Blog post for CP ecosystem

_posts/2021-09-18-cp-0-6.md

@@ -0,0 +1,55 @@
+---
+layout: post
+title: "ConstraintProgrammingExtensions 0.6 is released"
+date: 2021-09-18
+categories: [releases]
+permalink: /blog/cpe-0.6-release/
+---
+
+JuMP and MathOptInterface are mostly oriented towards mathematical optimisation. 
+However, their interface is amenable to other kinds of formalism, including 
+constraint programming (CP). Bridging the gap between MOI and constraint programming
+is the goal of the 
+[ConstraintProgrammingExtensions.jl](https://github.com/dourouc05/ConstraintProgrammingExtensions.jl)
+package. 
+
+Discussions for support of constraint programming in the JuMP ecosystem started
+[in 2019](https://github.com/jump-dev/JuMP.jl/issues/2014). The original scope
+was quite limited, as it only planned to propose some modeling capabilities to 
+JuMP. However, quite soon, it morphed into a larger proposal of a full 
+[constraint programming environment](https://github.com/jump-dev/JuMP.jl/issues/2227)
+accessible from the same API as JuMP. It also meant giving access to CP solvers. 
+
+Mathematical optimisation models are only based on equality and inequality constraints, 
+typically linear or convex, with an objective function. Constraint programming 
+allows more variety in the kind of constraints that are supported, with a 
+focus on discrete structure: the best-known constraint is probably `all_different`, 
+which models the fact that several integer-valued variables should take different
+values (no two variables are allowed to have the same value). Extremely efficient
+solvers can be written for this kind of model.
+
+In the last few months, the CP ecosystem in JuMP made tremendous progress. 
+
+- The lowest layer, 
+  [ConstraintProgrammingExtensions.jl](https://github.com/dourouc05/ConstraintProgrammingExtensions.jl),
+  has [dozens of constraint types available](http://tcuvelier.be/ConstraintProgrammingExtensions.jl/dev/reference/sets/)
+  and [dozens of bridges between constraints](http://tcuvelier.be/ConstraintProgrammingExtensions.jl/dev/reference/bridges_sets/).
+  It supports the FlatZinc file format, a standard in constraint programming to exchange
+  models (similar to the LP and MPS formats for linear programming), both to read and 
+  write models, but also to communicate with solvers. It has recently been updated
+  to be compatible with MOI 0.10.
+
+- Several solvers are accessible using this API: the pure-Julia 
+  [ConstraintSolver.jl](https://github.com/Wikunia/ConstraintSolver.jl/), the industry standard 
+  [IBM CPLEX CP Optimizer](https://github.com/dourouc05/CPLEXCP.jl/), and the open-source
+  [Chuffed](https://github.com/dourouc05/Chuffed.jl/).
+
+- Other pure-Julia CP solvers have been released, but they do not yet use the same API:
+  [JuliaConstraints](https://github.com/JuliaConstraints) and 
+  [SeaPerl](https://github.com/corail-research/SeaPearl.jl). 
+
+This is not the end of the journey. Although MOI and JuMP are nearing their 1.0 
+release, the CP ecosystem has still a long way to go. In the near term, more
+modelling features should be added as extensions to JuMP and more solvers should
+be made accessible. In the longer term, CP modelling will also benefit from 
+the new generation of nonlinear support.