souffle-haskell: Souffle Datalog bindings for Haskell

[ bindings, foreign-binding, library, logic-programming, mit ] [ Propose Tags ]

Souffle Datalog bindings for Haskell.

[Skip to Readme]
Versions [RSS] [faq] 0.0.1, 0.1.0, 0.2.0, 0.2.1, 0.2.2, 0.2.3, 1.0.0, 1.1.0, 2.0.0, 2.0.1, 2.1.0, 3.0.0, 3.1.0
Change log
Dependencies base (>=4.12 && <5), deepseq (>=1.4.4 && <2), directory (>=1.3.3 && <2), filepath (>=1.4.2 && <2), mtl (>=2.0 && <3), process (>=1.6 && <2), template-haskell (==2.*), temporary (>=1.3 && <2), text (>=1.0 && <2), type-errors-pretty (>= && <1), vector (<=1.0) [details]
License MIT
Copyright 2020 Luc Tielen
Author Luc Tielen
Category Logic Programming, Foreign Binding, Bindings
Home page
Bug tracker
Source repo head: git clone
Uploaded by luc_tielen at 2020-04-25T11:46:25Z
Distributions NixOS:3.0.0
Downloads 1810 total (21 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Hackage Matrix CI
Docs uploaded by user
Build status unknown [no reports yet]


[Index] [Quick Jump]


Maintainer's Corner

For package maintainers and hackage trustees


Readme for souffle-haskell-0.2.1

[back to package description]


License: MIT CircleCI Hackage

This repo provides Haskell bindings for performing analyses with the Souffle Datalog language.

Fun fact: this library combines both functional programming (Haskell), logic programming (Datalog / Souffle) and imperative / OO programming (C / C++).

Motivating example

Let's first write a datalog program that can check if one point is reachable from another:

// We define 2 data types:
.decl edge(n: symbol, m: symbol)
.decl reachable(n: symbol, m: symbol)

// We indicate we are interested in "reachable" facts.
// NOTE: If you forget to add outputs, the souffle compiler will
//       try to be smart and remove most generated code!
.output reachable

// We write down some pre-defined facts on the datalog side.
edge("a", "b").
edge("b", "c").
edge("c", "e").
edge("e", "f").
edge("c", "d").

// And we tell datalog how to check if 1 point is reachable from another.
reachable(x, y) :- edge(x, y).                  // base rule
reachable(x, z) :- edge(x, y), reachable(y, z). // inductive rule

Now that we have the datalog code, we can generate a path.cpp from it using souffle -g path.cpp path.dl. souffle-haskell can bind to this program in the following way:

-- Enable some necessary extensions:
{-# LANGUAGE TemplateHaskell, ScopedTypeVariables, DataKinds, TypeFamilies, DeriveGeneric #-}

module Main ( main ) where

import Data.Foldable ( traverse_ )
import Control.Monad.IO.Class
import GHC.Generics
import Data.Vector
import qualified Language.Souffle.TH as Souffle
import qualified Language.Souffle.Compiled as Souffle

-- We only use template haskell for directly embedding the .cpp file into this file.
-- If we do not do this, it will link incorrectly due to the way the
-- C++ code is generated.
Souffle.embedProgram "/path/to/path.cpp"

-- We define a data type representing our datalog program.
data Path = Path

-- Facts are represent in Haskell as simple product types,
-- Numbers map to Int32, symbols to Strings / Text.

data Edge = Edge String String
  deriving (Eq, Show, Generic)

data Reachable = Reachable String String
  deriving (Eq, Show, Generic)

-- By making Path an instance of Program, we provide Haskell with information
-- about the datalog program. It uses this to perform compile-time checks to
-- limit the amount of possible programmer errors to a minimum.
instance Souffle.Program Path where
  type ProgramFacts Path = [Edge, Reachable]
  programName = const "path"

-- By making a data type an instance of Edge, we give Haskell the
-- necessary information to bind to the datalog fact.
instance Souffle.Fact Edge where
  factName = const "edge"

instance Souffle.Fact Reachable where
  factName = const "reachable"

