SMT Based Verification in Haskell. Express properties about Haskell programs and automatically prove them using SMT solvers.

View the Project on GitHub LeventErkok/sbv

SBV: SMT Based Verification in Haskell

Express properties about Haskell programs and automatically prove them using SMT solvers.

$ ghci
ghci> :m Data.SBV
ghci> prove $ \x -> x `shiftL` 2 .== 4 * (x::SWord8)
ghci> prove $ \x -> x `shiftL` 2 .== 2 * (x::SWord8)
Falsifiable. Counter-example:
  s0 = 32 :: Word8

The function prove establishes theorem-hood, while sat finds any satisfying model. All satisfying models can be computed using allSat. SBV can also perform static assertion checks, such as absence of division-by-0, and other user given properties. Furthermore, SBV can perform optimization, minimizing/maximizing arithmetic goals for their optimal values.

SBV also allows for an incremental mode: Users are given a handle to the SMT solver as their programs execute, and they can issue SMTLib commands programmatically, query values, and direct the interaction using a high-level typed API. The incremental mode also allows for creation of constraints based on the current model, and access to internals of SMT solvers for advanced users. See the runSMT and query commands for details.


SBV library provides support for dealing with symbolic values in Haskell. It introduces the types:

The user can construct ordinary Haskell programs using these types, which behave like ordinary Haskell values when used concretely. However, when used with symbolic arguments, functions built out of these types can also be:

Picking the SMT solver to use

The SBV library uses third-party SMT solvers via the standard SMT-Lib interface. The following solvers are supported:

Most functions have two variants: For instance prove/proveWith. The former uses the default solver, which is currently Z3. The latter expects you to pass it a configuration that picks the solver. The valid values are abc, boolector, bitwuzla, cvc4, cvc5, dReal, mathSAT, yices, and z3

See versions for a listing of the versions of these tools SBV has been tested with. Please report if you see any discrepancies!

Other SMT solvers can be used with SBV as well, with a relatively easy hook-up mechanism. Please do get in touch if you plan to use SBV with any other solver.

Using multiple solvers, simultaneously

SBV also allows for running multiple solvers at the same time, either picking the result of the first to complete, or getting results from all. See proveWithAny/proveWithAll and satWithAny/satWithAll functions. The function sbvAvailableSolvers can be used to query the available solvers at run-time.

The SBV library is distributed with the BSD3 license. See COPYRIGHT for details. The LICENSE file contains the BSD3 verbiage.


The following people made major contributions to SBV, by developing new features and contributing to the design in significant ways: Joel Burget, Brian Huffman, Brian Schroeder, and Jeffrey Young.

The following people reported bugs, provided comments/feedback, or contributed to the development of SBV in various ways: Ara Adkins, Kanishka Azimi, Markus Barenhoff, Reid Barton, Ben Blaxill, Ian Blumenfeld, Guillaume Bouchard, Martin Brain, Ian Calvert, Oliver Charles, Christian Conkle, Matthew Danish, Iavor Diatchki, Alex Dixon, Robert Dockins, Thomas DuBuisson, Trevor Elliott, Gergő Érdi, John Erickson, Adam Foltzer, Joshua Gancher, Remy Goldschmidt, Brad Hardy, Tom Hawkins, Greg Horn, Jan Hrcek, Georges-Axel Jaloyan, Anders Kaseorg, Tom Sydney Kerckhove, Piërre van de Laar, Pablo Lamela, Ken Friis Larsen, Joe Leslie-Hurd, Brett Letner, Sirui Lu, Georgy Lukyanov, Martin Lundfall, John Matthews, Curran McConnell, Philipp Meyer, Joshua Moerman, Matt Parker, Jan Path, Matt Peddie, Matthew Pickering, Lee Pike, Gleb Popov, Rohit Ramesh, Geoffrey Ramseyer, Jaro Reinders, Stephan Renatus, Dan Rosén, Ryan Scott, Eric Seidel, Austin Seipp, Andrés Sicard-Ramírez, Don Stewart, Greg Sullivan, Josef Svenningsson, Matt Torrence, Daniel Wagner, Sean Weaver, Nis Wegmann, and Jared Ziegler.