There are three ways to build SWI-Prolog from source for Windows. Here are some notes regarding the different approaches:

• Compiler. MinGW (GCC) produces 20-30% faster binaries than MSVC. The resulting library can be used with MinGW as well as MSVC. Debugging can be tricky because the GNU gdb debugger does not understand the MSVC debugging information and Microsoft debugger does not understand the GCC debugging information.
• Cross compiling Windows is notoriously slow in starting programs and the compilation starts many programs. CMake does not support symbolic links on Windows, which implies that we need to do a lot of copying to establish the target file structure. Cross compiling using Linux is much faster.
• Dependencies Getting the dependencies is fairly well handled in all three scenarios. They are built into the Docker image when using docker based cross compilation. For MSYS2, the dependencies are produced by the MSYS packages and for Microsoft Visual C++ we use vcpkg.

While building for Windows has improved a lot with the migration to CMake for SWI-Prolog and much improved dependency handling, it is still a tedious process that provides little benefits over using the binary installers that can be downloaded from https://www.swi-prolog.org/Download.html

Cross-compiling

We provide a Docker file that is based on Fedora Linux and builds SWI-Prolog using the MinGW cross compilers and Wine emulation to run the Prolog steps. This should work on any machine capable of running Docker.

The Docker image uses a host filesystem directory to access the source tree and create the result.

See the Makefile of the repo for using it.

This is used to build the binary releases for Windows as well as the daily builds.

Using MSYS

There are also swi-prolog packages for MSYS2 and you can use MSYS2 to compile SWI-Prolog from source using the dependencies from MSYS2.

Using Microsoft Visual C++ (MSVC)

This is a recent addition. The process is based MSVC and the vcpkg package manager.

Preparation

Install Microsoft Visual C++. You need the command line tools and optionally Visual Studio. The the supplied shortcuts to start a cmd.exe or PowerShell prompt setup for running the MSVC command line tools.

Install vcpkg if you do not have it. The vcpkg site suggests to install the package manager into a folder close to the device root, suggesting c:\src\vcpkg or c:\dev\vcpkg. We use the latter in our examples.

c:
cd \
mkdir dev
git clone https://github.com/Microsoft/vcpkg.git
cd vcpkg
.\bootstrap-vcpkg.bat

Install the dependencies we need. Run these commands inside the c:\dev\vcpkg directory. The --triplet x64-windows flag demands for the 64 bit versions of the dependencies. The zlib and pthreads dependencies are obligatory. The others are to support the packages.

.\vcpkg install --triplet x64-windows zlib pthreads
.\vcpkg install --triplet x64-windows pcre2 libjpeg-turbo libyaml libarchive

Configure the system

Go to the checked out SWI-Prolog sources and run the commands below. Adjust the tool chain script if you installed vcpkg elsewhere.

mkdir build
cd build
cmake -DCMAKE_TOOLCHAIN_FILE=/dev/vcpkg/scripts/buildsystems/vcpkg.cmake ..

Build the system

To build using the command line tools, run the following from the build directory. Alternatively you may open the SWI-Prolog.sln file that opens Visual Studio.

cmake --build . --config Release

Test the system

The system may be tested using ctest using the -C option to specify the configuration. The -j 4 below specifies concurrency. Typically one should use the number of cores as most tests are CPU bound. The --output-on-failure option writes the Prolog output to the terminal in case a test fails.

ctest -j 4 --output-on-failure -C Release

Status

The Microsoft Visual C++ port is experimental. The above builds the core system and all packages except for JPL and bdb. The tests pass except for a spurious failure in swipl:basics.

Packaging has not been tested.