-- For simple product types, we can automatically generate the
-- marshalling/unmarshalling code of data between Haskell and datalog.
instance Souffle.Marshal Edge
instance Souffle.Marshal Reachable

main :: IO ()
main = Souffle.runSouffle $ do
  maybeProgram <- Souffle.init Path  -- Initializes the Souffle program.
  case maybeProgram of
    Nothing -> liftIO $ putStrLn "Failed to load program."
    Just prog -> do
      Souffle.addFact prog $ Edge "d" "i"   -- Adding a single fact from Haskell side
      Souffle.addFacts prog [ Edge "e" "f"  -- Adding multiple facts
                            , Edge "f" "g"
                            , Edge "f" "g"
                            , Edge "f" "h"
                            , Edge "g" "i"
                            ] prog  -- Run the Souffle program

      -- NOTE: You can change type param to fetch different relations
      --       Here it requires an annotation since we directly print it
      --       to stdout, but if passed to another function, it can infer
      --       the correct type automatically.
      --       A list of facts can also be returned here.
      results :: Vector Reachable <- Souffle.getFacts prog
      liftIO $ traverse_ print results

      -- We can also look for a specific fact:
      maybeFact <- Souffle.findFact prog $ Reachable "a" "c"
      liftIO $ print $ maybeFact

For more examples of how to use the top level API, you can also take a look at the tests.

Getting started

This library assumes that the Souffle include paths are properly set. This is needed in order for the C++ code to be compiled correctly. The easiest way to do this (that I know of) is via Nix. Add souffle to the build inputs of your derivation and everything will be set correctly. Without Nix, you will have to follow the manual install instructions on the Souffle website.

In your package.yaml / *.cabal file, make sure to add the following options (assuming package.yaml here):

# ...


# ...

This will instruct the Souffle compiler to compile the C++ in such a way that it can be linked with other languages (including Haskell!).

Supported modes

Souffle programs can be run in 2 ways. They can either run in interpreted mode (using the souffle CLI command), or they can be compiled to C++-code and called from a host program for improved efficiency. This library supports both modes (since version 0.2.0). The two variants have only a few minor differences and can be swapped fairly easily.

Interpreted mode

This is probably the mode you want to start out with if you are developing a program that uses Datalog for computing certain relations. Interpreted mode offers quick development iterations (no compiling of C++ code each time you change your Datalog code). However because the Souffle code is interpreted, it can't offer the same speed as in compiled mode.

The main differences with compiled mode are the following:

  1. You need to import Language.Souffle.Interpreted
  2. You need to call Souffle.cleanup after you no longer need the Souffle functionality. This will clean up the generated CSV fact files located in a temporary directory.
  3. You don't need to import Language.Souffle.TH to embed a Datalog program.

Interpreter configuration

The interpreter uses CSV files to read or write facts. The configuration allows specifiying where the fact directory is located. With the default configuration, it will try to lookup DATALOG_DIR in the environment and fall back to the current directory (or .).

You can also configure which souffle executable will be used. By default, it will first look at the SOUFFLE_BIN environment variable. If this is not set, it will try to find the executable using the which shell-command. If it also can't find the executable this way, then it will fail to initialize the interpreter.

For more information regarding configuration, take a look at the runSouffleWith function.

The separators in the CSV fact files cannot be configured at the moment. A tab character ('\t') is used to separate the different columns.

Compiled mode

Once the prototyping phase of the Datalog algorithm is over, it is advised to switch over to the compiled mode. It offers much improved performance compared to the interpreted mode, at the cost of having to recompile your Datalog algorithm each time it changes.

The main differences with interpreted mode are the following:

  1. Compile the Datalog code with souffle -g.
  2. You need to import Language.Souffle.TH to embed a Datalog program using Language.Souffle.TH.embedProgram, as shown in the motivating example.
  3. Remove Souffle.cleanup if it is present in your code, compiled mode leaves no CSV artifacts.

The motivating example is a complete example for the compiled mode.


TLDR: Nix-based project; the Makefile contains the most commonly used commands.

Long version:

The project makes use of Nix to setup the development environment. Setup your environment by entering the following command:

$ nix-shell

After this command, you can build the project:

$ make configure  # configures the project
$ make build      # builds the haskell code
$ make lint       # runs the linter
$ make hoogle     # starts a local hoogle webserver


Found an issue or missing a piece of functionality? Please open an issue with a description of the problem.