0))
+      {
+      access::rw(out.values[count]) = out_val;
+      
+      const uword out_row = (use_y_loc == false) ? x_it_row : y_it_row;
+      const uword out_col = (use_y_loc == false) ? x_it_col : y_it_col;
+      
+      access::rw(out.row_indices[count]) = out_row;
+      access::rw(out.col_ptrs[out_col + 1])++;
+      ++count;
+      }
     }
   
   const uword out_n_cols = out.n_cols;

include/armadillo_bits/spop_normalise_meat.hpp

@@ -50,7 +50,7 @@ spop_normalise::apply_direct(SpMat<eT>& out, const SpMat<eT>& X, const uword p,
   
   X.sync();
   
-  if( X.is_empty() || (X.n_nonzero == 0) )  { return; }
+  if( X.is_empty() || (X.n_nonzero == 0) )  { out.zeros(X.n_rows, X.n_cols); return; }
   
   if(dim == 0)
     {

mex_interface/armaMex.hpp

@@ -708,7 +708,9 @@ armaSetSparsePr(mxArray *matlabMatrix, const SpMat<double>& armaMatrix)
   double  *sr  = mxGetPr(matlabMatrix);
   mwIndex *irs = mxGetIr(matlabMatrix);
   mwIndex *jcs = mxGetJc(matlabMatrix);
-    
+  
+  armaMatrix.sync();
+  
   mwSize n_nonzero = armaMatrix.n_nonzero;
   mwSize n_cols    = armaMatrix.n_cols;
 
@@ -732,7 +734,9 @@ armaSetSparsePi(mxArray *matlabMatrix, const SpMat<double>& armaMatrix)
   double  *si  = mxGetPi(matlabMatrix);
   mwIndex *irs = mxGetIr(matlabMatrix);
   mwIndex *jcs = mxGetJc(matlabMatrix);
-    
+  
+  armaMatrix.sync();
+  
   mwSize n_nonzero = armaMatrix.n_nonzero;
   mwSize n_cols    = armaMatrix.n_cols;
 

tests2/Makefile

@@ -0,0 +1,27 @@
+
+LIB_FLAGS = -larmadillo
+#LIB_FLAGS = -lblas -llapack 
+#LIB_FLAGS = -lopenblas -llapack 
+
+CXX_FLAGS = -std=c++11 -Wshadow -Wall -pedantic -O0
+#CXX_FLAGS = -std=c++11 -Wshadow -Wall -pedantic -O0 -fopenmp
+#CXX_FLAGS = -std=c++11 -Wshadow -Wall -pedantic -O0 -DARMA_DONT_USE_WRAPPER
+#CXX_FLAGS = -std=c++11 -Wshadow -Wall -pedantic -O0 -fsanitize=address -fsanitize=leak -fsanitize=undefined -fsanitize=bounds -fsanitize=bounds-strict -g
+#CXX_FLAGS = -std=c++11 -Wshadow -Wall -pedantic -O2
+
+
+OBJECTS = $(patsubst %.cpp,%.o,$(wildcard *.cpp))
+
+%.o: %.cpp $(DEPS)
+	$(CXX) $(CXX_FLAGS) -o $@ -c $<
+
+main: $(OBJECTS)
+	$(CXX) $(CXX_FLAGS) -o $@ $(OBJECTS) $(LIB_FLAGS)
+
+
+all: main
+
+.PHONY: clean
+
+clean:
+	rm -f main *.o

tests2/catch.hpp

@@ -1,6 +1,6 @@
 /*
- *  Catch v1.10.0
- *  Generated: 2017-08-26 15:16:46.676990
+ *  Catch v1.12.2
+ *  Generated: 2018-05-14 15:10:01.112442
  *  ----------------------------------------------------------
  *  This file has been merged from multiple headers. Please don't edit it directly
  *  Copyright (c) 2012 Two Blue Cubes Ltd. All rights reserved.
@@ -8,6 +8,11 @@
  *  Distributed under the Boost Software License, Version 1.0. (See accompanying
  *  file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
  */
+
+// NOTE: added backport of fix for macOS on ARM, as per https://github.com/catchorg/Catch2/commit/bb6d08323f23a39eb65dd86671e68f4f5d3f2d6c
+
+// NOTE: SIGSTKSZ is no longer a constant in glibc 2.34; see https://github.com/catchorg/Catch2/issues/2178
+
 #ifndef TWOBLUECUBES_SINGLE_INCLUDE_CATCH_HPP_INCLUDED
 #define TWOBLUECUBES_SINGLE_INCLUDE_CATCH_HPP_INCLUDED
 
