Skip to content

Latest commit

 

History

History
564 lines (471 loc) · 19.6 KB

COMPILING.md

File metadata and controls

564 lines (471 loc) · 19.6 KB
Raw

What architecture?

CPROVER now needs a C++11 compliant compiler and is known to work in the following environments:

  • Linux
  • MacOS X
  • Windows

The above environments are currently tested as part of our continuous integration system. It separately tests both the CMake build system and the hand-written make files. The latest build steps being used in CI can be found here.

The environments below have been used successfully in the past, but are not actively tested:

  • Solaris 11
  • FreeBSD 11

Building using CMake

Building with CMake is supported across Linux, MacOS X and Windows with Visual Studio 2019. There are also hand-written make files which can be used to build separate targets independently. Usage of the hand-written make files is explained in a separate section. The CMake build can build the complete repository in fewer steps and supports better integration with various IDEs and static-analysis tools. On Windows, the CMake build has the advantage of not depending on Cygwin or MinGW, and doesn't require manual modification of build files.

  1. Ensure you have all the build dependencies installed. Build dependencies are the same as for the makefile build, but with the addition of CMake version 3.2 or higher. The installed CMake version can be queried with cmake --version. To install CMake:

    • On Debian-like distributions, do 
      apt-get install g++ gcc flex bison make git curl patch cmake
    • On Red Hat/Fedora or derivates, do 
      dnf install gcc gcc-c++ flex bison curl patch cmake
    • On macOS, do 
      xcode-select --install
      You should also install Homebrew, after which you can run brew install cmake to install CMake.
    • On Windows, ensure you have Visual Studio 2019 or later installed. The developer command line that comes with Visual Studio 2019 has CMake already available. You will also need to ensure that you have winflexbison installed and available in the path. winflexbison is available from the github release page or through the chocolatey package manager. Installing strawberryperl is advised if you want to run tests on Windows. This is available from the Strawberry Perl website or through the the chocolatey package manager.
    • Use of CMake has not been tested on Solaris or FreeBSD. However, it should be possible to install CMake from the system package manager or the official download page on those systems. The dependencies (as listed in the relevant sections above) will also be required, and should be installed using the suggested methods.

  2. Navigate to the top level folder of the project. This is different from the Makefile build, which requires you to navigate to the src directory first.

  3. Update git submodules:

    git submodule update --init

  4. Generate build files with CMake:

    cmake -S . -Bbuild

    This command tells CMake to use the configuration in the current directory, and to generate build files into the build directory. This is the point to specify custom build settings, such as compilers and build back-ends. You can use clang (for example) by adding the argument -DCMAKE_CXX_COMPILER=clang++ to the command line. You can also tell CMake to generate IDE projects by supplying the -G flag. Run cmake -G for a comprehensive list of supported back-ends.

    As part of this step, CMake will download the back-end solver (see Section "Compiling with alternative SAT solvers" in this document for configuration options). Should it be necessary to perform this step without network access, a solver can be downloaded ahead of the above cmake invocation as follows:

    mkdir -p build/minisat2-download/minisat2-download-prefix/src/
wget http://ftp.debian.org/debian/pool/main/m/minisat2/minisat2_2.2.1.orig.tar.gz \
  -O build/minisat2-download/minisat2-download-prefix/src/minisat2_2.2.1.orig.tar.gz

    On macOS >10.14, the build will fail unless you explicitly specify the full path to the compiler. This issue is being tracked here. The invocation thus looks like this:

    cmake -S. -Bbuild -DCMAKE_C_COMPILER=/usr/bin/clang

    Generally it is not necessary to manually specify individual compiler or linker flags, as CMake defines a number of "build modes" including Debug and Release modes. To build in a particular mode, add the flag -DCMAKE_BUILD_TYPE=Debug (or RelWithDebInfo) to the initial invocation. The default is to perform an optimized build via the Release configuration.

    If you do need to manually add flags, use -DCMAKE_CXX_FLAGS=... and -DCMAKE_EXE_LINKER_FLAGS=.... This is useful for enabling clang's sanitizers.

    Note: Building without JBMC: On platforms where installing the Java Development Kit and Maven is difficult, you can avoid needing these dependencies by not building JBMC. Just pass -DWITH_JBMC=OFF to cmake.

