antoinemine / apron

Apron Numerical Abstract Domain Library
Other
112 stars 33 forks source link

The Apron Numerical Abstract Domain Library

Introduction

Apron is a library to represent properties of numeric variables, such as variable bounds or linear relations between variables, and to manipulate these properties through semantic operations, such as variable assignments, tests, conjunctions, entailment.

Apron is intended to be used in static program analyzers, in order to infer invariants of numeric variables, i.e., properties that hold for all executions of a program. It is based on the theory of Abstract Interpretation.

The API documentation is available on the GitHub page for Apron.

Overview

The Apron library includes several numeric abstract domains, corresponding to different classes of numeric properties with their own internal representation and algorithms, achieving various trade-offs between precision, expressiveness, and efficiency.

Apron includes the following numeric domains:

Additional domains are made available through the optional PPL and PPLite third-party libraries:

The domains are made available under a common interface, so that changing the abstract domain of interpretation in a static analysis should only take a one-line change.

The core API is in C, but optional API wrappers for additional languages are provided:

Dependencies

Base

Compiling the built-in domains with the C interface requires:

Additional domains

Compiling the PPL-based domains requires the Parma Polyhedra Library (tested with version 1.2) and gcc (no clang).

Compiling the PPLite-based domains requires the PPLite library, which also depends on Flint. Note that building the PPLite library from sources requires using a C++ compiler (g++ or clang++) that supports the c++17 language standard; however, starting from PPLite version 0.11, the Apron wrapper for PPLite can be compiled using a C++ compiler supporting the c++11 language standard.

Additional language support

Additional language wrappers require additional components:

Installation instructions

Installation with Opam

If you are using OCaml and the Opam package manager, then you could install the latest version of Apron in Opam with just opam install apron. Only Opam 2.x is supported.

To compile from this source tree, you can install the dependencies with opam install --deps-only . and follow the instructions below.

Installation on Linux

On deb-based Linux distributions (Debian, Ubuntu) a sudo apt-get install libgmp-dev libmpfr-dev should suffice to get the dependencies for the basic C library.

On Opam-based OCaml distributions, a opam install ocamlfind camlidl mlgmpidl should suffice to get the dependencies for the OCaml API.

Compilation from source could be as simple as:

./configure automatically generates a Makefile.config file. It is also possible to write a Makefile.config by hand by taking some inspiration from Makefile.config.model.

In case some components fail to compile, it is possible to disable them through ./configure options:

See ./configure -help for more options.

Debug versions

By default, make install now only install non-debug versions of the C libraries. Moreover, these are striped of symbols.

Use the -debug ./configure option to also install debug (non-stripped) C versions, and -no-strip to avoid stripping the non-debug C versions. The C debug versions have a _debug suffix (such as libapron_debug.so).

When installing with opam, debug versions are always available. OCaml debug libraries use the .d suffix (such as apron.d.cmxa).

Installation on MacOS X

Help needed for this section.

The README.mac file is not up to date.

Installation on Windows

See the Windows README.

Documentation compilation

You can build the documentation with make doc. You will need the following tools:

Note that some generated documentation may not be up-to-date.

A generated copy of the documentation is available on-line on the GitHub page for Apron.