CMake is a "meta build system" that reads a description
of the project written in the CMakeLists.txt
files and emits a build
system for that project of your choice using one of CMake's "generators".
This allows CMake to support many different platforms and build tools.
You can run cmake --help
to see the list of supported "generators"
on your platform. Example generators include "UNIX Makefiles" and "Visual Studio
12 2013".
If you have never used the python build system you can skip this step.
The existing Python build system creates generated source files in the source tree. The CMake build system will refuse to work if it detects this so you need to clean your source tree first.
To do this run the following in the root of the repository
git clean -nx src
This will list everything that will be removed. If you are happy with this then run.
git clean -fx src
which will remove the generated source files.
Run the following in the top level directory of the Z3 repository.
mkdir build
cd build
cmake -G "Unix Makefiles" ../
make -j4 # Replace 4 with an appropriate number
Note that on some platforms "Unix Makefiles" is the default generator so on those
platforms you don't need to pass -G "Unix Makefiles"
command line option to
cmake
.
Note there is nothing special about the build
directory name here. You can call
it whatever you like.
Note the "Unix Makefile" generator is a "single" configuration generator which
means you pick the build type (e.g. Debug
, Release
) when you invoke CMake.
You can set the build type by passing it to the cmake
invocation like so:
cmake -G "Unix Makefiles" -DCMAKE_BUILD_TYPE=Release ../
See the section on "Build Types" for the different CMake build types.
If you wish to use a different compiler set the CXX and CC environment variables
passed to cmake. This must be done at the very first invocation to cmake
in the build directory because once configuration has happened the compiler
is fixed. If you want to use a different compiler to the one you have already
configured you either need to make a new build directory or delete the contents
of the current build directory and start again.
For example to use clang the cmake
line would be
CC=clang CXX=clang++ cmake ../
Note that CMake build will detect the target architecture that compiler is set up
to build for and the generated build system will build for that architecture.
If there is a way to tell your compiler to build for a different architecture via
compiler flags then you can set the CFLAGS
and CXXFLAGS
environment variables
to have the build target that architecture.
For example if you are on a x86_64 machine and you want to do a 32-bit build and have
a multilib version of GCC you can run cmake
like this
CFLAGS="-m32" CXXFLAGS="-m32" CC=gcc CXX=g++ cmake ../
Note like with the CC
and CXX
flags this must be done on the very first invocation
to CMake in the build directory.
Ninja is a simple build system that is built for speed. It can be significantly faster than "UNIX Makefile"s because it is not a recursive build system and thus doesn't create a new process everytime it traverses into a directory. Ninja is particularly appropriate if you want fast incremental building.
Basic usage is as follows:
mkdir build
cd build
cmake -G "Ninja" ../
ninja
Note the discussion of the CC
, CXX
, CFLAGS
and CXXFLAGS
for "Unix Makefiles"
also applies here.
Note also that like the "Unix Makefiles" generator, the "Ninja" generator is a single configuration
generator so you pick the build type when you invoke cmake
by passing CMAKE_BUILD_TYPE=<build_type>
to cmake
. See the section on "Build Types".
Note that Ninja runs in parallel by default. Use the -j
flag to change this.
Note that Ninja also runs on Windows. You just need to run cmake
in an
environment where the compiler can be found. If you have Visual Studio
installed it typically ships with a "Developer Command Prompt Window" that you
can use which has the environment variables setup for you.
NMake is a build system that ships with Visual Studio. You are advised to use
Ninja instead which is significantly faster due to supporting concurrent
builds. However CMake does support NMake if you wish to use it. Note that
NMake is a single configuration generator so you must set CMAKE_BUILD_TYPE
to set the build type.
Basic usage:
- Launch the "Developer Command Prompt Windows"
- Change to the root of the Z3 repository
mkdir build
cd build
cmake -G "NMake Makefiles" ../
nmake
Visual Studio 19 comes with integrated support for CMake. It suffices to open the (z3) folder where this file and the Z3 project CMakeLists.txt resides, and Visual Studio does the rest.
For legacy versions of Visual Studio a process is as follows: For the Visual Studio generators you need to know which version of Visual Studio you wish to use and also what architecture you want to build for.