@@ -228,7 +233,12 @@
     ( defined __GNUC__  && ( __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3 )) ) || \
     ( defined __clang__ && __clang_major__ >= 3 )
 
-#define CATCH_INTERNAL_CONFIG_COUNTER
+// Use of __COUNTER__ is suppressed during code analysis in CLion/AppCode 2017.2.x and former,
+// because __COUNTER__ is not properly handled by it.
+// This does not affect compilation
+#if ( !defined __JETBRAINS_IDE__ || __JETBRAINS_IDE__ >= 20170300L )
+    #define CATCH_INTERNAL_CONFIG_COUNTER
+#endif
 
 #endif
 
@@ -309,10 +319,7 @@
 #if defined(CATCH_INTERNAL_CONFIG_CPP11_UNIQUE_PTR) && !defined(CATCH_CONFIG_CPP11_NO_UNIQUE_PTR) && !defined(CATCH_CONFIG_CPP11_UNIQUE_PTR) && !defined(CATCH_CONFIG_NO_CPP11)
 #   define CATCH_CONFIG_CPP11_UNIQUE_PTR
 #endif
-// Use of __COUNTER__ is suppressed if __JETBRAINS_IDE__ is #defined (meaning we're being parsed by a JetBrains IDE for
-// analytics) because, at time of writing, __COUNTER__ is not properly handled by it.
-// This does not affect compilation
-#if defined(CATCH_INTERNAL_CONFIG_COUNTER) && !defined(CATCH_CONFIG_NO_COUNTER) && !defined(CATCH_CONFIG_COUNTER) && !defined(__JETBRAINS_IDE__)
+#if defined(CATCH_INTERNAL_CONFIG_COUNTER) && !defined(CATCH_CONFIG_NO_COUNTER) && !defined(CATCH_CONFIG_COUNTER)
 #   define CATCH_CONFIG_COUNTER
 #endif
 #if defined(CATCH_INTERNAL_CONFIG_CPP11_SHUFFLE) && !defined(CATCH_CONFIG_CPP11_NO_SHUFFLE) && !defined(CATCH_CONFIG_CPP11_SHUFFLE) && !defined(CATCH_CONFIG_NO_CPP11)
@@ -1318,10 +1325,12 @@ namespace Internal {
     template<> struct OperatorTraits<IsGreaterThanOrEqualTo>{ static const char* getName(){ return ">="; } };
 
     template<typename T>
-    T& removeConst(T const &t) { return const_cast<T&>(t); }
+    T& opCast(T const& t) { return const_cast<T&>(t); }
+
+// nullptr_t support based on pull request #154 from Konstantin Baumann
 #ifdef CATCH_CONFIG_CPP11_NULLPTR
-    inline std::nullptr_t removeConst(std::nullptr_t) { return nullptr; }
-#endif
+    inline std::nullptr_t opCast(std::nullptr_t) { return nullptr; }
+#endif // CATCH_CONFIG_CPP11_NULLPTR
 
     // So the compare overloads can be operator agnostic we convey the operator as a template
     // enum, which is used to specialise an Evaluator for doing the comparison.
@@ -1331,90 +1340,161 @@ namespace Internal {
     template<typename T1, typename T2>
     struct Evaluator<T1, T2, IsEqualTo> {
         static bool evaluate( T1 const& lhs, T2 const& rhs) {
-            return bool(removeConst(lhs) == removeConst(rhs) );
+            return bool( opCast( lhs ) ==  opCast( rhs ) );
         }
     };
     template<typename T1, typename T2>
     struct Evaluator<T1, T2, IsNotEqualTo> {
         static bool evaluate( T1 const& lhs, T2 const& rhs ) {
-            return bool(removeConst(lhs) != removeConst(rhs) );
+            return bool( opCast( lhs ) != opCast( rhs ) );
         }
     };
     template<typename T1, typename T2>
     struct Evaluator<T1, T2, IsLessThan> {
         static bool evaluate( T1 const& lhs, T2 const& rhs ) {
-            return bool(removeConst(lhs) < removeConst(rhs) );
+            return bool( opCast( lhs ) < opCast( rhs ) );
         }
     };
     template<typename T1, typename T2>
     struct Evaluator<T1, T2, IsGreaterThan> {
         static bool evaluate( T1 const& lhs, T2 const& rhs ) {
-            return bool(removeConst(lhs) > removeConst(rhs) );
+            return bool( opCast( lhs ) > opCast( rhs ) );
         }
     };
     template<typename T1, typename T2>
     struct Evaluator<T1, T2, IsGreaterThanOrEqualTo> {
         static bool evaluate( T1 const& lhs, T2 const& rhs ) {
-            return bool(removeConst(lhs) >= removeConst(rhs) );
+            return bool( opCast( lhs ) >= opCast( rhs ) );
         }
     };
     template<typename T1, typename T2>
     struct Evaluator<T1, T2, IsLessThanOrEqualTo> {
         static bool evaluate( T1 const& lhs, T2 const& rhs ) {
-            return bool(removeConst(lhs) <= removeConst(rhs) );
+            return bool( opCast( lhs ) <= opCast( rhs ) );
         }
     };
 
