Skip to content

Latest commit

 

History

History
40 lines (34 loc) · 2.17 KB

README.md

File metadata and controls

40 lines (34 loc) · 2.17 KB

Examples

This directory contains example usage of smt-switch.

QF_UFBV

QF_UFBV stands for quantifier-free uninterpreted functions and bit-vectors. The files cvc5_qf_ufbv.cpp and btor_qf_ufbv.cpp demonstrate some common usage of the API with this logic.

The build.sh script will setup and install smt-switch in this directory and then build the examples. The script has comments explaining each step for building smt-switch. The script assumes you already have dependencies installed and that boolector and cvc5 have been built in the top-level repository under deps/cvc5 and deps/boolector. You can use the helper scripts setup-cvc5.sh and setup-btor.sh to automate this. If the build script fails and you haven't recently installed the solvers, you might have an old version. You can try deleting them from deps and rebuilding them.

You can look at the Makefile to understand how to link smt-switch to a binary. If the script succeeds in building everything, you can run each of the files with ./cvc5_qf_ufbv.out and ./btor_qf_ufbv.out, respectively. Note that there may be a delay for looking up the shared libraries on some systems the first time they are run. Importantly, notice that the files differ by only two lines of code (excluding comments), demonstrating that changing the solver is straightforward.

To explore the API, you can change either of those files and run make again. To remove the built binaries, run make clean. To clean up all the build and install files for smt-switch in this directory, run make clean-all.

Python bindings

You can also run the same example through the Python bindings with the file, python_qf_ufbv.py. This requires building the Python bindings, which can be done by running ./build.sh --python. If you have already built without Python bindings, please run make clean-all before rebuilding. Python3 along with the modules Cython and pytest are required for building the bindings and running the example. It is recommended to use a Python3 virtual environment.