We'll use the cmake-gui
here as it is easier to pick the right generator but this can
be scripted if need be.
Here are the basic steps:
- Create an empty build directory
- Start the cmake-gui program
- Set "where is the source code" to the root of the Z3 project repository. You can do this by pressing the "Browse Source..." button and picking the directory.
- Set "where to build the binaries" to the empty build directory you just created. You can do this by pressing the "Browse build..." button and picking the directory.
- Press the "Configure" button
- A window will appear asking you to pick the generator to use. Pick the
generator that matches the version of Visual Studio you are using. Note also
that some of the generator names contain
Win64
(e.g.Visual Studio 12 2013 Win64
) this indicates a x86 64-bit build. Generator names without this (e.g.Visual Studio 12 2013
) are x86 32-bit build. - Press the "Finish" button and wait for CMake to finish it's first configure.
- A set of configuration options will appear which will affect various aspects of the build. Change them as you desire. If you change a set of options press the "Configure" again. Additional options may appear when you do this.
- When you have finished changing configuration options press the "Generate" button.
- When generation is done close cmake-gui.
- In the build directory open the generated
Z3.sln
solution file created by CMake with Visual Studio. - In Visual Studio pick the build type (e.g.
Debug
,Release
) you want. - Click "BUILD > Build Solution".
Note that unlike the "Unix Makefile" and "Ninja" generators the Visual Studio generators are multi-configuration generators which means you don't set the build type when invoking CMake. Instead you set the build type inside Visual Studio. See the "Build Type" section for more information.
The general workflow when using CMake is the following
- Create a new build directory
- Configure the project
- Generate the build system
- Invoke the build system to build the project
To perform steps 2 and 3 you can choose from three different tools
- cmake
- ccmake
- cmake-gui
cmake
is a command line tool and is what you should use if you are
writing a script to build Z3. This tool performs steps 1 and 2 in one
go without user interaction. The ccmake
and cmake-gui
tools are
more interactive and allow you to change various options. In both these
tools the basic steps to follow are:
- Configure.
- Change any options you wish. Everytime you change a set of options You should configure again. This may cause new options to appear
- Generate.
For information see https://cmake.org/runningcmake/
Note when invoking CMake you give it the path to the source directory.
This is the top-level directory in the Z3 repository containing a
CMakeLists.txt
. That file should contain the line project(Z3 C CXX)
.
If you give it a path deeper into the Z3 repository (e.g. the src
directory)
the configure step will fail.
Several build types are supported.
- Release
- Debug
- RelWithDebInfo
- MinSizeRel
For the single configuration generators (e.g. "Unix Makefile" and "Ninja") you set the
build type when invoking cmake
by passing -DCMAKE_BUILD_TYPE=<build_type>
where
<build_type>
is one of the build types specified above.
For multi-configuration generators (e.g. Visual Studio) you don't set the build type when invoking CMake and instead set the build type within Visual Studio itself.
The following useful options can be passed to CMake whilst configuring.