-    // Special case for comparing a pointer to an int (deduced for p==0)
-    template<typename T>
-    struct Evaluator<int const&, T* const&, IsEqualTo> {
-        static bool evaluate( int lhs, T* rhs) {
-            return reinterpret_cast<void const*>( lhs ) ==  rhs;
-        }
-    };
-    template<typename T>
-    struct Evaluator<T* const&, int const&, IsEqualTo> {
-        static bool evaluate( T* lhs, int rhs) {
-            return lhs == reinterpret_cast<void const*>( rhs );
-        }
-    };
-    template<typename T>
-    struct Evaluator<int const&, T* const&, IsNotEqualTo> {
-        static bool evaluate( int lhs, T* rhs) {
-            return reinterpret_cast<void const*>( lhs ) !=  rhs;
-        }
-    };
-    template<typename T>
-    struct Evaluator<T* const&, int const&, IsNotEqualTo> {
-        static bool evaluate( T* lhs, int rhs) {
-            return lhs != reinterpret_cast<void const*>( rhs );
-        }
-    };
+    template<Operator Op, typename T1, typename T2>
+    bool applyEvaluator( T1 const& lhs, T2 const& rhs ) {
+        return Evaluator<T1, T2, Op>::evaluate( lhs, rhs );
+    }
 
