Timeline

2026: Colibri2 0.7

Many domains are now written in WhyML, proved with Why3 and extracted to OCaml (the proved library): the domain of the reals (unions of intervals and congruences), the bit-vector domain (BDD), known bits, floating-point numbers with the rounding of Flocq, the numbers, the ring normal form.

Also new:

  • DSeq, a common theory for arrays and sequences (with OCamlPro)
  • Integer bitwise operators and non-linear bit-vector identities
  • Floating-point rules ported from COLIBRI
  • Rewrite rules for quantified equalities (:rewrite-rule)
  • About 870 more goals proved than 0.6 on our benchmarks of program verification

2021: Colibri2 0.1

The solver supports:

  • Boolean
  • Linear arithmetic (using Fourier-Mostkin)
  • Uninterpreted symbols
  • Quantifiers
  • Polymorphism

In a limited way:

  • Floating Point numbers
  • Function ceil and floor

2020: Colibri2

The learning developed in Popop is removed to keep only the propagation part as in COLIBRI. So the solver is named Colibri2. The first features to be available are:

  • Boolean
  • Linear arithmetic (using Fourier-Mostkin)
  • Uninterpreted symbols
  • Quantifiers

2019: Colibrics

Colibrics started as an experiment on the possibility to prove a simple CP solver in Why3

2016: Popop

Popop, developped during the SOPRANO project, is a framework that combined the learning of SMT and the propagation of CP. However the learning framework was to cumbersome to allow the developpement of more complicated theories than booleans and linear arithmetic.

2014: SOPRANO

The ANR SOPRANO project allowed to gather researchers working on SMT (OcamlPro, Université Paris-Saclay) and CP (CEA, INRIA) and industrials partner (Adacore) in the domain of software verification. The goal was to merge the powerful learning of SMT and the powerful propagations of CP.

2000: First COLIBRI

COLIBRI started as a COnstraint solving LIBrary in Eclipse Prolog