CMAKE_BUILD_TYPE
- STRING. The build type to use. Only relevant for single configuration generators (e.g. "Unix Makefile" and "Ninja").CMAKE_INSTALL_BINDIR
- STRING. The path to install z3 binaries (relative toCMAKE_INSTALL_PREFIX
), e.g.bin
.CMAKE_INSTALL_INCLUDEDIR
- STRING. The path to install z3 include files (relative toCMAKE_INSTALL_PREFIX
), e.g.include
.CMAKE_INSTALL_LIBDIR
- STRING. The path to install z3 libraries (relative toCMAKE_INSTALL_PREFIX
), e.g.lib
.CMAKE_INSTALL_PREFIX
- STRING. The install prefix to use (e.g./usr/local/
).CMAKE_INSTALL_PKGCONFIGDIR
- STRING. The path to install pkgconfig files.CMAKE_INSTALL_PYTHON_PKG_DIR
- STRING. The path to install the z3 python bindings. This can be relative (toCMAKE_INSTALL_PREFIX
) or absolute.CMAKE_INSTALL_Z3_CMAKE_PACKAGE_DIR
- STRING. The path to install CMake package files (e.g./usr/lib/cmake/z3
).CMAKE_INSTALL_API_BINDINGS_DOC
- STRING. The path to install documentation for API bindings.PYTHON_EXECUTABLE
- STRING. The python executable to use during the build.Z3_ENABLE_TRACING_FOR_NON_DEBUG
- BOOL. If set toTRUE
enable tracing in non-debug builds, if set toFALSE
disable tracing in non-debug builds. Note in debug builds tracing is always enabled.Z3_BUILD_LIBZ3_SHARED
- BOOL. If set toTRUE
build libz3 as a shared library otherwise build as a static library.Z3_ENABLE_EXAMPLE_TARGETS
- BOOL. If set toTRUE
add the build targets for building the API examples.Z3_USE_LIB_GMP
- BOOL. If set toTRUE
use the GNU multiple precision library. If set toFALSE
use an internal implementation.Z3_BUILD_PYTHON_BINDINGS
- BOOL. If set toTRUE
then Z3's python bindings will be built.Z3_INSTALL_PYTHON_BINDINGS
- BOOL. If set toTRUE
andZ3_BUILD_PYTHON_BINDINGS
isTRUE
then running theinstall
target will install Z3's Python bindings.Z3_BUILD_DOTNET_BINDINGS
- BOOL. If set toTRUE
then Z3's .NET bindings will be built.Z3_INSTALL_DOTNET_BINDINGS
- BOOL. If set toTRUE
andZ3_BUILD_DOTNET_BINDINGS
isTRUE
then running theinstall
target will install Z3's .NET bindings.Z3_DOTNET_CSC_EXECUTABLE
- STRING. The path to the C# compiler to use. Only relevant ifZ3_BUILD_DOTNET_BINDINGS
is set toTRUE
.Z3_DOTNET_GACUTIL_EXECUTABLE
- STRING. The path to the gacutil program to use. Only relevant ifBUILD_DOTNET_BINDINGS
is set toTRUE
.Z3_BUILD_JAVA_BINDINGS
- BOOL. If set toTRUE
then Z3's Java bindings will be built.Z3_INSTALL_JAVA_BINDINGS
- BOOL. If set toTRUE
andZ3_BUILD_JAVA_BINDINGS
isTRUE
then running theinstall
target will install Z3's Java bindings.Z3_JAVA_JAR_INSTALLDIR
- STRING. The path to directory to install the Z3 Java.jar
file. This path should be relative toCMAKE_INSTALL_PREFIX
.Z3_JAVA_JNI_LIB_INSTALLDIRR
- STRING. The path to directory to install the Z3 Java JNI bridge library. This path should be relative toCMAKE_INSTALL_PREFIX
.Z3_INCLUDE_GIT_DESCRIBE
- BOOL. If set toTRUE
and the source tree of Z3 is a git repository then the output ofgit describe
will be included in the build.Z3_INCLUDE_GIT_HASH
- BOOL. If set toTRUE
and the source tree of Z3 is a git repository then the git hash will be included in the build.Z3_BUILD_DOCUMENTATION
- BOOL. If set toTRUE
then documentation for the API bindings can be built by invoking theapi_docs
target.Z3_INSTALL_API_BINDINGS_DOCUMENTATION
- BOOL. If set toTRUE
andZ3_BUILD_DOCUMENTATION` is
TRUEthen documentation for API bindings will be installed when running the
install`` target.Z3_ALWAYS_BUILD_DOCS
- BOOL. If set toTRUE
andZ3_BUILD_DOCUMENTATION
isTRUE
then documentation for API bindings will always be built. Disabling this is useful for faster incremental builds. The documentation can be manually built by invoking theapi_docs
target.Z3_LINK_TIME_OPTIMIZATION
- BOOL. If set toTRUE
link time optimization will be enabled.Z3_ENABLE_CFI
- BOOL. If set toTRUE
will enable Control Flow Integrity security checks. This is only supported by MSVC and Clang and will fail on other compilers. This requires Z3_LINK_TIME_OPTIMIZATION to also be enabled.Z3_API_LOG_SYNC
- BOOL. If set toTRUE
will enable experimental API log sync feature.WARNINGS_AS_ERRORS
- STRING. If set toTRUE
compiler warnings will be treated as errors. If set toFalse
compiler warnings will not be treated as errors. If set toSERIOUS_ONLY
a subset of compiler warnings will be treated as errors.Z3_C_EXAMPLES_FORCE_CXX_LINKER
- BOOL. If set toTRUE
the C API examples will request that the C++ linker is used rather than the C linker.Z3_BUILD_EXECUTABLE
- BOOL. If set toTRUE
build the z3 executable. Defaults toTRUE
unless z3 is being built as a submodule in which case it defaults toFALSE
.Z3_BUILD_TEST_EXECUTABLES
- BOOL. If set toTRUE
build the z3 test executables. Defaults toTRUE
unless z3 is being built as a submodule in which case it defaults toFALSE
.Z3_SAVE_CLANG_OPTIMIZATION_RECORDS
- BOOL. If set toTRUE
saves Clang optimization records by setting the compiler flag-fsave-optimization-record
.Z3_SINGLE_THREADED
- BOOL. If set toTRUE
compiles Z3 for single threaded mode.