-    template<typename T>
-    struct Evaluator<long const&, T* const&, IsEqualTo> {
-        static bool evaluate( long lhs, T* rhs) {
-            return reinterpret_cast<void const*>( lhs ) ==  rhs;
-        }
-    };
-    template<typename T>
-    struct Evaluator<T* const&, long const&, IsEqualTo> {
-        static bool evaluate( T* lhs, long rhs) {
-            return lhs == reinterpret_cast<void const*>( rhs );
-        }
-    };
-    template<typename T>
-    struct Evaluator<long const&, T* const&, IsNotEqualTo> {
-        static bool evaluate( long lhs, T* rhs) {
-            return reinterpret_cast<void const*>( lhs ) !=  rhs;
-        }
-    };
-    template<typename T>
-    struct Evaluator<T* const&, long const&, IsNotEqualTo> {
-        static bool evaluate( T* lhs, long rhs) {
-            return lhs != reinterpret_cast<void const*>( rhs );
-        }
-    };
+    // This level of indirection allows us to specialise for integer types
+    // to avoid signed/ unsigned warnings
+
+    // "base" overload
+    template<Operator Op, typename T1, typename T2>
+    bool compare( T1 const& lhs, T2 const& rhs ) {
+        return Evaluator<T1, T2, Op>::evaluate( lhs, rhs );
+    }
+
+    // unsigned X to int
+    template<Operator Op> bool compare( unsigned int lhs, int rhs ) {
+        return applyEvaluator<Op>( lhs, static_cast<unsigned int>( rhs ) );
+    }
+    template<Operator Op> bool compare( unsigned long lhs, int rhs ) {
+        return applyEvaluator<Op>( lhs, static_cast<unsigned int>( rhs ) );
+    }
+    template<Operator Op> bool compare( unsigned char lhs, int rhs ) {
+        return applyEvaluator<Op>( lhs, static_cast<unsigned int>( rhs ) );
+    }
+
+    // unsigned X to long
+    template<Operator Op> bool compare( unsigned int lhs, long rhs ) {
+        return applyEvaluator<Op>( lhs, static_cast<unsigned long>( rhs ) );
+    }
+    template<Operator Op> bool compare( unsigned long lhs, long rhs ) {
+        return applyEvaluator<Op>( lhs, static_cast<unsigned long>( rhs ) );
+    }
+    template<Operator Op> bool compare( unsigned char lhs, long rhs ) {
+        return applyEvaluator<Op>( lhs, static_cast<unsigned long>( rhs ) );
+    }
+
+    // int to unsigned X
+    template<Operator Op> bool compare( int lhs, unsigned int rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned int>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( int lhs, unsigned long rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned int>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( int lhs, unsigned char rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned int>( lhs ), rhs );
+    }
+
+    // long to unsigned X
+    template<Operator Op> bool compare( long lhs, unsigned int rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( long lhs, unsigned long rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( long lhs, unsigned char rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned long>( lhs ), rhs );
+    }
+
+    // pointer to long (when comparing against NULL)
+    template<Operator Op, typename T> bool compare( long lhs, T* rhs ) {
+        return Evaluator<T*, T*, Op>::evaluate( reinterpret_cast<T*>( lhs ), rhs );
+    }
+    template<Operator Op, typename T> bool compare( T* lhs, long rhs ) {
+        return Evaluator<T*, T*, Op>::evaluate( lhs, reinterpret_cast<T*>( rhs ) );
+    }
+
+    // pointer to int (when comparing against NULL)
+    template<Operator Op, typename T> bool compare( int lhs, T* rhs ) {
+        return Evaluator<T*, T*, Op>::evaluate( reinterpret_cast<T*>( lhs ), rhs );
+    }
+    template<Operator Op, typename T> bool compare( T* lhs, int rhs ) {
+        return Evaluator<T*, T*, Op>::evaluate( lhs, reinterpret_cast<T*>( rhs ) );
+    }
+
+#ifdef CATCH_CONFIG_CPP11_LONG_LONG
+    // long long to unsigned X
+    template<Operator Op> bool compare( long long lhs, unsigned int rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( long long lhs, unsigned long rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( long long lhs, unsigned long long rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( long long lhs, unsigned char rhs ) {
+        return applyEvaluator<Op>( static_cast<unsigned long>( lhs ), rhs );
+    }
+
+    // unsigned long long to X
+    template<Operator Op> bool compare( unsigned long long lhs, int rhs ) {
+        return applyEvaluator<Op>( static_cast<long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( unsigned long long lhs, long rhs ) {
+        return applyEvaluator<Op>( static_cast<long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( unsigned long long lhs, long long rhs ) {
+        return applyEvaluator<Op>( static_cast<long>( lhs ), rhs );
+    }
+    template<Operator Op> bool compare( unsigned long long lhs, char rhs ) {
+        return applyEvaluator<Op>( static_cast<long>( lhs ), rhs );
+    }
+
+    // pointer to long long (when comparing against NULL)
+    template<Operator Op, typename T> bool compare( long long lhs, T* rhs ) {
+        return Evaluator<T*, T*, Op>::evaluate( reinterpret_cast<T*>( lhs ), rhs );
+    }
+    template<Operator Op, typename T> bool compare( T* lhs, long long rhs ) {
+        return Evaluator<T*, T*, Op>::evaluate( lhs, reinterpret_cast<T*>( rhs ) );
+    }
+#endif // CATCH_CONFIG_CPP11_LONG_LONG
+
+#ifdef CATCH_CONFIG_CPP11_NULLPTR
+    // pointer to nullptr_t (when comparing against nullptr)
+    template<Operator Op, typename T> bool compare( std::nullptr_t, T* rhs ) {
+        return Evaluator<T*, T*, Op>::evaluate( nullptr, rhs );
+    }
+    template<Operator Op, typename T> bool compare( T* lhs, std::nullptr_t ) {
+        return Evaluator<T*, T*, Op>::evaluate( lhs, nullptr );
+    }
+#endif // CATCH_CONFIG_CPP11_NULLPTR
 
 } // end of namespace Internal
 } // end of namespace Catch
@@ -1835,7 +1915,7 @@ public:
 
     void endExpression() const {
         m_rb
-            .setResultType(  Internal::Evaluator<LhsT, RhsT, Op>::evaluate( m_lhs, m_rhs ) )
+            .setResultType( Internal::compare<Op>( m_lhs, m_rhs ) )
             .endExpression( *this );
     }
 
@@ -2054,6 +2134,9 @@ namespace Catch{
         #define CATCH_TRAP() \
                 __asm__("li r0, 20\nsc\nnop\nli r0, 37\nli r4, 2\nsc\nnop\n" \
                 : : : "memory","r0","r3","r4" ) /* NOLINT */
+    #elif defined(__aarch64__)
+        // Backport of https://github.com/catchorg/Catch2/commit/a25c1a24af8bffd35727a888a307ff0280cf9387
+        #define CATCH_TRAP() __asm__(".inst 0xd4200000")
     #else
         #define CATCH_TRAP() __asm__("int $3\n" : : /* NOLINT */ )
     #endif
@@ -2756,7 +2839,8 @@ namespace Detail {
             if (relativeOK) {
                 return true;
             }
-            return std::fabs(lhs_v - rhs.m_value) < rhs.m_margin;
+
+            return std::fabs(lhs_v - rhs.m_value) <= rhs.m_margin;
         }
 
         template <typename T, typename = typename std::enable_if<std::is_constructible<double, T>::value>::type>
@@ -2828,7 +2912,7 @@ namespace Detail {
             if (relativeOK) {
                 return true;
             }
-            return std::fabs(lhs - rhs.m_value) < rhs.m_margin;
+            return std::fabs(lhs - rhs.m_value) <= rhs.m_margin;
         }
 
         friend bool operator == ( Approx const& lhs, double rhs ) {
@@ -5279,10 +5363,6 @@ namespace Catch {
             .describe( "should output be colourised" )
             .bind( &setUseColour, "yes|no" );
 
