Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?


Failed to load latest commit information.
Latest commit message
Commit time


Mangle is a programming language for deductive database programming. It is an extension of Datalog, with various extensions like aggregation, function calls and optional type-checking.

Deductive database programming is useful for bringing data from multiple data sources together since it enables us to represent and query that data in a uniform way. It can also be used to model domain knowledge, similar to machine-readable ontology but without being restricted to binary predicates.

Datalog is an expressive declarative language similar to relational calculus (think SQL and relational views). Unlike relational calculus, it also supports recursive rules and program structuring in a straightforward way.

Mangle contains Datalog as a fragment and adds extensions that make its use more practical. Some of the good properties like guaranteed termination are lost when such extensions are used.

The goal of Mangle as an open source project is to convey the concepts in a way that is accessible to developers and lends itself to easy experimentation. This repository contains an implementation of Mangle as a go library that can be easily embedded into applications.

Check out the docs and the GitHub discussions for more information. There is also a Q&A section.

This is not an officially supported Google product.

Table of Contents


Simple Queries

Imagine you were asked to spot software affected by the log4j vulnerability discovered in late 2021. We want to look for projects that contain a Java archive (jar file) of log4j that is not updated to the patched version.

projects_with_vulnerable_log4j(P) :-
  contains_jar(P, "log4j", Version),
  Version != "2.17.1",
  Version != "2.12.4",
  Version != "2.3.2".

This is a Mangle rule: conceptually, the implementation retrieve all possible values for variables P and Version that make all the subgoals true.

Simple Mangle rules like this correspond to select-project-join relational queries. The same query in SQL would look like this:

FROM projects JOIN contains_jar ON = contains_jar.project_id
WHERE contains_jar.version NOT IN ("2.17.1", "2.12.4", "2.3.2")

Unlike SQL, our Mangle rule projects_with_vulnerable_log4j has a name and can be referenced in other queries.

(If translating non-recursive Datalog into SQL queries sounds interesting, you should check out the Logica open source project.)


In practice, querying is rarely enough and we also need grouping and aggregation.

count_projects_with_vulnerable_log4j(Num) :-
  projects_with_vulnerable_log4j(P) |> do fn:group_by(), let Num = fn:Count().

Recursive Queries

The example does not specify what contains_jar does. Here is a possible implementation for contains_jar that walks a dependency graph. This shows that Mangle rules can be recursive.

contains_jar(P, Name, Version) :-
  contains_jar_directly(P, Name, Version).

contains_jar(P, Name, Version) :-
  project_depends(P, Q),
  contains_jar(Q, Name, Version).

The two rules correspond to two cases in which a project may "contain" a jar: either directly, or through some dependency.

Knowledge Graphs, Property Graphs

In requirements engineering, one needs to captures real world concepts in a domain model and controlled vocabulary. Description logics use roles to describe how concepts interact, but these relationships are always binary. Mangle can represent binary predicates, but also arbitrary n-ary relations. Moreover it also has support for structured data.

one_or_two_leg_trip(Codes, Start, Destination, Price) :-
  direct_conn(Code, Start, Destination, Price)
  |> let Codes = [Code].

one_or_two_leg_trip(Codes, Start, Destination, Price) :-
  direct_conn(FirstCode, Start, Connecting, FirstLegPrice).
  direct_conn(SecondCode, Connecting, Destination, SecondLegPrice)
  |> let Code = [FirstCode, SecondCode],
     let Price = fn:plus(FirstLegPrice, SecondLegPrice).

graph LR
    /zurich -->|/code/ZL <br /> 60 CHF| /lausanne
    /zurich -->|/code/ZB <br /> 30 CHF| /bern
    /bern -->|/code/BL <br /> 30 CHF| /lausanne


If you want to build from source, you need to set up ANTLR first, which requires Java runtime environment.

alias antlr='java -jar $PWD/antlr-4.11.1-complete.jar'

Then you can generate the parser sources

antlr -Dlanguage=Go -package gen -o ./ parse/gen/Mangle.g4 -visitor

... and finally get the dependencies (see go.mod) and build the library:

go get ./...
go build ./...


The Mangle maintainers welcome external contributions to spec, documentation and this implementation (see and also other implementations. Pull requests will be handled like for tensorflow, to ensure our internal usage and tests will pass.


No description, website, or topics provided.



Code of conduct

Security policy





No releases published


No packages published