On the command line these can be passed to cmake
using the -D
option. In ccmake
and cmake-gui
these can be set in the user interface.
Example
cmake -DCMAKE_BUILD_TYPE=Release -DZ3_ENABLE_TRACING_FOR_NON_DEBUG=FALSE ../
Z3 exposes various language bindings for its API. Below are some notes on building and/or installing these bindings when building Z3 with CMake.
The CMake build uses the FindJava
and FindJNI
cmake modules to detect the
installation of Java. If CMake fails to find your installation of Java set the
JAVA_HOME
environment variable when invoking CMake so that it points at the
correct location. For example
JAVA_HOME=/usr/lib/jvm/default cmake -DZ3_BUILD_JAVA_BINDINGS=ON ../
Note that the built .jar
file is named com.microsoft.z3-VERSION.jar
where VERSION
is the Z3 version. Under non Windows systems a
symbolic link named com.microsoft.z3.jar
is provided. This symbolic
link is not created when building under Windows.
These notes are help developers and packagers of Z3.
Install and uninstall targets are supported. Use CMAKE_INSTALL_PREFIX
to
set the install prefix. If you also need need to control which directories are
used for install set the documented CMAKE_INSTALL_*
options.
To install run
make install
To uninstall run
make uninstall
Note that DESTDIR
is supported for staged installs.
To install
mkdir staged
make install DESTDIR=/full/path/to/staged/
to uninstall
make uninstall DESTDIR=/full/path/to/staged
The above also works for Ninja but DESTDIR
must be an environment variable instead.
If you are using the "UNIX Makefiles" generator and want to see exactly the commands that are
being run you can pass VERBOSE=1
to make.
make VERBOSE=1
If you are using Ninja you can use the -v
flag.
To see the list of targets run
make help
There are a few special targets:
clean
all the built targets in the current directory and belowedit_cache
will invoke one of the CMake tools (depending on which is available) to let you change configuration options.rebuild_cache
will reinvokecmake
for the project.api_docs
will build the documentation for the API bindings.
The build system supports single configuration and multi-configuration generators. This means
it is not possible to know the build type at configure time and therefore ${CMAKE_BUILD_TYPE}
should not be conditionally used to set compiler flags or definitions. Instead you should use Generator expressions which are evaluated by the generator.
For example
$<$<CONFIG:Debug>:Z3DEBUG>
If the build type at build time is Debug
this evaluates to Z3DEBUG
but evaluates to nothing for all other configurations. You can see examples of this in the CMakeLists.txt
files.
It is tempting use file-globbing in CMakeLists.txt
to find a set for files matching a pattern and
use them as the sources to build a target. This however is a bad idea because it prevents CMake from knowing when it needs to rerun itself. This is why source file names are explicitly listed in the CMakeLists.txt
so that when changes are made the source files used to build a target automatically triggers a rerun of CMake.
Long story short. Don't use file globbing.
By default the WARNINGS_AS_ERRORS
flag is set to SERIOUS_ONLY
which means
some warnings will be treated as errors. These warnings are controlled by the
relevant *_WARNINGS_AS_ERRORS
list defined in
cmake/compiler_warnings.cmake
.
Additional warnings should only be added here if the warnings has no false positives.