    Finally, to enable building universal binaries on macOS, you can pass the flag -DCMAKE_OSX_ARCHITECTURES=i386;x86_64. If you don't supply this flag, the built binaries will only work on the architecture of the machine being used to do the build.

  5. Run the build:

    cmake --build build

    This command tells CMake to invoke the correct tool to run the build in the build directory. You can also use the build back-end directly by invoking make, ninja, or opening the generated IDE project as appropriate. The complete set of built binaries can be found in build/bin/ once the build is complete.

    Parallel building: You can pass -j<numjobs> to make to indicate how many jobs to run simultaneously. ninja defaults to building with # of cores + 2 jobs at the same time.

#Building using Make

The rest of this section is split up based on the platform being built on. Please read the section appropriate for your platform.

Compiling with Make on Linux

We assume that you have a Debian/Ubuntu or Red Hat-like distribution.

  1. You need a C/C++ compiler, Flex and Bison, and GNU make. The GNU Make needs to be version 3.81 or higher. On Debian-like distributions, do as root:

    apt-get install g++ gcc flex bison make git curl patch

    On Red Hat/Fedora or derivates, do as root:

    dnf install gcc gcc-c++ flex bison curl patch

    Note that you need g++ version 5.0 or newer.

    On Amazon Linux and similar distributions, do as root:

    yum install gcc72-c++ flex bison curl patch tar

    To compile JBMC, you additionally need the JDK and Maven 3. You also need jq if you wish to run the entire test suite. On Debian-like distributions, do as root:

    apt-get install openjdk-8-jdk maven jq

    On Red Hat/Fedora or derivates, do as root:

    dnf install java-1.8.0-openjdk-devel maven jq

  2. As a user, get the CBMC source via

    git clone https://github.com/diffblue/cbmc cbmc-git
cd cbmc-git

  3. To compile, do

    make -C src minisat2-download
make -C src

    See doc/architectural/compilation-and-development.md for instructions on how to use a SAT solver other than MiniSat 2.

  4. To compile JBMC, do

    make -C jbmc/src setup-submodules
make -C jbmc/src

Compiling with Make on MacOS

Follow these instructions:

  1. You need a C/C++ compiler, Flex and Bison, and GNU make. To this end, first install the XCode from the App-store and then type 
    xcode-select --install
    in a terminal window.
  2. Then get the CBMC source via 
    git clone https://github.com/diffblue/cbmc cbmc-git
cd cbmc-git
  3. To compile CBMC, do 
    make -C src minisat2-download
make -C src
  4. To compile JBMC, you additionally need Maven 3, which has to be installed manually. Then do 
    make -C jbmc/src setup-submodules
make -C jbmc/src

Compiling with Make on Solaris

We assume Solaris 11.4 or newer. To build JBMC, you'll need to install Maven 3 manually.

  1. As root, get the necessary development tools: 
    pkg install gcc-c++-7 bison flex
  2. As a user, get the CBMC source via 
    git clone https://github.com/diffblue/cbmc cbmc-git
cd cbmc-git
  3. To compile CBMC, type 
    gmake -C src minisat2-download DOWNLOADER=wget TAR=gtar
gmake -C src
  4. To compile JBMC, type 
    gmake -C jbmc/src setup-submodules
gmake -C jbmc/src

Compiling with Make on FreeBSD

  1. As root, get the necessary tools: 
    pkg install bash gmake git www/p5-libwww patch flex bison
    To compile JBMC, additionally install 
    pkg install openjdk8 wget maven3
  2. As a user, get the CBMC source via 
    git clone https://github.com/diffblue/cbmc cbmc-git
cd cbmc-git
  3. To compile CBMC, do 
    gmake -C src minisat2-download
gmake -C src
  4. To compile JBMC, do 
    gmake -C jbmc/src setup-submodules
gmake -C jbmc/src

#Working with IDEs and Docker

Working with Visual Studio on Windows

Follow these instructions to work on Windows with Visual Studio:

  1. Open the cbmc folder in Visual Studio. Visual Studio 2017 and later have automated support for CMake projects, so you need to give sometime to Visual Studio to index the project and load its own plugins, and then it's going to be ready to build cbmc.
  2. Once indexing and plugin loading has finished, a menu Build should have appeared on the top bar. From there, select Build All.
  3. After the build has finished, there should be a folder out present. Navigating out/build/<build_parameters>/bin should get you to the binaries and other artifacts built, with <build_parameters> corresponding to something like x64-Debug (Default) or whatever the equivalent is for your system.