-        cli["--use-colour"]
-            .describe( "should output be colourised" )
-            .bind( &setUseColour, "yes|no" );
-
         cli["--libidentify"]
             .describe( "report name and version according to libidentify standard" )
             .bind( &ConfigData::libIdentify );
@@ -6527,6 +6607,7 @@ namespace Catch {
 
 #endif // not Windows
 
+#include <cassert>
 #include <set>
 #include <string>
 
@@ -7215,14 +7296,14 @@ namespace Catch {
 namespace Catch {
 
     struct RandomNumberGenerator {
-        typedef std::ptrdiff_t result_type;
+        typedef unsigned int result_type;
 
         result_type operator()( result_type n ) const { return std::rand() % n; }
 
 #ifdef CATCH_CONFIG_CPP11_SHUFFLE
-        static constexpr result_type min() { return 0; }
-        static constexpr result_type max() { return 1000000; }
-        result_type operator()() const { return std::rand() % max(); }
+        static constexpr result_type (min)() { return 0; }
+        static constexpr result_type (max)() { return 1000000; }
+        result_type operator()() const { return std::rand() % (max)(); }
 #endif
         template<typename V>
         static void shuffle( V& vector ) {
@@ -8136,7 +8217,7 @@ namespace Catch {
 
     std::string AssertionResult::getExpression() const {
         if( isFalseTest( m_info.resultDisposition ) )
-            return '!' + capturedExpressionWithSecondArgument(m_info.capturedExpression, m_info.secondArg);
+            return "!(" + capturedExpressionWithSecondArgument(m_info.capturedExpression, m_info.secondArg) + ")";
         else
             return capturedExpressionWithSecondArgument(m_info.capturedExpression, m_info.secondArg);
     }
@@ -8394,7 +8475,7 @@ namespace Catch {
     }
 
     inline Version libraryVersion() {
-        static Version version( 1, 10, 0, "", 0 );
+        static Version version( 1, 12, 2, "", 0 );
         return version;
     }
 
@@ -8429,11 +8510,18 @@ namespace Catch {
     : m_info( other.m_info )
     {}
 
+#if defined(_MSC_VER)
+#pragma warning(push)
+#pragma warning(disable:4996) // std::uncaught_exception is deprecated in C++17
+#endif
     ScopedMessage::~ScopedMessage() {
         if ( !std::uncaught_exception() ){
             getResultCapture().popScopedMessage(m_info);
         }
     }
+#if defined(_MSC_VER)
+#pragma warning(pop)
+#endif
 
 } // end namespace Catch
 
@@ -9092,6 +9180,8 @@ std::string toString( std::nullptr_t ) {
 // #included from: catch_result_builder.hpp
 #define TWOBLUECUBES_CATCH_RESULT_BUILDER_HPP_INCLUDED
 
+#include <cassert>
+
 namespace Catch {
 
     ResultBuilder::ResultBuilder(   char const* macroName,
@@ -10208,12 +10298,12 @@ namespace Catch {
 
             bool includeResults = m_config->includeSuccessfulResults() || !result.isOk();
 
-            if( includeResults ) {
+            if( includeResults || result.getResultType() == ResultWas::Warning ) {
                 // Print any info messages in <Info> tags.
                 for( std::vector<MessageInfo>::const_iterator it = assertionStats.infoMessages.begin(), itEnd = assertionStats.infoMessages.end();
                      it != itEnd;
                      ++it ) {
-                    if( it->type == ResultWas::Info ) {
+                    if( it->type == ResultWas::Info && includeResults ) {
                         m_xml.scopedElement( "Info" )
                                 .writeText( it->message );
                     } else if ( it->type == ResultWas::Warning ) {
@@ -10585,6 +10675,7 @@ namespace Catch {
 // #included from: ../reporters/catch_reporter_console.hpp
 #define TWOBLUECUBES_CATCH_REPORTER_CONSOLE_HPP_INCLUDED
 
+#include <cassert>
 #include <cfloat>
 #include <cstdio>