The above instructions have been tested against Visual Studio 2019.

Working with Eclipse

To work with Eclipse, do the following:

  1. Select File -> New -> Makefile Project with Existing Code
  2. Type "cprover" as "Project Name"
  3. Select the "src" subdirectory as "Existing Code Location"
  4. Select a toolchain appropriate for your platform
  5. Click "Finish"
  6. Select Project -> Build All

You can also select the repository's root folder as the "Existing Code Location". This has the advantage that you can build both CBMC and JBMC without the need to integrate JBMC as a separate project. Be aware that you need to change the build location (Select project in Eclipse -> Properties -> C/C++ Build) to one of the src directories.

Working with Docker

To compile and run the tools in a Docker container, do the following:

  1. From the root folder of the project, run $ docker build -t cbmc .

  2. After the building phase has finished, there should be a new image with the CProver binaries installed under /usr/local/bin/.

    To start a container using that image as a base, run $ docker run -i -t cbmc This will result in dropping you to a new terminal inside the container. To load files for analysis into the container, one way is by mounting the folder that contains the tests to the container. A possible invocation that does that is: $ docker run --mount type=bind,source=local/path/with/files,target=/mnt/analysis -i t cbmc. In the resulting container, the files present in the local file system under local/path/with/files will be present under /mnt/analysis.

#Compilation options and configuration

Compiling with CUDD

Cudd is a BDD library which can be used instead of the builtin miniBDD implementation.

If compiling with make:

  1. Download and compile Cudd: 
    make -C src cudd-download
  2. Uncomment the definition of CUDD in the file src/config.inc.
  3. Compile with make -C src

If compiling with CMake:

  1. Add the -DCMAKE_USE_CUDD=true flag to the cmake configuration phase. For instance: 
    cmake -S . -Bbuild -DCMAKE_USE_CUDD=true
  2. Run the build: 
    cmake --build build

Use BDDs for guards

There are two implementation for symex guards. The default one uses the internal representation of expression. The other one uses BDDs and though experimental, it is expected to have better performance, in particular when used in conjunction with CUDD.

To use the BDD implementation of guards, add the BDD_GUARDS compilation flag:

  • If compiling with make: 
    make -C src CXXFLAGS="-O2 -DBDD_GUARDS"
  • If compiling with CMake: 
    cmake -S . -Bbuild -DCMAKE_CXX_FLAGS="-DBDD_GUARDS"
    and then cmake --build build

Compiling with alternative SAT solvers

For the packaged builds of CBMC on our release page we currently build CBMC with the MiniSat2 SAT solver statically linked at compile time. However it is also possible to build CBMC using alternative SAT solvers.

Compiling CBMC Using Solver Native Interfaces

The following solvers are supported by CBMC using custom interfaces and can by downloaded and compiled by the build process: MiniSAT2, CaDiCaL, and Glucose.

For make alternatives to the default (i.e. not MiniSAT) can be built with the following commands for CaDiCaL:

make -C src cadical-download
make -C src CADICAL=../../cadical

and for glucose

make -C src glucose-download
make -C src GLUCOSE=../../glucose-syrup

For CMake the alternatives can be built with the following arguments to cmake for CaDiCaL -Dsat_impl=cadical and for glucose -Dsat_impl=glucose.

Compiling with IPASIR Interface

The below compiling instructions allow linking of an arbitrary IPASIR compatible SAT solver when compiling CBMC.

The general command using make is to compile with

make -C src LIBS="$PWD/SATOBJ SATLINKFLAGS" IPASIR=$PWD/SATPATH

Where SATOBJ is the pre-compiled IPASIR compatible SAT binary, SATLINKFLAGS are any flags required by the SAT object file, and SATPATH is the path to the SAT interface.

The rest of this section provides detailed instructions for some example SAT solvers.

Compiling with CaDiCaL via IPASIR

Note that CaDiCaL can also be built using CBMC's CaDiCaL native interface as described above. This section is to use CaDiCaL with the IPASIR interface in CBMC.

The CaDiCaL solver supports the IPASIR C interface to incremental SAT solvers, which is also supported by CBMC. So the process for producing a CBMC with CaDiCaL build is to build CaDiCaL as a static library then compile CBMC with the IPASIR build options and link to the CaDiCaL static library.

Note that at the time of writing this has been tested to work with the CaDiCaL 1.4.0 on Ubuntu 18.04 & 20.04 and MacOS.

  1. Download CaDiCaL:

    git clone --branch rel-1.4.0 https://github.com/arminbiere/cadical.git

    This will clone the CaDiCaL repository into a cadical subdirectory and checkout release 1.4.0, which has been checked for compatibility with CBMC at the time these instructions were written.

  2. Build CaDiCaL:

    cd cadical
./configure
make cadical
cd ..

    This will create a build directory called build inside the clone of the CaDiCaL repository. The cadical make target is specified in this example in order to avoid building targets which are not required by CBMC. The built static library will be placed in cadical/build/libcadical.a.

  3. Build CBMC:

    make -C src LIBS="$PWD/cadical/build/libcadical.a" IPASIR=$PWD/cadical/src

    This links the CaDiCaL library as part of the build. Passing the IPASIR parameter tells the build system to build for the IPASIR interface. The argument for the IPASIR parameter gives the build system the location for the IPASIR headers, which is needed for the cbmc includes of ipasir.h. The compiled binary will be placed in cbmc/src/cbmc/cbmc.

It's also possible to build CBMC using CaDiCaL through IPASIR via cmake, controlled with the flag -Dsat_impl=ipasir-cadical, like so:

$ cmake -Bbuild_ipasir -S. -Dsat_impl=ipasir-cadical

An advanced user may also take a more adventurous route, trying to link CBMC against any solver supporthing the IPASIR interface. To do that, an invocation like this is needed:

$ cmake -Bbuild_ipasir -S . -Dsat_impl=ipasir-custom -DIPASIR=<source_location> -DIPASIR_LIB=<lib_location>

with <source_location> being the absolute path to the folder containing the solver implementation and <lib_location> being the absolute path that contains a precompiled static library of the solver (.a file).

Compiling with Riss via IPASIR

The Riss solver supports the IPASIR C interface to incremental SAT solvers, which is also supported by CBMC. So the process for producing a CBMC with Riss build is to build Riss as a static library then compile CBMC with the IPASIR build options and link to the Riss static library. The following instructions have been confirmed to work on Ubuntu 20 at the time of writing. They can also be made to work on Ubuntu 18, by using a debug build of Riss. Riss uses Linux specific functionality / API calls. Therefore it can't be compiled successfully on Windows or macOS.

  1. Download Riss:

    git clone https://github.com/conp-solutions/riss riss

    This will clone the Riss repository into a riss subdirectory.

  2. Build Riss:

    cd riss
cmake -H. -Brelease -DCMAKE_BUILD_TYPE=Release
cd release
make riss-coprocessor-lib-static
cd ../..

    This will create a build directory called release inside the clone of the Riss repository and build the static library in riss/release/lib/libriss-coprocessor.a.

  3. Build CBMC:

    make -C src LIBS="$PWD/riss/release/lib/libriss-coprocessor.a -lpthread" IPASIR=$PWD/riss/riss

    This links the Riss library and the pthreads library. The pthreads library is needed because the Riss library uses it for multithreading. Passing the IPASIR parameter tells the build system to build for the IPASIR interface. The argument for the IPASIR parameter gives the build system the location for the IPASIR headers, which is needed for the cbmc includes of ipasir.h. The compiled binary will be placed in cbmc/src/cbmc/cbmc. This document assumes you have already been able to build CPROVER on your chosen architecture.

#RUNNING REGRESSION AND UNIT TESTS

Regression and unit tests can be run using cmake or make. Your choice here should be the same as the compiling of the project itself.

Note that running all regression and unit tests can be slow when a debug build of CPROVER is used.

CMAKE

This can be done by changing to the directory you built the project in with cmake and running ctest as follows.

cd <build_dir>
ctest . -V -L CORE

You can also specify a pattern of tests to run as follows.

ctest . -V -L CORE -R <pattern>

For example

ctest . -V -L CORE -R goto

that will run all CORE tests that include goto in their name.

MAKE

The regression and unit tests are handled differently in the make system. To run the regressions tests change to the regression directory and simply running make as follows.

cd regression
make test

To run the unit tests, change into the unit directory and then run make as follows.

cd unit
make test